Megatest

           process.scm runs.scm tests.scm genexample.scm	\
           http-transport.scm filedb.scm tdb.scm client.scm mt.scm	\
           ezsteps.scm lock-queue.scm sdb.scm rmt.scm api.scm		\
           subrun.scm archive.scm env.scm		\
           diff-report.scm

# module source files
MSRCFILES = dbmod.scm servermod.scm apimod.scm commonmod.scm rmtmod.scm \
            ods.scm configfmod.scm transport.scm portlogger.scm tasks.scm \
            pgdb.scm margsmod.scm debugprint.scm

MOFILES = $(addprefix mofiles/,$(MSRCFILES:%.scm=%.o))
MOIMPFILES = $(MSRCFILES:%.scm=%.import.o)

# ftail.scm rmtmod.scm commonmod.scm removed
# MSRCFILES = ducttape-lib.scm pkts.scm stml2.scm cookie.scm mutils.scm	\
#             mtargs.scm commonmod.scm dbmod.scm adjutant.scm ulex.scm	\
#             rmtmod.scm apimod.scm

;;======================================================================

(declare (unit apimod))
(declare (uses commonmod))
(declare (uses dbmod))
(declare (uses tasks))
;; (declare (uses servermod))

(module apimod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69)
(import commonmod)
(import dbmod)
(import tasks)
;; (import servermod)

;; allow these queries through without starting a server
;;
(define api:read-only-queries
  '(get-key-val-pairs
    get-var
    get-keys

                   ;;===============================================
                   ;; READ/WRITE QUERIES
                   ;;===============================================

                   ((get-keys-write)                        (db:get-keys dbstruct)) ;; force a dummy "write" query to force server; for debug in -repl
                   
                   ;; SERVERS
                   ;; ((start-server)                    (apply server:kind-run params))
                   ((kill-server)                     (set! *server-run* #f))

                   ;; TESTS

                   ;;((test-set-state-status-by-id)     (apply mt:test-set-state-status-by-id dbstruct params))
                   ;;BB - commented out above because it was calling below, eventually, incorrectly (dbstruct passed to mt:test-set-state-status-by-id, which previosly did more, but now only passes thru to db:set-state-status-and-roll-up-items.
                   ((test-set-state-status-by-id)

# To regenerate this file do:
#   (cd utils/;ck52 csc gendeps.scm) && ./utils/gendeps allunits *scm
#   cp allunits.inc build.inc
#




mofiles/apimod.o : mofiles/tasks.o







mofiles/transport.o : mofiles/configfmod.o




















mofiles/tasks.o : mofiles/configfmod.o



mofiles/servermod.o : mofiles/configfmod.o






mofiles/portlogger.o : mofiles/configfmod.o










mofiles/dcommonmod.o : mofiles/configfmod.o
















mofiles/dbmod.o : mofiles/configfmod.o
mofiles/dbmod.o : mofiles/ods.o
mofiles/transport.o : mofiles/commonmod.o

mofiles/tasks.o : mofiles/commonmod.o
mofiles/servermod.o : mofiles/commonmod.o




mofiles/rmtmod.o : mofiles/commonmod.o

mofiles/portlogger.o : mofiles/commonmod.o


mofiles/ods.o : mofiles/commonmod.o







mofiles/dcommonmod.o : mofiles/commonmod.o



mofiles/dbmod.o : mofiles/commonmod.o

mofiles/configfmod.o : mofiles/commonmod.o



mofiles/apimod.o : mofiles/commonmod.o
mofiles/rmtmod.o : mofiles/apimod.o


mofiles/tasks.o : mofiles/margsmod.o



mofiles/commonmod.o : mofiles/margsmod.o

mofiles/tasks.o : mofiles/dbmod.o



mofiles/servermod.o : mofiles/dbmod.o



mofiles/rmtmod.o : mofiles/dbmod.o



mofiles/portlogger.o : mofiles/dbmod.o



mofiles/apimod.o : mofiles/dbmod.o


mofiles/transport.o : mofiles/portlogger.o

(declare (uses servermod))
(import servermod)

(declare (uses margsmod))
(import margsmod)














































































































































































;;======================================================================
;; logic for getting homehost. Returns (host . at-home)
;; IF *toppath* is not set, wait up to five seconds trying every two seconds
;; (this is to accomodate the watchdog)
;;
(define (common:get-homehost #!key (trynum 5))
  ;; called often especially at start up. use mutex to eliminate collisions
  (mutex-lock! *homehost-mutex*)
  (cond
   (*home-host*
    (mutex-unlock! *homehost-mutex*)
    *home-host*)
   ((not *toppath*)
    (mutex-unlock! *homehost-mutex*)
    (launch:setup) ;; safely mutexed now
    (if (> trynum 0)

	(begin
	  (thread-sleep! 2)
	  (common:get-homehost trynum: (- trynum 1)))
	#f))
   (else
    (let* ((currhost (get-host-name))
	   (bestadrs (server:get-best-guess-address currhost))

					 (car (common:get-homehost))))
				     #f))))))
	   (at-home  (or (equal? homehost currhost)
			 (equal? homehost bestadrs))))
      (set! *home-host* (cons homehost at-home))
      (mutex-unlock! *homehost-mutex*)
      *home-host*))))
(define (common:wait-for-homehost-load maxnormload msg)
  (let* ((hh-dat (if (common:on-homehost?) ;; if we are on the homehost then pass in #f so the calls are local.
                     #f
                     (common:get-homehost)))
         (hh     (if hh-dat (car hh-dat) #f)))
    (common:wait-for-normalized-load maxnormload msg hh)))

;;======================================================================
;; am I on the homehost?
;;======================================================================
;;======================================================================
;;======================================================================
;;
(define (common:on-homehost?)
  (let ((hh (common:get-homehost)))
    (if hh
	(cdr hh)
	#f)))





















































































































































































(define (common:run-sync?)
    (and (common:on-homehost?)
	 (args:get-arg "-server")))

;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;;======================================================================

(declare (unit commonmod))
(declare (uses margsmod))
(declare (uses debugprint))

(module commonmod
	*
	
;;(import scheme chicken data-structures extras files ports)
(import scheme chicken)
(import margsmod)
(import debugprint)

(use data-structures extras files ports)
(use 
  (prefix base64 base64:)
  (prefix sqlite3 sqlite3:) 
  (srfi 18)
  (prefix dbi dbi:)
  directory-utils
  format
  matchable
  md5
  message-digest
  pkts
  posix 
  regex
  regex-case
  sparse-vectors
  srfi-1
  srfi-13
  srfi-69
  stack
  stml2
  typed-records 
  z3
 )
	
(include "common_records.scm")

;;======================================================================
;; (require-library margs)
;; (include "margs.scm")

;; (define old-exit exit)
;; 
;; (define (exit . code)
;;   (if (null? code)
;;       (old-exit)
;;       (old-exit code)))

(define *numcpus-cache* (make-hash-table))

;;======================================================================
;; use to transition to area-name

(define (common:get-sync-lock-filepath)
  (let* ((tmp-area     (common:get-db-tmp-area))
         (lockfile     (conc tmp-area "/megatest.db.sync-lock")))
    lockfile))
    

(define (common:get-sync-lock-filepath)
  (let* ((tmp-area     (common:get-db-tmp-area))
         (lockfile     (conc tmp-area "/megatest.db.sync-lock")))
    lockfile))

;;======================================================================
;; (debug:print 0 *default-log-port* "Failed to find this executable! Using what can be found on the path")


(define (configf:lookup cfgdat section var)
  (if (hash-table? cfgdat)
      (let ((sectdat (hash-table-ref/default cfgdat section '())))
	(if (null? sectdat)
	    #f
	    (let ((match (assoc var sectdat)))
	      (if match ;; (and match (list? match)(> (length match) 1))
		  (cadr match)
		  #f))
	    ))
      #f))

(define (common:get-testsuite-name)
  (or (configf:lookup *configdat* "setup" "area-name") ;; megatest is a flexible tool, testsuite is too limiting a description.
      (configf:lookup *configdat* "setup" "testsuite" )
      (getenv "MT_TESTSUITE_NAME")
      (pathname-file (or (if (string? *toppath* )
			     (pathname-file *toppath*)
			     #f)
			 (common:get-toppath #f)))
      "please-set-setup-area-name")) ;; (pathname-file (current-directory)))))

(define common:get-area-name common:get-testsuite-name)


(define (get-area-name configdat toppath #!optional (short #f))
  ;; look up my area name in areas table (future)
  ;; generate auto name
  (conc (get-area-path-signature toppath short)
	"-"
	(common:get-testsuite-name toppath configdat)))


;;======================================================================
;; L O C K I N G   M E C H A N I S M S 
;;======================================================================
(define (common:get-db-tmp-area . junk)
  (if *db-cache-path*
      *db-cache-path*
      (if *toppath* ;; common:get-create-writeable-dir
	  (handle-exceptions
	      exn
	      (begin
		(debug:print-error 0 *default-log-port* "Couldn't create path to " *db-cache-path* ", exn=" exn)
		(exit 1))
	      (let* ((tsname (common:get-testsuite-name))
		     (dbpath (common:get-create-writeable-dir
			      (list (conc "/tmp/" (current-user-name)
					  "/megatest_localdb/"
					  tsname "/"
					  (string-translate *toppath* "/" "."))
				    (conc "/tmp/" (current-process-id) ;; just in case we have an issue with the dir by own user name
					  "/megatest_localdb/"
					  tsname
					  (string-translate *toppath* "/" "."))
				    ))))
		(set! *db-cache-path* dbpath)
		dbpath))
	  #f)))



;;======================================================================
;; CONTENTS
;;
;;  config file utils
;;  misc conversion, data manipulation functions
;;  testsuite and area utilites

  (let ((adat (get-section cfgdat "areas")))
    (map (lambda (entry)
	   `(,(car entry) . 
	     ,(val->alist (cadr entry))))
	 adat)))

;;======================================================================
;; debug
;;======================================================================










































(define (debug:setup debug-arg verbose-arg) ;; debug-arg= #f, #t or 'noprop
  (let ((debugstr (or debug-arg ;; (args:get-arg "-debug")
		      ;; (args:get-arg "-debug-noprop")












		      (get-environment-variable "MT_DEBUG_MODE"))))
    (debug:calc-verbosity debugstr verbose-arg)
    ;; (debug:check-verbosity *verbosity* debugstr)


    ;; if we were handed a bad verbosity rule then we will override it with 1 and continue
    (if (not (verbosity))(set! (verbosity) 1))



    (if (and (not (eq? debug-arg 'noprop))


	     (or debug-arg



		 (not (get-environment-variable "MT_DEBUG_MODE"))))

	(setenv "MT_DEBUG_MODE" (if (list? (verbosity))



				    (string-intersperse (map conc (verbosity)) ",")
				    (conc (verbosity)))))))








;;======================================================================
;; Safe utilities
;;======================================================================

(define (common:false-on-exception thunk #!key (message #f))
  (handle-exceptions exn

(define *testconfigs*        (make-hash-table)) ;; test-name => testconfig
(define *runconfigs*         (make-hash-table)) ;; target    => runconfig

;; This is a cache of pre-reqs met, don't re-calc in cases where called with same params less than
;; five seconds ago
(define *pre-reqs-met-cache* (make-hash-table))





(use posix-extras pathname-expand files)

;; this plugs a hole in posix-extras in recent chicken versions > 4.9)
(let-values (( (chicken-release-number chicken-major-version)
               (apply values
                      (map string->number
                           (take

;;     (cons 
;;      (if nstate  (symbol->string nstate)  nstate)
;;      (if nstatus (symbol->string nstatus) nstatus))))
               
;;======================================================================
;; D E B U G G I N G   S T U F F 
;;======================================================================

(define (common:debug-setup)
  (debug:setup (cond ;; debug arg
		((args:get-arg "-debug-noprop") 'noprop)
		((args:get-arg "-debug")         #t)
		(else #f))
	       (cond ;; verbosity arg
		((args:get-arg "-q") 'v)
		((args:get-arg "-q") 'q)
		(else #f))))

(define *verbosity*         1)
(define *logging*           #f)

(define (assoc/default key lst . default)
  (let ((res (assoc key lst)))
    (if res (cadr res)(if (null? default) #f (car default)))))

      (thread-sleep! 0.5)
      (if (common:file-exists? gzfile) (delete-file gzfile))
      (system (conc "gzip " logfile))
      
      (unsetenv "TARGETHOST_LOGF")
      (unsetenv "TARGETHOST"))))
    

;;======================================================================
;; hash-table tree to html list tree
;;
;;   tipfunc takes two parameters: y the tip value and path the path to that point
;;
(define (common:htree->html ht path tipfunc)
  (let ((datlist 	(sort (hash-table->alist ht)
;;======================================================================
;;======================================================================
                              (lambda (a b)
                                (string< (car a)(car b))))))
    (if (null? datlist)
    	(tipfunc #f path) ;; really shouldn't get here
	(s:ul
	 (map (lambda (x)
		(let* ((levelname (car x))
		       (y         (cdr x))
		       (newpath   (append path (list levelname)))
		       (leaf      (or (not (hash-table? y))
				      (null? (hash-table-keys y)))))
		  (if leaf
		      (s:li (tipfunc y newpath))
		      (s:li
		       (list 
			levelname
			(common:htree->html y newpath tipfunc))))))
	      datlist)))))

;;======================================================================


;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;; D A S H B O A R D   U S E R   V I E W S
;;======================================================================

;;======================================================================

(define (common:args-get-runname)
  (let ((res (or (args:get-arg "-runname")
		 (args:get-arg ":runname")
		 (getenv "MT_RUNNAME"))))
    ;; (if res (set-environment-variable "MT_RUNNAME" res)) ;; not sure if this is a good idea. side effect and all ...
    res))

(define (get-with-default val default)
  (let ((val (args:get-arg val)))
    (if val val default)))

(define (common:args-get-target #!key (split #f)(exit-if-bad #f))
  (let* ((keys    (if (hash-table? *configdat*) (common:get-fields *configdat*) '()))
	 (numkeys (length keys))
	 (target  (or (args:get-arg "-reqtarg")
		      (args:get-arg "-target")
		      (getenv "MT_TARGET")))
	 (tlist   (if target (string-split target "/" #t) '()))
	 (valid   (if target
		      (or (null? keys) ;; probably don't know our keys yet
			  (and (not (null? tlist))
			       (eq? numkeys (length tlist))
			       (null? (filter string-null? tlist))))
		      #f)))
    (if valid
	(if split
	    tlist
	    target)
	(if target
	    (begin
	      (debug:print-error 0 *default-log-port* "Invalid target, spaces or blanks not allowed \"" target "\", target should be: " (string-intersperse keys "/") ", have " tlist " for elements")
	      (if exit-if-bad (exit 1))
	      #f)
	    #f))))

;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;; do we honor the caches of the config files?
;;
(define (common:use-cache?)
  (let ((res #t)) ;; priority by order of evaluation
    (if *configdat* ;; sillyness here. can't use setup/use-cache to know if we can use the cached files!
	(if (equal? (configf:lookup *configdat* "setup" "use-cache") "no")
	    (set! res #f)
	    (if (equal? (configf:lookup *configdat* "setup" "use-cache") "yes")
		(set! res #t))))
    (if (args:get-arg "-no-cache")(set! res #f)) ;; overrides setting in "setup"
    (if (getenv "MT_USE_CACHE")
	(if (equal? (getenv "MT_USE_CACHE") "yes")
	    (set! res #t)
	    (if (equal? (getenv "MT_USE_CACHE") "no")
		(set! res #f))))    ;; overrides -no-cache switch
    res))

;;======================================================================
;; force use of server?
;;
(define (common:force-server?)
  (let* ((force-setting (configf:lookup *configdat* "server" "force"))
	 (force-type    (if force-setting (string->symbol force-setting) #f))
	 (force-result  (case force-type
			  ((#f)     #f)
			  ((always) #t)
			  ((test)   (if (args:get-arg "-execute") ;; we are in a test
					#t
					#f))
			  (else
			   (debug:print 0 *default-log-port* "ERROR: Bad server force setting " force-setting ", forcing server.")
			   #t)))) ;; default to requiring server
    (if force-result
	(begin
	  (debug:print-info 0 *default-log-port* "forcing use of server, force setting is \"" force-setting "\".")
	  #t)
	#f)))

(define (common:in-running-test?)
  (and (args:get-arg "-execute") (get-environment-variable "MT_CMDINFO")))
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================


;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================

;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================

;;======================================================================
;; Rotate logs, logic: 
;;                 if > 500k and older than 1 week:
;;                     remove previous compressed log and compress this log
;; WARNING: This proc operates assuming that it is in the directory above the
;;          logs directory you wish to log-rotate.
;;
(define (common:rotate-logs)
  (let* ((all-files (make-hash-table))
	 (stats     (make-hash-table))
	 (inc-stat  (lambda (key)
		      (hash-table-set! stats key (+ (hash-table-ref/default stats key 0) 1))))
	(max-allowed (string->number (or (configf:lookup *configdat* "setup" "max-logfiles") "300")))) ;; name -> age
    (if (not (directory-exists? "logs"))(create-directory "logs"))
    (directory-fold 
     (lambda (file rem)
       (handle-exceptions
	exn
	(begin
	  (debug:print-info 2 *default-log-port* "unable to rotate log " file ", probably handled by another process, this is safe to ignore. exn=" exn)
	  (debug:print 2 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	  ;; (print-call-chain (current-error-port)) ;; 
	  )
	(let* ((fullname  (conc "logs/" file))
	       (mod-time  (file-modification-time fullname))
	       (file-age  (- (current-seconds) mod-time))
	       (file-old  (> file-age (* 48 60 60)))
	       (file-big  (> (file-size fullname) 200000)))
	  (hash-table-set! all-files file mod-time)
	  (if (or (and (string-match "^.*.log" file)
		       file-old
		       file-big)
		  (and (string-match "^server-.*.log" file)
		       file-old))
	      (let ((gzfile (conc fullname ".gz")))
		(if (common:file-exists? gzfile)
		    (begin
		      (debug:print-info 0 *default-log-port* "removing " gzfile)
		      (delete-file* gzfile)
		      (hash-table-delete!  all-files gzfile) ;; needed?
		      ))
		(debug:print-info 0 *default-log-port* "compressing " file)
		(system (conc "gzip " fullname))
		(inc-stat "gzipped")
		(hash-table-set! all-files (conc file ".gz") file-age)  ;; add the .gz file and remove the base file
		(hash-table-delete! all-files file)
		)
	      (if (and (> file-age (* (string->number (or (configf:lookup *configdat* "setup" "log-expire-days") "30")) 24 3600))
		       (file-exists? fullname)) ;; just in case it was gzipped - will get it next time
		  (handle-exceptions
		   exn
		   #f
		   (if (directory? fullname)
		       (begin
			 (debug:print-info 0 *default-log-port* fullname " in logs directory is a directory! Cannot rotate it, it is best to not put subdirectories in the logs dir.")
			 (inc-stat "directories"))
		       (begin
			 (delete-file* fullname)
			 (inc-stat "deleted")))
		   (hash-table-delete! all-files file)))))))
     '()
     "logs")
    (for-each
     (lambda (category)
       (let ((quant (hash-table-ref/default stats category 0)))
	 (if (> quant 0)
	     (debug:print-info 0 *default-log-port* category " log files: " quant))))
     `("deleted" "gzipped" "directories"))
    (let ((num-logs (hash-table-size all-files)))
      (if (> num-logs max-allowed) ;; because NFS => don't let number of logs exceed 300
	  (let ((files (take (sort (hash-table-keys all-files)
				   (lambda (a b)
				     (< (hash-table-ref all-files a)(hash-table-ref all-files b))))
			     (- num-logs max-allowed))))
	    (for-each
	     (lambda (file)
	       (let* ((fullname (conc "logs/" file)))
		 (if (directory? fullname)
		     (debug:print-info 0 *default-log-port* fullname " in logs directory is a directory! Cannot rotate it, it is best to not put subdirectories in the logs dir.")
		     (handle-exceptions
		      exn
		      (debug:print-error 0 *default-log-port* "failed to remove " fullname ", exn=" exn)
		      (delete-file* fullname)))))
	     files)
	    (debug:print-info 0 *default-log-port* "Deleted " (length files) " files from logs, keeping " max-allowed " files."))))))
  
;;======================================================================
;; calculate a delay number based on a droop curve
;;    inputs are:
;;      - load-in, load as from uptime, NOT normalized
;;      - numcpus, number of cpus, ideally use the real cpus, not threads
;;
(define (common:get-delay load-in numcpus)
  (let* ((ratio (/ load-in numcpus))
         (new-option (configf:lookup *configdat* "load" "new-load-method"))
	 (paramstr   (or (configf:lookup *configdat* "load" "exp-params")
			 "15 12 1281453987.9543 0.75")) ;; 5 4 10 1"))
	 (paramlst   (map string->number (string-split paramstr))))
    (if new-option
	(begin
          (cond ((and (>= ratio 0) (< ratio .5))
		 0)
		((and (>= ratio 0.5) (<= ratio .9))
		 (* ratio (/ 5 .9)))
		((and (> ratio .9) (<= ratio 1.1))
		 (+ 5 (* (- ratio .9) (/ 55 .2)))) 
		((> ratio 1.1)
		 60)))
	(match paramlst
	  ((r1 r2 s1 s2)
	   (debug:print 3 *default-log-port* "Using params r1=" r1 " r2=" r2 " s1=" s1 " s2=" s2)
	   (min (max (/ (expt r1 (* r2 s2 ratio)) s1) 0) 30))
	  (else
	   (debug:print 0 *default-log-port* "BAD exp-params, should be \"r1 r2 s1 s2\" but got " paramstr)
	   30)))))

(define (common:print-delay-table)
  (let loop ((x 0))
    (print x "," (common:get-delay x 1))
    (if (< x 2)
	(loop (+ x 0.1)))))

;;======================================================================
;; DO NOT CALL THIS DIRECTLY. It is called from common:wait-for-normalized-load
;;   count     - count down to zero, at some point we'd give up if the load never drops
;;   num-tries - count down to zero number tries to get numcpus
;;
(define (common:wait-for-cpuload maxnormload numcpus-in
				 #!key (count 1000)
				 (msg #f)(remote-host #f)(num-tries 5))
  (let* ((loadavg     (common:get-cpu-load remote-host))
	  ;; not possible to have zero.  If we get 1, it's possible that we got the previous default, and we should check again
	 (numcpus     (if (<= 1 numcpus-in)
			  (common:get-num-cpus remote-host)
			  numcpus-in))
	 (first       (car loadavg))
	 (next        (cadr loadavg))
	 (adjmaxload  (* maxnormload (max 1 numcpus))) ;; possible bug
						   ;; where numcpus
						   ;; (or could be
						   ;; maxload) is
						   ;; zero, crude
						   ;; fallback is to
						   ;; at least use 1
	 ;; effective load accounts for load jumps, this should elminate all the first-next-avg, adjwait, load-jump-limit
	 ;; etc.
	 (effective-load    (common:get-intercept first next))
	 (recommended-delay (common:get-delay effective-load numcpus))
	 (effective-host    (or remote-host "localhost"))
	 (normalized-effective-load (/ effective-load numcpus))
	 (will-wait                 (> normalized-effective-load maxnormload)))
    (if (> recommended-delay 1) 
	(let* ((actual-delay (min recommended-delay 30)))
	  (if (common:low-noise-print 30 (conc (round actual-delay) "-safe-load"))
	      (debug:print-info 0 *default-log-port* "Load control, delaying "
				actual-delay " seconds to maintain safe load. current normalized effective load is "
				normalized-effective-load". maxnormload = " maxnormload " numcpus = " numcpus " loadavg = " loadavg " effective-load = " effective-load))
	  (thread-sleep! actual-delay)))
    
    (cond
     ;; bad data, try again to get the data
     ((not will-wait)
      (if (common:low-noise-print 3600 (conc (round normalized-effective-load) "-load-acceptable-" effective-host))
      	  (debug:print 0 *default-log-port* "Effective load on " effective-host " is acceptable at " effective-load " continuing.")))

     ((and (< first 0) ;; this indicates the loadavg data is bad - machine may not be reachable
	   (> num-tries 0))
      (debug:print 0 *default-log-port* "WARNING: received bad data from get-cpu-load "
		   first ", we'll sleep 10s and try " num-tries " more times.")
      (thread-sleep! 10)
      (common:wait-for-cpuload maxnormload numcpus-in
			       count: count remote-host: remote-host num-tries: (- num-tries 1)))

     ;; need to wait for load to drop
     ((and will-wait ;; (> first adjmaxload)
	   (> count 0))
      (debug:print-info 0 *default-log-port*
			"Delaying 15" ;; adjwait
			" seconds due to normalized effective load " normalized-effective-load ;; first
			" exceeding max of " adjmaxload
			" on server " (or remote-host (get-host-name))
			" (normalized load-limit: " maxnormload ") " (if msg msg ""))
      (thread-sleep! 15) ;; adjwait)
      (common:wait-for-cpuload maxnormload numcpus count: (- count 1) msg: msg remote-host: remote-host)
      ;; put the message here to indicate came out of waiting
      (debug:print-info 1 *default-log-port*
			"On host: " effective-host
			", effective load: " effective-load
			", numcpus: " numcpus
			", normalized effective load: " normalized-effective-load
			))
     ;; overloaded and count expired (i.e. went to zero)
     (else
      (if (> num-tries 0) ;; should be "num-tries-left". 
	  (if (common:low-noise-print 30 (conc (round effective-load) "-load-acceptable-" effective-host))
	      (debug:print 0 *default-log-port* "Load on " effective-host " is acceptable at effective normalized load of "
			   normalized-effective-load "  continuing."))
	  (debug:print 0 *default-log-port* "Load on " effective-host ", "
		       first" could not be retrieved. Giving up and continuing."))))))

;;======================================================================
;; wait for normalized cpu load to drop below maxload
;;
(define (common:wait-for-normalized-load maxnormload msg remote-host #!optional (rem-tries 5))
  (let ((num-cpus (common:get-num-cpus remote-host)))
    (if num-cpus
	(common:wait-for-cpuload maxnormload num-cpus 15 msg: msg remote-host: remote-host)
	(begin
	  (thread-sleep! (random 60)) ;; we failed to get num cpus. wait a bit and try again
	  (if (> rem-tries 0)
	      (common:wait-for-normalized-load maxnormload msg remote-host (- rem-tries 1))
	      #f)))))

;;======================================================================
;;======================================================================
;; given path get free space, allows override in [setup]
;; with free-space-script /path/to/some/script.sh
;;
(define (get-df path)
  (if (configf:lookup *configdat* "setup" "free-space-script")
      (with-input-from-pipe 
       (conc (configf:lookup *configdat* "setup" "free-space-script") " " path)
       (lambda ()
	 (let ((res (read-line)))
	   (if (string? res)
	       (string->number res)))))
      (get-unix-df path)))

(define (common:check-space-in-dir dirpath required)
  (let* ((dbspace  (if (directory? dirpath)
		       (get-df dirpath)
		       0)))
    (list (> dbspace required)
	  dbspace
	  required
	  dirpath)))

(define (get-free-inodes path)
  (if (configf:lookup *configdat* "setup" "free-inodes-script")
      (with-input-from-pipe 
	  (conc (configf:lookup *configdat* "setup" "free-inodes-script") " " path)
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================

	(lambda ()
	  (let ((res (read-line)))
	    (if (string? res)
		(string->number res)))))
      (get-unix-inodes path)))
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================

;; 		(let ((cmddat (make-qitem
;; 			       command:   command
;; 			       host-port: host-port
;; 			       params:    params)))
;; 		  (queue-push cmddat) 		;; put request into the queue
;; 		  (nn-send soc "queued"))         ;; reply with "queued"
;; 		(print "ERROR: ["(common:human-time)"] BAD request " dat))
;; 	    (loop (nn-recv soc)))))
;;     (nn-close soc)))
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
;;======================================================================
(define (common:get-pkts-dirs mtconf use-lt)
  (let* ((pktsdirs-str (or (configf:lookup mtconf "setup"  "pktsdirs")
			   (and use-lt
				(conc (or *toppath*
					  (current-directory))
				      "/lt/.pkts"))))
	 (pktsdirs  (if pktsdirs-str
			(string-split pktsdirs-str " ")
			#f)))
    pktsdirs))

;;======================================================================
(define (common:with-queue-db mtconf proc #!key (use-lt #f)(toppath-in #f))
  (let* ((pktsdirs (common:get-pkts-dirs mtconf use-lt))
	 (pktsdir  (if pktsdirs (car pktsdirs) #f))
	 (toppath  (or (configf:lookup mtconf "scratchdat" "toppath")
		       toppath-in))
	 (pdbpath  (or (configf:lookup mtconf "setup"  "pdbpath") pktsdir)))
    (cond
     ((not (and  pktsdir toppath pdbpath))
      (debug:print 0 *default-log-port* "ERROR: settings are missing in your megatest.config for area management.")
      (debug:print  0 *default-log-port* "  you need to have pktsdirs in the [setup] section."))
     ((not (common:file-exists? pktsdir))
      (debug:print 0 *default-log-port* "ERROR: pkts directory not found " pktsdir))
     ((not (equal? (file-owner pktsdir)(current-effective-user-id)))
      (debug:print 0 *default-log-port* "ERROR: directory " pktsdir " is not owned by " (current-effective-user-name)))
     (else
	(let* ((pdb  (open-queue-db pdbpath "pkts.db"
				    schema: '("CREATE TABLE groups (id INTEGER PRIMARY KEY,groupname TEXT, CONSTRAINT group_constraint UNIQUE (groupname));"))))
	  (proc pktsdirs pktsdir pdb)
	  (dbi:close pdb))))))

;;======================================================================
;; check space in dbdir and in megatest dir
;; returns: ok/not dbspace required-space
;;
(define (common:check-db-dir-space)
  (let* ((required (string->number 
                    ;; default is 1GB (or actually a billion bytes) This is the number of 1 kB blocks.
		    (or (configf:lookup *configdat* "setup" "dbdir-space-required")
			"1000000")))
	 (dbdir    (common:get-db-tmp-area)) ;; (db:get-dbdir))
	 (tdbspace (common:check-space-in-dir dbdir required))
	 (mdbspace (common:check-space-in-dir *toppath* required)))
    (sort (list tdbspace mdbspace) (lambda (a b)
				     (< (cadr a)(cadr b))))))

;;======================================================================
;; check available space in dbdir, exit if insufficient
;;
(define (common:check-db-dir-and-exit-if-insufficient)
  (let* ((spacedat (car (common:check-db-dir-space))) ;; look only at worst for now
	 (is-ok    (car spacedat))
	 (dbspace  (cadr spacedat))
	 (required (caddr spacedat))
	 (dbdir    (cadddr spacedat)))
    (if (not is-ok)
	(begin
	  (debug:print-error 0 *default-log-port* "Insufficient space in " dbdir ", require " required ", have " dbspace  ", exiting now.")
	  (exit 1)))))

;;======================================================================
;; paths is list of lists ((name path) ... )
;;
(define (common:get-disk-with-most-free-space disks minsize)
  (let* ((best     #f)
	(bestsize 0)
        (default-min-inodes-string "1000000")
        (default-min-inodes (string->number default-min-inodes-string))
        (min-inodes (or (string->number (if (configf:lookup *configdat* "setup" "min_inodes") (configf:lookup *configdat* "setup" "min_inodes") default-min-inodes-string)) default-min-inodes)))

    (for-each 
     (lambda (disk-num)
       (let* ((dirpath    (cadr (assoc disk-num disks)))
	      (freespc    (cond
			   ((not (directory? dirpath))
			    (if (common:low-noise-print 300 "disks not a dir " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a directory - ignoring it."))
			    -1)
			   ((not (file-write-access? dirpath))
			    (if (common:low-noise-print 300 "disks not writeable " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not writeable - ignoring it."))
			    -1)
			   ((not (eq? (string-ref dirpath 0) #\/))
			    (if (common:low-noise-print 300 "disks not a proper path " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a fully qualified path - ignoring it."))
			    -1)
			   (else
			    (get-df dirpath))))
	      (free-inodes (cond
			   ((not (directory? dirpath))
			    (if (common:low-noise-print 300 "disks not a dir " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a directory - ignoring it."))
			    -1)
			   ((not (file-write-access? dirpath))
			    (if (common:low-noise-print 300 "disks not writeable " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not writeable - ignoring it."))
			    -1)
			   ((not (eq? (string-ref dirpath 0) #\/))
			    (if (common:low-noise-print 300 "disks not a proper path " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a fully qualified path - ignoring it."))
			    -1)
			   (else
			    (get-free-inodes dirpath))))
             ;;(free-inodes (get-free-inodes dirpath))
             )
             (debug:print 2 *default-log-port* "INFO: disk " disk-num " path " dirpath " free space " freespc " free inodes " free-inodes)
	 (if (and (> freespc bestsize)(> free-inodes min-inodes ))
	     (begin
	       (set! best     (cons disk-num dirpath))
	       (set! bestsize freespc)))
        ;;(print "Processing: " disk-num " bestsize: " bestsize " best: " best " freespc: " freespc " min-inodes: " min-inodes " free-inodes: " free-inodes)
      ))
     (map car disks))
    (if (and best (> bestsize minsize))
	best
	#f))) ;; #f means no disk candidate found

(define (common:load-pkts-to-db mtconf #!key (use-lt #f))
  (common:with-queue-db
   mtconf
   (lambda (pktsdirs pktsdir pdb)
     (for-each
      (lambda (pktsdir) ;; look at all
	(cond
	 ((not (common:file-exists? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory " pktsdir " does not exist."))
	 ((not (directory? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not a directory."))
	 ((not (file-read-access? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not readable."))
	 (else
	  (debug:print-info 0 *default-log-port* "Loading packets found in " pktsdir)
	  (let ((pkts (glob (conc pktsdir "/*.pkt"))))
	    (for-each
	     (lambda (pkt)
	       (let* ((uuid    (cadr (string-match ".*/([0-9a-f]+).pkt" pkt)))
		      (exists  (lookup-by-uuid pdb uuid #f)))
		 (if (not exists)
		     (let* ((pktdat (string-intersperse
				     (with-input-from-file pkt read-lines)
				     "\n"))
			    (apkt   (pkt->alist pktdat))
			    (ptype  (alist-ref 'T apkt)))
		       (add-to-queue pdb pktdat uuid (or ptype 'cmd) #f 0)
		       (debug:print 4 *default-log-port* "Added " uuid " of type " ptype " to queue"))
		     (debug:print 4 *default-log-port* "pkt: " uuid " exists, skipping...")
		     )))
	     pkts)))))
      pktsdirs))
   use-lt: use-lt))

;;======================================================================
;; use-lt is use linktree "lt" link to find pkts dir
(define (common:save-pkt pktalist-in mtconf use-lt #!key (add-only #f)) ;; add-only saves the pkt only if there is a parent already
  (if (or add-only
	  (hash-table-exists? *pkts-info* 'last-parent))
      (let* ((parent   (hash-table-ref/default *pkts-info* 'last-parent #f))
	     (pktalist (if parent
			   (cons `(parent . ,parent)
				 pktalist-in)
			   pktalist-in)))
	(let-values (((uuid pkt)
		      (alist->pkt pktalist common:pkts-spec)))
	  (hash-table-set! *pkts-info* 'last-parent uuid)
	  (let ((pktsdir (or (hash-table-ref/default *pkts-info* 'pkts-dir #f)
			     (let* ((pktsdirs (common:get-pkts-dirs mtconf use-lt))
				    (pktsdir   (car pktsdirs))) ;; assume it is there
			       (hash-table-set! *pkts-info* 'pkts-dir pktsdir)
			       pktsdir))))
            (handle-exceptions
             exn
             (debug:print-info 0 "failed to write out packet to " pktsdir ", exn=" exn) ;; don't care if this failed for now but MUST FIX - BUG!!
             (if (not (file-exists? pktsdir))
                 (create-directory pktsdir #t))
             (with-output-to-file
                 (conc pktsdir "/" uuid ".pkt")
               (lambda ()
                 (print pkt)))))))))
;;        (set! *common:telemetry-log-socket* #f)))))
	

(define (common:get-linktree)
  (or (getenv "MT_LINKTREE")
      (if *configdat*
	  (configf:lookup *configdat* "setup" "linktree")
	  #f)
      (if (or *toppath* (getenv "MT_RUN_AREA_HOME"))
	  (conc (or *toppath* (getenv "MT_RUN_AREA_HOME")) "/lt")
	  #f)
      (let* ((tp (common:get-toppath #f))
	     (lt (conc tp "/lt")))
	(if (not tp)(debug:print 0 *default-log-port* "WARNING: cannot calculate best path for linktree, using " lt))
	lt)))



;;======================================================================the end
)

(import configfmod)

(declare (uses margsmod))
(import margsmod)

(include "common_records.scm")






















































































































































































































































































































































































































































































































































































































































































































































































;; if 
(define (configf:read-alist fname)
  (handle-exceptions
      exn
    (begin
      (debug:print 0 *default-log-port* "read of alist " fname " failed. exn=" exn)
      #f)

			(debug:print 0 *default-log-port* "WARNING: content " dat " for cache " fname " is not readable. Deleting generated file.")
			(delete-file fname))
                      #f))
                #f))))
    (common:faux-unlock fname)
    res))
  

;; 
;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;;======================================================================

(declare (unit configfmod))
(declare (uses margsmod))

(module configfmod
	*
	
(import scheme chicken data-structures extras files ports)
(import margsmod)

(use
 (prefix base64 base64:)
 (prefix dbi dbi:)
 (prefix sqlite3 sqlite3:)
 (srfi 18)
 format
 matchable

 srfi-69
 stack
 typed-records
 directory-utils
 z3
 )



;;======================================================================
;; move debug stuff to separate module then put these back where they belong
;;======================================================================





























































;;======================================================================
;; lookup routines - replicated from configf
;;======================================================================

(define (configf:assoc-safe-add alist key val #!key (metadata #f))
  (let ((newalist (filter (lambda (x)(not (equal? key (car x)))) alist)))
    (append newalist (list (if metadata
			       (list key val metadata)
			       (list key val))))))

(define (configf:section-var-set! cfgdat section-name var value #!key (metadata #f))
  (hash-table-set! cfgdat section-name
		   (configf:assoc-safe-add
		    (hash-table-ref/default cfgdat section-name '())
		    var value metadata: metadata)))













;; use to have definitive setting:
;;  [foo]
;;  var yes
;;
;;  (configf:var-is? cfgdat "foo" "var" "yes") => #t
;;
(define (configf:var-is? cfgdat section var expected-val)

(define (configf:set-section-var cfgdat section var val)
  (let ((sectdat (configf:get-section cfgdat section)))
    (hash-table-set! cfgdat section
                     (configf:assoc-safe-add sectdat var val))))

;;======================================================================the end

;; return list (path fullpath configname)
(define (find-config configname #!key (toppath #f))
  (if toppath
      (let ((cfname (conc toppath "/" configname)))
	(if (common:file-exists? cfname)
	    (list toppath cfname configname)
	    (list #f      #f     #f)))
      (let* ((cwd (string-split (current-directory) "/")))
	(let loop ((dir cwd))
	  (let* ((path     (conc "/" (string-intersperse dir "/")))
		 (fullpath (conc path "/" configname)))
	    (if (common:file-exists? fullpath)
		(list path fullpath configname)
		(let ((remcwd (take dir (- (length dir) 1))))
		  (if (null? remcwd)
		      (list #f #f #f) ;;  #f #f) 
		  (loop remcwd)))))))))

(define (configf:eval-string-in-environment str)
  ;; (if (or (string-null? str)
  ;;	  (equal? "!" (substring str 0 1))) ;; null string or starts with ! are preserved but NOT set in the environment
      str
      (handle-exceptions
       exn
       (begin
	 (debug:print-error 0 *default-log-port* "problem evaluating \"" str "\" in the shell environment, exn=" exn)
	 #f)
       (let ((cmdres (process:cmd-run->list (conc "echo " str))))
	 (if (null? cmdres) ""
	     (caar cmdres))))) ;; )

;;======================================================================
;; Make the regexp's needed globally available
;;======================================================================

(define configf:include-rx (regexp "^\\[include\\s+(.*)\\]\\s*$"))
(define configf:script-rx  (regexp "^\\[scriptinc\\s+(\\S+)([^\\]]*)\\]\\s*$")) ;; include output from a script
(define configf:section-rx (regexp "^\\[(.*)\\]\\s*$"))
(define configf:blank-l-rx (regexp "^\\s*$"))
(define configf:key-sys-pr (regexp "^(\\S+)\\s+\\[system\\s+(\\S+.*)\\]\\s*$"))
(define configf:key-val-pr (regexp "^(\\S+)(\\s+(.*)|())$"))
(define configf:key-no-val (regexp "^(\\S+)(\\s*)$"))
(define configf:comment-rx (regexp "^\\s*#.*"))
(define configf:cont-ln-rx (regexp "^(\\s+)(\\S+.*)$"))
(define configf:settings   (regexp "^\\[configf:settings\\s+(\\S+)\\s+(\\S+)]\\s*$"))

;; read a line and process any #{ ... } constructs

(define configf:var-expand-regex (regexp "^(.*)#\\{(scheme|system|shell|getenv|get|runconfigs-get|rget|scm|sh|rp|gv|g|mtrah)\\s+([^\\}\\{]*)\\}(.*)"))

(define (configf:system ht cmd)
  (system cmd)
  )

(define (configf:process-line l ht allow-system #!key (linenum #f))
  (let loop ((res l))
    (if (string? res)
	(let ((matchdat (string-search configf:var-expand-regex res)))
	  (if matchdat
	      (let* ((prestr  (list-ref matchdat 1))
		     (cmdtype (list-ref matchdat 2)) ;; eval, system, shell, getenv
		     (cmd     (list-ref matchdat 3))
		     (poststr (list-ref matchdat 4))
		     (result  #f)
		     (start-time (current-seconds))
		     (cmdsym  (string->symbol cmdtype))
		     (fullcmd (case cmdsym
				((scheme scm) (conc "(lambda (ht)" cmd ")"))
				((system)     (conc "(lambda (ht)(configf:system ht \"" cmd "\"))"))
				((shell sh)   (conc "(lambda (ht)(string-translate (shell \""  cmd "\") \"\n\" \" \"))"))
				((realpath rp)(conc "(lambda (ht)(common:nice-path \"" cmd "\"))"))
				((getenv gv)  (conc "(lambda (ht)(get-environment-variable \"" cmd "\"))"))
				((mtrah)      (conc "(lambda (ht)"
                                                    "    (let ((extra \"" cmd "\"))"
						    "       (conc (or *toppath* (get-environment-variable \"MT_RUN_AREA_HOME\"))"
						    "             (if (string-null? extra) \"\" \"/\")"
						    "             extra)))"))
				((get g)   
				 (match (string-split cmd)
					((sect var)(conc "(lambda (ht)(configfmod#configf:lookup ht \"" sect "\" \"" var "\"))"))
					(else
					 (debug:print-error 0 *default-log-port* "#{get ...} used with only one parameter, \"" cmd "\", two needed.")
					 "(lambda (ht) #f)")))
				((runconfigs-get rget) (conc "(lambda (ht)(runconfigs-get ht \"" cmd "\"))"))
				;; ((rget)           (conc "(lambda (ht)(runconfigs-get ht \"" cmd "\"))"))
				(else "(lambda (ht)(print \"ERROR\") \"ERROR\")"))))
		;; (print "fullcmd=" fullcmd)
		(handle-exceptions
		 exn
		 (begin
		   (debug:print 0 *default-log-port* "WARNING: failed to process config input \"" l "\", exn=" exn)
		   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
		   ;; (print "exn=" (condition->list exn))
		   (set! result (conc "#{( " cmdtype ") " cmd "}, full expansion: " fullcmd)))
		 (if (or allow-system
			 (not (member cmdtype '("system" "shell" "sh"))))
		     (with-input-from-string fullcmd
		       (lambda ()
			 (set! result ((eval (read)) ht))))
		     (set! result (conc "#{(" cmdtype ") "  cmd "}"))))
		(case cmdsym
		  ((system shell scheme)
		   (let ((delta (- (current-seconds) start-time)))
		     (if (> delta 2)
			 (debug:print-info 0 *default-log-port* "for line \"" l "\"\n command:  " cmd " took " delta " seconds to run with output:\n   " result)
			 (debug:print-info 9 *default-log-port* "for line \"" l "\"\n command:  " cmd " took " delta " seconds to run with output:\n   " result)))))
		(loop (conc prestr result poststr)))
	      res))
	res)))

;; Run a shell command and return the output as a string
(define (shell cmd)
  (let* ((output (process:cmd-run->list cmd))
	 (res    (car output))
	 (status (cadr output)))
    (if (equal? status 0)
	(let ((outres (string-intersperse 
		       res
		       "\n")))
	  (debug:print-info 4 *default-log-port* "shell result:\n" outres)
	  outres)
	(begin
	  (with-output-to-port (current-error-port)
	    (lambda ()
	      (print "ERROR: " cmd " returned bad exit code " status)))
	  ""))))

;; this was inline but I'm pretty sure that is a hold over from when it was *very* simple ...
;;
(define (configf:read-line p ht allow-processing settings)
  (let loop ((inl (read-line p)))
    (let ((cont-line (and (string? inl)
			  (not (string-null? inl))
			  (equal? "\\" (string-take-right inl 1)))))
      (if cont-line ;; last character is \ 
	  (let ((nextl (read-line p)))
	    (if (not (eof-object? nextl))
		(loop (string-append (if cont-line 
					 (string-take inl (- (string-length inl) 1))
					 inl)
				     nextl))))
	  (let ((res (case allow-processing ;; if (and allow-processing 
		       ;;	   (not (eq? allow-processing 'return-string)))
		       ((#t #f)
			(configf:process-line inl ht allow-processing))
		       ((return-string)
			inl)
		       (else
			(configf:process-line inl ht allow-processing)))))
	    (if (and (string? res)  ;; must set to "no" to force NOT trimming trailing spaces
		     (not (equal? (hash-table-ref/default settings "trim-trailing-spaces" "yes") "no")))
		(string-substitute "\\s+$" "" res)
		res))))))

(define (configf:cfgdat->env-alist section cfgdat-ht allow-system)
  (filter
   (lambda (pair)
     (let* ((var (car pair))
            (val (cdr pair)))
       (cons var
             (cond
              ((and allow-system (procedure? val)) ;; if we decided to use something other than #t or #f for allow-system ('return-procs or 'return-string) , this may become problematic
               (val))
              ((procedure? val) #f)
              ((string? val) val)
              (else "#f")))))
   (append
    (hash-table-ref/default cfgdat-ht "default" '())
    (if (equal? section "default") '() (hash-table-ref/default cfgdat-ht section '())))))

(define (calc-allow-system allow-system section sections)
  (if sections
      (and (or (equal? "default" section)
	       (member section sections))
	   allow-system) ;; account for sections and return allow-system as it might be a symbol such as return-strings
      allow-system))
    
;; given a config hash and a section name, apply that section to all matching sections (using wildcard % or regex if /..../)
;; remove the section when done so that there is no downstream clobbering
;;
(define (configf:apply-wildcards ht section-name)
  (if (hash-table-exists? ht section-name)
      (let* ((vars  (hash-table-ref ht section-name))
	     (rxstr (if (string-contains section-name "%")
			(string-substitute (regexp "%") ".*" section-name)
			(string-substitute (regexp "^/(.*)/$") "\\1" section-name)))
	     (rx    (regexp rxstr)))
	;; (print "\nsection-name: " section-name " rxstr: " rxstr)
        (for-each
         (lambda (section)
	   (if section
	       (let ((same-section (string=? section-name section))
		     (rx-match     (string-match rx section)))
		 ;; (print "section: " section " vars: " vars " same-section: " same-section " rx-match: " rx-match)
		 (if (and (not same-section) rx-match)
		     (for-each
		      (lambda (bundle)
			;; (print "bundle: " bundle)
			(let ((key  (car bundle))
			      (val  (cadr bundle))
			      (meta (if (> (length bundle) 2)(caddr bundle) #f)))
			  (hash-table-set! ht section (configf:assoc-safe-add (hash-table-ref ht section) key val metadata: meta))))
		      vars)))))
         (hash-table-keys ht))))
  ht)

;; read a config file, returns hash table of alists

;; read a config file, returns hash table of alists
;; adds to ht if given (must be #f otherwise)
;; allow-system:
;;    #f - do not evaluate [system
;;    #t - immediately evaluate [system and store result as string
;;    'return-procs -- return a proc taking ht as an argument that may be evaulated at some future time
;;    'return-string -- return a string representing a proc taking ht as an argument that may be evaulated at some future time
;; envion-patt is a regex spec that identifies sections that will be eval'd
;; in the environment on the fly
;; sections: #f => get all, else list of sections to gather
;; post-section-procs alist of section-pattern => proc, where: (proc section-name next-section-name ht curr-path)
;; apply-wildcards: #t/#f - apply vars from targets with % wildcards to all matching sections
;;
(define (read-config path ht allow-system #!key (environ-patt #f)            (curr-section #f)   
		     (sections #f)              (settings (make-hash-table)) (keep-filenames #f)
		     (post-section-procs '())   (apply-wildcards #t) )
  (debug:print 9 *default-log-port* "START: " path)
;; (if *configdat*
;;     (common:save-pkt `((action . read-config)
;;       		 (f      . ,(cond ((string? path) path)
;;       				  ((port?   path) "port")
;;       				  (else (conc path))))
;;                        (T      . configf))
;;       	       *configdat* #t add-only: #t))
  (if (and (not (port? path))
	   (not (common:file-exists? path))) ;; for case where we are handed a port
      (begin 
	(debug:print-info 1 *default-log-port* "read-config - file not found " path " current path: " (current-directory))
	;; WARNING: This is a risky change but really, we should not return an empty hash table if no file read?
	#f) ;; (if (not ht)(make-hash-table) ht))
      (let ((inp        (if (string? path)
			    (open-input-file path)
			      path)) ;; we can be handed a port
	    (res        (if (not ht)(make-hash-table) ht))
	    (metapath   (if (or (debug:debug-mode 9)
				keep-filenames)
			    path #f))
            (process-wildcards  (lambda (res curr-section-name)
                                  (if (and apply-wildcards
                                           (or (string-contains curr-section-name "%")   ;; wildcard
                                               (string-match "/.*/" curr-section-name))) ;; regex
                                      (begin
                                        (configf:apply-wildcards res curr-section-name)
                                        (hash-table-delete! res curr-section-name))))))  ;; NOTE: if the section is a wild card it will be REMOVED from res 
	(let loop ((inl               (configf:read-line inp res (calc-allow-system allow-system curr-section sections) settings)) ;; (read-line inp))
		   (curr-section-name (if curr-section curr-section "default"))
		   (var-flag #f);; turn on for key-var-pr and cont-ln-rx, turn off elsewhere
		   (lead     #f))
	  (debug:print-info 8 *default-log-port* "curr-section-name: " curr-section-name " var-flag: " var-flag "\n   inl: \"" inl "\"")
	  (if (eof-object? inl) 
	      (begin
                ;; process last section for wildcards
                (process-wildcards res curr-section-name)
		(if (string? path) ;; we received a path, not a port, thus we are responsible for closing it.
		    (close-input-port inp))
		(if (list? sections) ;; delete all sections except given when sections is provided
		    (for-each
		     (lambda (section)
		       (if (not (member section sections))
			   (hash-table-delete! res section))) ;; we are using "" as a dumping ground and must remove it before returning the ht
		     (hash-table-keys res)))
		(debug:print 9 *default-log-port* "END: " path)
                res
                ) ;; retval
	      (regex-case 
	       inl 
	       (configf:comment-rx _                  (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings)
                                                            curr-section-name #f #f))
               
	       (configf:blank-l-rx _                  (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings)
                                                            curr-section-name #f #f))
	       (configf:settings   ( x setting val  )
                                   (begin
                                     (hash-table-set! settings setting val)
                                     (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings)
                                           curr-section-name #f #f)))
               
	       (configf:include-rx ( x include-file )
                                   (let* ((curr-conf-dir (pathname-directory path))
                                          (full-conf     (if (and (absolute-pathname? include-file) (file-exists? include-file))
                                                             include-file
                                                             (common:nice-path 
                                                              (conc (if curr-conf-dir
                                                                        curr-conf-dir
                                                                        ".")
                                                                    "/" include-file)))))
				     (let ((all-matches (sort (handle-exceptions exn
								(begin
								 (debug:print '(2 9) *default-log-port* "glob of " full-conf " gave no match. , exn=" exn)
								 (list))
								(glob full-conf)) string<=?)))
				       (if (null? all-matches)
					   (begin
					     (debug:print '(2 9) #f "INFO: include file(s) matching " include-file " not found (called from " path ")")
					     (debug:print 2 *default-log-port* "        " full-conf))
					   (for-each
					    (lambda (fpath)
					      ;; (push-directory conf-dir)
					      (debug:print 9 *default-log-port* "Including: " full-conf)
					      (read-config fpath res allow-system environ-patt: environ-patt
							   curr-section: curr-section-name sections: sections settings: settings
							   keep-filenames: keep-filenames))
					    all-matches))
				       (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings)
					     curr-section-name #f #f))))
	       (configf:script-rx ( x include-script params);; handle-exceptions
                                  ;;    exn
                                  ;;    (begin
                                  ;;      (debug:print '(0 2 9) #f "INFO: include from script " include-script " failed.")
                                  ;;      (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f))
                                  (if (and (common:file-exists? include-script)(file-execute-access? include-script))
                                      (let* ((local-allow-system  (calc-allow-system allow-system curr-section-name sections))
                                             (env-delta  (configf:cfgdat->env-alist curr-section-name res local-allow-system))
                                             (new-inp-port
                                              (common:with-env-vars
                                               env-delta
                                               (lambda ()
                                                 (open-input-pipe (conc include-script " " params))))))
                                        (debug:print '(2 9) *default-log-port* "Including from script output: " include-script)
                                        ;;  (print "We got here, calling read-config next. Port is: " new-inp-port)
                                        (read-config new-inp-port res allow-system environ-patt: environ-patt curr-section: curr-section-name sections: sections settings: settings keep-filenames: keep-filenames)
                                        (close-input-port new-inp-port)
                                        (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f))
                                      (begin
                                        (debug:print 0 *default-log-port* "Script not found or not exectutable: " include-script)
                                        (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f)))
                                  ) ;; )
	       (configf:section-rx ( x section-name )
                                   (begin
                                     ;; call post-section-procs
                                     (for-each 
                                      (lambda (dat)
                                        (let ((patt (car dat))
                                              (proc (cdr dat)))
                                          (if (string-match patt curr-section-name)
                                              (proc curr-section-name section-name res path))))
                                      post-section-procs)
                                     ;; after gathering the vars for a section and if apply-wildcards is true and if there is a wildcard in the section name process wildcards
                                     ;; NOTE: we are processing the curr-section-name, NOT section-name.
                                     (process-wildcards res curr-section-name)
                                     (if (not (hash-table-ref/default res section-name #f))(hash-table-set! res section-name '())) ;; ensure that mere mention of a section is not lost
                                     (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings)
                                           ;; if we have the sections list then force all settings into "" and delete it later?
                                           ;; (if (or (not sections) 
                                           ;;	      (member section-name sections))
                                           ;;	  section-name "") ;; stick everything into "". NOPE: We need new strategy. Put stuff in correct sections and then delete all sections later.
                                           section-name
                                           #f #f)))
	       (configf:key-sys-pr ( x key cmd      )
                                   (if (calc-allow-system allow-system curr-section-name sections)
                                       (let ((alist    (hash-table-ref/default res curr-section-name '()))
                                             (val-proc (lambda ()
                                                         (let* ((start-time (current-seconds))
                                                                (local-allow-system  (calc-allow-system allow-system curr-section-name sections))
                                                                (env-delta  (configf:cfgdat->env-alist curr-section-name res local-allow-system))
                                                                (cmdres     (process:cmd-run->list cmd delta-env-alist-or-hash-table: env-delta)) ;; BB: here is where [system is exec'd.  needs to have env from other vars!
                                                                (delta      (- (current-seconds) start-time))
                                                                (status     (cadr cmdres))
                                                                (res        (car  cmdres)))
                                                           (debug:print-info 4 *default-log-port* "" inl "\n => " (string-intersperse res "\n"))
                                                           (if (not (eq? status 0))
                                                               (begin
                                                                 (debug:print-error 0 *default-log-port* "problem with " inl ", return code " status
                                                                                    " output: " cmdres)))
                                                           (if (> delta 2)
                                                               (debug:print-info 0 *default-log-port* "for line \"" inl "\"\n  command: " cmd " took " delta " seconds to run with output:\n   " res)
                                                               (debug:print-info 9 *default-log-port* "for line \"" inl "\"\n  command: " cmd " took " delta " seconds to run with output:\n   " res))
                                                           (if (null? res)
                                                               ""
                                                               (string-intersperse res " "))))))
                                         (hash-table-set! res curr-section-name 
                                                          (configf:assoc-safe-add alist
                                                                                 key 
                                                                                 (case (calc-allow-system allow-system curr-section-name sections)
                                                                                   ((return-procs) val-proc)
                                                                                   ((return-string) cmd)
                                                                                   (else (val-proc)))
                                                                                 metadata: metapath))
                                         (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f))
                                       (loop (configf:read-line inp res
                                                                (calc-allow-system allow-system curr-section-name sections)
                                                                settings)
                                             curr-section-name #f #f)))
               
	       (configf:key-no-val ( x key val)
                                   (let* ((alist   (hash-table-ref/default res curr-section-name '()))
                                          (fval    (or (if (string? val) val #f) ""))) ;; fval should be either "" or " " (one or more spaces)
                                     (debug:print 10 *default-log-port* "   setting: [" curr-section-name "] " key " = #t")
                                     (safe-setenv key fval)
                                     (hash-table-set! res curr-section-name 
                                                      (configf:assoc-safe-add alist key fval metadata: metapath))
                                     (loop (configf:read-line inp res
                                                              (calc-allow-system allow-system curr-section-name sections)
                                                              settings)
                                           curr-section-name key #f)))
               
	       (configf:key-val-pr ( x key unk1 val unk2 )
                                   (let* ((alist   (hash-table-ref/default res curr-section-name '()))
                                          (envar   (and environ-patt
							(string-search (regexp environ-patt) curr-section-name) ;; does the section match the envionpatt?
							(and (not (string-null? key))
							     (not (equal? "!" (substring key 0 1)))) ;; ! as leading character is a signature to NOT export to the environment
							;; (string-match "^.*:.*:.*$" key) ;; ;; something:something:something reserved for triggers in runconfigs
							)) 
                                          (realval (if envar
                                                       (configf:eval-string-in-environment val)
                                                       val)))
                                     (debug:print-info 6 *default-log-port* "read-config env setting, envar: " envar " realval: " realval " val: " val " key: " key " curr-section-name: " curr-section-name)
                                     (if envar (safe-setenv key realval))
                                     (debug:print 10 *default-log-port* "   setting: [" curr-section-name "] " key " = " val)
                                     (hash-table-set! res curr-section-name 
                                                      (configf:assoc-safe-add alist key realval metadata: metapath))
                                     (loop (configf:read-line inp res
                                                              (calc-allow-system allow-system curr-section-name sections) settings)
                                           curr-section-name key #f)))
	       ;; if a continued line
	       (configf:cont-ln-rx ( x whsp val     )
                                   (let ((alist (hash-table-ref/default res curr-section-name '())))
                                     (if var-flag             ;; if set to a string then we have a continued var
                                         (let ((newval (conc 
                                                        (configf:lookup res curr-section-name var-flag) "\n"
                                                        ;; trim lead from the incoming whsp to support some indenting.
                                                        (if lead
                                                            (string-substitute (regexp lead) "" whsp)
                                                            "")
                                                        val)))
                                           ;; (print "val: " val "\nnewval: \"" newval "\"\nvarflag: " var-flag)
                                           (hash-table-set! res curr-section-name 
                                                            (configf:assoc-safe-add alist var-flag newval metadata: metapath))
                                           (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name var-flag (if lead lead whsp)))
                                         (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f))))
	       (else (debug:print-error 0 *default-log-port* "problem parsing " path ",\n   \"" inl "\"")
		     (set! var-flag #f)
		     (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f))))
          ) ;; end loop
        )))
  
;; pathenvvar will set the named var to the path of the config
(define (find-and-read-config fname #!key (environ-patt #f)(given-toppath #f)(pathenvvar #f))
  (let* ((curr-dir   (current-directory))
         (configinfo (find-config fname toppath: given-toppath))
	 (toppath    (car configinfo))
	 (configfile (cadr configinfo))
	 (set-fields (lambda (curr-section next-section ht path)
		       (let ((field-names (if ht (common:get-fields ht) '()))
			     (target      (or (getenv "MT_TARGET")(args:get-arg "-reqtarg")(args:get-arg "-target"))))
			 (debug:print-info 9 *default-log-port* "set-fields with field-names=" field-names " target=" target " curr-section=" curr-section " next-section=" next-section " path=" path " ht=" ht)
			 (if (not (null? field-names))(keys:target-set-args field-names target #f))))))
    (if toppath (change-directory toppath)) 
    (if (and toppath pathenvvar)(setenv pathenvvar toppath))
    (let ((configdat  (if configfile 
			  (read-config configfile #f #t environ-patt: environ-patt post-section-procs: (list (cons "^fields$" set-fields)) #f))))
      (if toppath (change-directory curr-dir))
      (list configdat toppath configfile fname))))

;;======================================================================
;; lookup and manipulation routines
;;======================================================================

;; (define (configf:assoc-safe-add alist key val #!key (metadata #f))
;;   (let ((newalist (filter (lambda (x)(not (equal? key (car x)))) alist)))
;;     (append newalist (list (if metadata
;; 			       (list key val metadata)
;; 			       (list key val))))))
;; 
;; (define (configf:section-var-set! cfgdat section-name var value #!key (metadata #f))
;;   (hash-table-set! cfgdat section-name
;; 		   (configf:assoc-safe-add
;; 		    (hash-table-ref/default cfgdat section-name '())
;; 		    var value metadata: metadata)))
;; 
;; (define (configf:lookup cfgdat section var)
;;   (if (hash-table? cfgdat)
;;       (let ((sectdat (hash-table-ref/default cfgdat section '())))
;; 	(if (null? sectdat)
;; 	    #f
;; 	    (let ((match (assoc var sectdat)))
;; 	      (if match ;; (and match (list? match)(> (length match) 1))
;; 		  (cadr match)
;; 		  #f))
;; 	    ))
;;       #f))
;; 
;; ;; use to have definitive setting:
;; ;;  [foo]
;; ;;  var yes
;; ;;
;; ;;  (configf:var-is? cfgdat "foo" "var" "yes") => #t
;; ;;
;; (define (configf:var-is? cfgdat section var expected-val)
;;   (equal? (configf:lookup cfgdat section var) expected-val))
;; 
;; (define config-lookup configf:lookup)
(define configf:read-file read-config)

;; ;; safely look up a value that is expected to be a number, return
;; ;; a default (#f unless provided)
;; ;;
;; (define (configf:lookup-number cfdat section varname #!key (default #f))
;;   (let* ((val (configf:lookup *configdat* section varname))
;;          (res (if val
;;                   (string->number (string-substitute "\\s+" "" val #t))
;;                   #f)))
;;     (cond
;;      (res  res)
;;      (val  (debug:print 0 *default-log-port* "ERROR: no number found for [" section "], " varname ", got: " val))
;;      (else default))))
;; 
;; (define (configf:section-vars cfgdat section)
;;   (let ((sectdat (hash-table-ref/default cfgdat section '())))
;;     (if (null? sectdat)
;; 	'()
;; 	(map car sectdat))))
;; 
;; (define (configf:get-section cfgdat section)
;;   (hash-table-ref/default cfgdat section '()))
;; 
;; (define (configf:set-section-var cfgdat section var val)
;;   (let ((sectdat (configf:get-section cfgdat section)))
;;     (hash-table-set! cfgdat section
;;                      (configf:assoc-safe-add sectdat var val))))
;; 
;;     ;;(append (filter (lambda (x)(not (assoc var sectdat))) sectdat)
;;     ;;	    (list var val))))
;; 
;;======================================================================
;; setup
;;======================================================================
;;======================================================================

(define (setup)
  (let* ((configf (find-config "megatest.config"))
	 (config  (if configf (read-config configf #f #t) #f)))
    (if config
	(setenv "RUN_AREA_HOME" (pathname-directory configf)))
    config))

;;======================================================================
;; Non destructive writing of config file
;;======================================================================

(define (configf:compress-multi-lines fdat)
  ;; step 1.5 - compress any continued lines
  (if (null? fdat) fdat
	(let loop ((hed (car fdat))
		   (tal (cdr fdat))
		   (cur "")
		   (led #f)
		   (res '()))
	  ;; ALL WHITESPACE LEADING LINES ARE TACKED ON!!
	  ;;  1. remove led whitespace
	  ;;  2. tack on to hed with "\n"
	  (let ((match (string-match configf:cont-ln-rx hed)))
	    (if match ;; blast! have to deal with a multiline
		(let* ((lead (cadr match))
		       (lval (caddr match))
		       (newl (conc cur "\n" lval)))
		  (if (not led)(set! led lead))
		  (if (null? tal) 
		      (set! fdat (append fdat (list newl)))
		      (loop (car tal)(cdr tal) newl led res))) ;; NB// not tacking newl onto res
		(let ((newres (if led 
				  (append res (list cur hed))
				  (append res (list hed)))))
		  ;; prev was a multiline
		  (if (null? tal)
		      newres
		      (loop (car tal)(cdr tal) "" #f newres))))))))

;; note: I'm cheating a little here. I merely replace "\n" with "\n         "
(define (configf:expand-multi-lines fdat)
  ;; step 1.5 - compress any continued lines
  (if (null? fdat) fdat
      (let loop ((hed (car fdat))
		 (tal (cdr fdat))
		 (res '()))
	(let ((newres (append res (list (string-substitute (regexp "\n") "\n         " hed #t)))))
	  (if (null? tal)
	      newres
	      (loop (car tal)(cdr tal) newres))))))

(define (configf:file->list fname)
  (if (common:file-exists? fname)
      (let ((inp (open-input-file fname)))
	(let loop ((inl (read-line inp))
		   (res '()))
	  (if (eof-object? inl)
	      (begin
		(close-input-port inp)
		(reverse res))
	      (loop (read-line inp)(cons inl res)))))
      '()))

;;======================================================================
;; Write a config
;;   0. Given a refererence data structure "indat"
;;   1. Open the output file and read it into a list
;;   2. Flatten any multiline entries
;;   3. Modify values per contents of "indat" and remove absent values
;;   4. Append new values to the section (immediately after last legit entry)
;;   5. Write out the new list 
;;======================================================================

(define (configf:write-config indat fname #!key (required-sections '()))
  (let* (;; step 1: Open the output file and read it into a list
	 (fdat       (configf:file->list fname))
	 (refdat  (make-hash-table))
	 (sechash (make-hash-table)) ;; current section hash, init with hash for "default" section
	 (new     #f) ;; put the line to be used in new, if it is to be deleted the set new to #f
	 (secname #f))

    ;; step 2: Flatten multiline entries
    (if (not (null? fdat))(set! fdat (configf:compress-multi-lines fdat)))

    ;; step 3: Modify values per contents of "indat" and remove absent values
    (if (not (null? fdat))
	(let loop ((hed  (car fdat))
		   (tal  (cadr fdat))
		   (res  '())
		   (lnum 0))
	  (regex-case 
	   hed
	   (configf:comment-rx _                  (set! res (append res (list hed)))) ;; (loop (read-line inp) curr-section-name #f #f))
	   (configf:blank-l-rx _                  (set! res (append res (list hed)))) ;; (loop (read-line inp) curr-section-name #f #f))
	   (configf:section-rx ( x section-name ) (let ((section-hash (hash-table-ref/default refdat section-name #f)))
					    (if (not section-hash)
						(let ((newhash (make-hash-table)))
						  (hash-table-set! refdat section-name newhash) ;; was refhash - not sure that refdat is correct here
						  (set! sechash newhash))
						(set! sechash section-hash))
					    (set! new hed) ;; will append this at the bottom of the loop
					    (set! secname section-name)
					    ))
	   ;; No need to process key cmd, let it fall though to key val
	   (configf:key-val-pr ( x key val      )
		       (let ((newval (configf:lookup indat secname key))) ;; was sec, bug or correct?
			 ;; can handle newval == #f here => that means key is removed
			 (cond 
			  ((equal? newval val)
			   (set! res (append res (list hed))))
			  ((not newval) ;; key has been removed
			   (set! new #f))
			  ((not (equal? newval val))
			     (hash-table-set! sechash key newval)
			     (set! new (conc key " " newval)))
			  (else
			   (debug:print-error 0 *default-log-port* "problem parsing line number " lnum "\"" hed "\"")))))
	   (else
	    (debug:print-error 0 *default-log-port* "Problem parsing line num " lnum " :\n   " hed )))
	  (if (not (null? tal))
	      (loop (car tal)(cdr tal)(if new (append res (list new)) res)(+ lnum 1)))
	  ;; drop to here when done processing, res contains modified list of lines
	  (set! fdat res)))

    ;; step 4: Append new values to the section
    (for-each 
     (lambda (section)
       (let ((sdat   '()) ;; append needed bits here
	     (svars  (configf:section-vars indat section)))
	 (for-each 
	  (lambda (var)
	    (let ((val (configf:lookup refdat section var)))
	      (if (not val) ;; this one is new
		  (begin
		    (if (null? sdat)(set! sdat (list (conc "[" section "]"))))
		    (set! sdat (append sdat (list (conc var " " val))))))))
	  svars)
	 (set! fdat (append fdat sdat))))
     (delete-duplicates (append required-sections (hash-table-keys indat))))

    ;; step 5: Write out new file
    (with-output-to-file fname 
      (lambda ()
	(for-each 
	 (lambda (line)
	   (print line))
	 (configf:expand-multi-lines fdat))))))

;;======================================================================
;; refdb
;;======================================================================

;; reads a refdb into an assoc array of assoc arrays
;;   returns (list dat msg)
(define (configf:read-refdb refdb-path)
  (let ((sheets-file  (conc refdb-path "/sheet-names.cfg")))
    (if (not (common:file-exists? sheets-file))
	(list #f (conc "ERROR: no refdb found at " refdb-path))
	(if (not (file-read-access? sheets-file))
	    (list #f (conc "ERROR: refdb file not readable at " refdb-path))
	    (let* ((sheets (with-input-from-file sheets-file
			     (lambda ()
			       (let loop ((inl (read-line))
					  (res '()))
				 (if (eof-object? inl)
				     (reverse res)
				     (loop (read-line)(cons inl res)))))))
		   (data   '()))
	      (for-each 
	       (lambda (sheet-name)
		 (let* ((dat-path  (conc refdb-path "/" sheet-name ".dat"))
			(ref-dat   (configf:read-file dat-path #f #t))
			(ref-assoc (map (lambda (key)
					  (list key (hash-table-ref ref-dat key)))
					(hash-table-keys ref-dat))))
				   ;; (hash-table->alist ref-dat)))
		   ;; (set! data (append data (list (list sheet-name ref-assoc))))))
		   (set! data (cons (list sheet-name ref-assoc) data))))
	       sheets)
	      (list data "NO ERRORS"))))))

;; map over all pairs in a three level hierarchial alist and apply a function to the keys/val
;;
(define (configf:map-all-hier-alist data proc #!key (initproc1 #f)(initproc2 #f)(initproc3 #f))
  (for-each 
   (lambda (sheetname)
     (let* ((sheettmp  (assoc sheetname data))
	    (sheetdat  (if sheettmp (cadr sheettmp) '())))
       (if initproc1 (initproc1 sheetname))
       (for-each 
	(lambda (sectionname)
	  (let* ((sectiontmp  (assoc sectionname sheetdat))
		 (sectiondat  (if sectiontmp (cadr sectiontmp) '())))
	    (if initproc2 (initproc2 sheetname sectionname))
	    (for-each
	     (lambda (varname)
	       (let* ((valtmp (assoc varname sectiondat))
		      (val    (if valtmp (cadr valtmp) "")))
		 (proc sheetname sectionname varname val)))
	     (map car sectiondat))))
	(map car sheetdat))))
   (map car data))
  data)

;;======================================================================
;;  C O N F I G   T O / F R O M   A L I S T
;;======================================================================

(define (configf:config->alist cfgdat)
  (hash-table->alist cfgdat))

(define (configf:alist->config adat)
  (let ((ht (make-hash-table)))
    (for-each
     (lambda (section)
       (hash-table-set! ht (car section)(cdr section)))
     adat)
    ht))

;; convert hierarchial list to ini format
;;
(define (configf:config->ini data)
  (map 
   (lambda (section)
     (let ((section-name (car section))
	   (section-dat  (cdr section)))
       (print "\n[" section-name "]")
       (map (lambda (dat-pair)
	      (let* ((var (car dat-pair))
		     (val (cadr dat-pair))
		     (fname (if (> (length dat-pair) 2)(caddr dat-pair) #f)))
		(if fname (print "# " var "=>" fname))
		(print var " " val)))
	    section-dat))) ;;       (print "section-dat: " section-dat))
   (hash-table->alist data)))

(define (tests:get-tests-search-path cfgdat)
  (let ((paths (let ((section (if cfgdat
				  (configf:get-section cfgdat "tests-paths")
				  #f)))
		 (if section
		     (map cadr section)
		     '()))))
    (filter (lambda (d)
	      (if (directory-exists? d)
		  d
		  (begin
		    ;; (if (common:low-noise-print 60 "tests:get-tests-search-path" d)
		    ;;	(debug:print 0 *default-log-port* "WARNING: problem with directory " d ", dropping it from tests path"))
		    #f)))
	    (append paths (list (conc *toppath* "/tests"))))))


)

(define (dboard:commondat-add-updater commondat updater #!key (tab-num #f))
  (let* ((tnum          (or tab-num
			     (dboard:commondat-curr-tab-num commondat)))
	 (curr-updaters (hash-table-ref/default (dboard:commondat-updaters commondat) tnum '())))
    (hash-table-set! (dboard:commondat-updaters commondat)
		     tnum
		     (cons updater curr-updaters))))



(define (common:run-a-command cmd #!key (with-vars #f) (with-orig-env #f))
  (let* ((pre-cmd  (dtests:get-pre-command))
         (post-cmd (dtests:get-post-command))
         (fullcmd  (if (or pre-cmd post-cmd)
                       (conc pre-cmd cmd post-cmd)
                       (conc "viewscreen " cmd))))
    (debug:print-info 02 *default-log-port* "Running command: " fullcmd)
    (cond
     (with-vars     (common:without-vars  fullcmd))
     (with-orig-env (common:with-orig-env fullcmd))
     (else          (common:without-vars  fullcmd "MT_.*")))))

(use (srfi 18) extras tcp stack)
(use sqlite3 srfi-1 posix regex regex-case srfi-69 csv-xml s11n md5 message-digest
     base64 format dot-locking z3 typed-records matchable)
(import (prefix sqlite3 sqlite3:))
(import (prefix base64 base64:))

(declare (unit db))
;; (declare (uses common))
;; (declare (uses client))
;; (declare (uses mt))
;; (declare (uses margsmod))
;; (import margsmod)























;; 










































































;; (declare (uses commonmod))







;; (import commonmod)


;; 





















;; (declare (uses configfmod))



















;; (import configfmod)



















;; 




















;; (declare (uses dbmod))















;; (import dbmod)

;; 













;; (declare (uses servermod))



;; (import servermod)

;; 











































;; (include "common_records.scm")

;; (include "db_records.scm")



;; (include "key_records.scm")




















;; (include "run_records.scm")























;; 







;; 

;;======================================================================

(declare (unit dbmod))
(declare (uses commonmod))
(declare (uses ods))
(declare (uses configfmod))
(declare (uses margsmod))

(module dbmod
	*
	
(import commonmod)
(import ods)
(import configfmod)
(import margsmod)

(import scheme chicken data-structures extras ports)
(import
  (prefix base64 base64:)
  (prefix sqlite3 sqlite3:)
  csv
  csv-xml

  regex
  s11n
  srfi-1
  srfi-13
  stack
  z3
  )

(import (prefix dbi dbi:)
	pkts)

;;======================================================================
;;  R E C O R D S
;;======================================================================

(include "db_records.scm")

(define (db:writeable dbdir dbname)
  (let* ((dbfile (conc dbdir "/" dbname)))
    (if (file-exists? dbfile)
	(file-write-access? dbfile)
	(file-write-access? *toppath*))))

;;======================================================================the end

;; state is the priority rollup of all states
;; status is the priority rollup of all completed statesfu
;;
;; if test-name is an integer work off that instead of test-name test-path
;;
(define (db:set-state-status-and-roll-up-items dbstruct run-id test-name item-path state status comment)
  ;; establish info on incoming test followed by info on top level test
  ;; BBnote - for mode itemwait, linkage between upstream test & matching item status is propagated to run queue in db:prereqs-not-met
  (let* ((testdat      (if (number? test-name)
			   (db:get-test-info-by-id dbstruct run-id test-name) ;; test-name is actually a test-id
			   (db:get-test-info       dbstruct run-id test-name item-path)))
	 (test-id      (db:test-get-id testdat))
	 (test-name    (if (number? test-name)
			   (db:test-get-testname testdat)
			   test-name))
	 (item-path    (db:test-get-item-path testdat))
         (tl-testdat   (db:get-test-info dbstruct run-id test-name ""))
         (tl-test-id   (if tl-testdat
			   (db:test-get-id tl-testdat)
			   #f)))
    (if (member state '("LAUNCHED" "REMOTEHOSTSTART")) 
	(db:general-call dbstruct 'set-test-start-time (list test-id)))
    (mutex-lock! *db-transaction-mutex*)
    (db:with-db
     dbstruct #f #f
     (lambda (db)
       (let ((tr-res
              (sqlite3:with-transaction
               db
               (lambda ()
                 ;; NB// Pass the db so it is part fo the transaction
                 (db:test-set-state-status db run-id test-id state status comment) ;; this call sets the item state/status
                 (if (not (equal? item-path "")) ;; only roll up IF incoming test is an item
                     (let* ((state-status-counts  (db:get-all-state-status-counts-for-test dbstruct run-id test-name item-path state status)) ;; item-path is used to exclude current state/status of THIS test
			      						  (state-stauses (db:roll-up-rules state-status-counts state status))
                          (newstate (car state-stauses))
                          (newstatus (cadr state-stauses)))
                       (debug:print 4 *default-log-port* "BB> tl-test-id="tl-test-id" ; "test-name":"item-path" newstate="newstate" newstatus="newstatus" len(sscs)="(length state-status-counts)  " state-status-counts: "
							(apply conc
                  (map (lambda (x)
                     (conc
                     		(with-output-to-string (lambda () (pp (dbr:counts->alist x)))) " | "))
                              state-status-counts))); end debug:print
   
                       (if tl-test-id
			   (db:test-set-state-status db run-id tl-test-id newstate newstatus #f)) ;; we are still in the transaction - must access the db and not the dbstruct
		       ))))))
         (mutex-unlock! *db-transaction-mutex*)
         (if (and test-id state status (equal? status "AUTO")) 
             (db:test-data-rollup dbstruct run-id test-id status))
         tr-res)))))

;; ;; speed up for common cases with a little logic
;; ;; NB// Ultimately this will be deprecated in deference to mt:test-set-state-status-by-id
;;
;;      NOTE: run-id is not used
;; ;;
(define (db:test-set-state-status dbstruct run-id test-id newstate newstatus newcomment)
  (db:with-db
   dbstruct
   ;; run-id
   #f
   #t
   (lambda (db)
     (cond
      ((and newstate newstatus newcomment)
       (sqlite3:execute db "UPDATE tests SET state=?,status=?,comment=? WHERE id=?;" newstate newstatus newcomment ;; (sdb:qry 'getid newcomment)
			test-id))
      ((and newstate newstatus)
       (sqlite3:execute db "UPDATE tests SET state=?,status=? WHERE id=?;" newstate newstatus test-id))
      (else
       (if newstate   (sqlite3:execute db "UPDATE tests SET state=?   WHERE id=?;" newstate   test-id))
       (if newstatus  (sqlite3:execute db "UPDATE tests SET status=?  WHERE id=?;" newstatus  test-id))
       (if newcomment (sqlite3:execute db "UPDATE tests SET comment=? WHERE id=?;" newcomment ;; (sdb:qry 'getid newcomment)
				       test-id))))))
  (mt:process-triggers dbstruct run-id test-id newstate newstatus))

(define (db:find-and-mark-incomplete dbstruct run-id ovr-deadtime)
  (let* ((incompleted '())
	 (oldlaunched '())
	 (toplevels   '())
          ;; The default running-deadtime is 720 seconds = 12 minutes.
          ;; "(running-deadtime-default (+ server-start-allowance (* 2 launch-monitor-period)))" = 200 + (2 * (200 + 30 + 30))
         (deadtime-trim (or ovr-deadtime (configf:lookup-number *configdat* "setup" "deadtime")))
         (server-start-allowance 200)
         (server-overloaded-budget 200)
         (launch-monitor-off-time (or (configf:lookup-number *configdat* "setup" "test-stats-update-period") 30))
         (launch-monitor-on-time-budget 30)
         (launch-monitor-period (+ launch-monitor-off-time launch-monitor-on-time-budget server-overloaded-budget))
         (remotehoststart-deadtime-default (+ server-start-allowance server-overloaded-budget 30))
         (remotehoststart-deadtime (or deadtime-trim remotehoststart-deadtime-default))
         (running-deadtime-default (+ server-start-allowance (* 2 launch-monitor-period)))
         (running-deadtime (or deadtime-trim running-deadtime-default)) ;; two minutes (30 seconds between updates, this leaves 3x grace period)
         )
    (debug:print-info 4  *default-log-port* "running-deadtime = " running-deadtime)
    (debug:print-info 4  *default-log-port* "deadtime-trim = " deadtime-trim)

    (db:with-db 
     dbstruct #f #f
     (lambda (db)
       (let* ((stmth1 (db:get-cache-stmth
		       dbstruct db
		       "SELECT id,rundir,uname,testname,item_path,event_time,run_duration FROM tests 
                           WHERE run_id=? AND (strftime('%s','now') - event_time) > (run_duration + ?)
                                          AND state IN ('RUNNING');"))
	      (stmth2 (db:get-cache-stmth
		       dbstruct db
		       "SELECT id,rundir,uname,testname,item_path,event_time,run_duration FROM tests 
                           WHERE run_id=? AND (strftime('%s','now') - event_time) > (run_duration + ?)
                                          AND state IN ('REMOTEHOSTSTART');"))
	      (stmth3 (db:get-cache-stmth
		       dbstruct db
		       "SELECT id,rundir,uname,testname,item_path FROM tests
                           WHERE run_id=? AND (strftime('%s','now') - event_time) > 86400
                                          AND state IN ('LAUNCHED');")))
	 ;; in RUNNING or REMOTEHOSTSTART for more than 10 minutes
	 ;;
	 ;; HOWEVER: this code in run:test seems to work fine
	 ;;              (> (- (current-seconds)(+ (db:test-get-event_time testdat)
	 ;;                     (db:test-get-run_duration testdat)))
	 ;;                    600) 
	 ;; (db:delay-if-busy dbdat)
	 (sqlite3:for-each-row 
	  (lambda (test-id run-dir uname testname item-path event-time run-duration)
	    (if (and (equal? uname "n/a")
		     (equal? item-path "")) ;; this is a toplevel test
		;; what to do with toplevel? call rollup?
		(begin
		  (set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
		  (debug:print-info 0 *default-log-port* "Found old toplevel test in RUNNING state, test-id=" test-id))
		(begin
		  (set! incompleted (cons (list test-id run-dir uname testname item-path run-id) incompleted))
		  (debug:print-info 0 *default-log-port* "Found old test in RUNNING state, test-id="
				    test-id" exceeded running-deadtime "running-deadtime" now="(current-seconds)
				    " event-time="event-time" run-duration="run-duration))))
	  stmth1
	  run-id running-deadtime) ;; default time 720 seconds
       
	 (sqlite3:for-each-row 
	  (lambda (test-id run-dir uname testname item-path event-time run-duration)
	    (if (and (equal? uname "n/a")
		     (equal? item-path "")) ;; this is a toplevel test
		;; what to do with toplevel? call rollup?
		(begin
		  (set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
		  (debug:print-info 0 *default-log-port* "Found old toplevel test in RUNNING state, test-id=" test-id))
		(begin
		  (debug:print-info 0 *default-log-port* "Found old test in REMOTEHOSTSTART state, test-id=" test-id
				    " exceeded running-deadtime "running-deadtime" now="(current-seconds)" event-time="event-time
				    " run-duration="run-duration)
		  (set! incompleted (cons (list test-id run-dir uname testname item-path run-id) incompleted)))))
	  stmth2
	  run-id remotehoststart-deadtime) ;; default time 230 seconds
	 
	 ;; in LAUNCHED for more than one day. Could be long due to job queues TODO/BUG: Need override for this in config
	 ;;
	 ;; (db:delay-if-busy dbdat)
	 (sqlite3:for-each-row
	  (lambda (test-id run-dir uname testname item-path)
	    (if (and (equal? uname "n/a")
		     (equal? item-path "")) ;; this is a toplevel test
		;; what to do with toplevel? call rollup?
		(set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
		(begin
		  (debug:print-info 0 *default-log-port* "Found old test in LAUNCHED state, test-id=" test-id
				    " 1 day since event_time marked")
                (set! oldlaunched (cons (list test-id run-dir uname testname item-path run-id) oldlaunched)))))
	  stmth3
	  run-id)
       
	 (debug:print-info 18 *default-log-port* "Found " (length oldlaunched) " old LAUNCHED items, "
			   (length toplevels) " old LAUNCHED toplevel tests and "
			   (length incompleted) " tests marked RUNNING but apparently dead."))

	 ;; These are defunct tests, do not do all the overhead of set-state-status. Force them to INCOMPLETE.
       ;;
	 ;; (db:delay-if-busy dbdat)
	 (let* ((min-incompleted-ids (map car incompleted)) ;; do 'em all
		(all-ids             (append min-incompleted-ids (map car oldlaunched))))
	   (if (> (length all-ids) 0)
	       (begin
		 ;; (launch:is-test-alive "localhost" 435)
		 (debug:print 0 *default-log-port* "WARNING: Marking test(s); " (string-intersperse (map conc all-ids) ", ")
			      " as DEAD")
		 (for-each
                  (lambda (test-id)
                    (let* (;; (run-dir (db:test-get-rundir-from-test-id dbstruct run-id test-id))
			   (tinfo   (db:get-test-info-by-id dbstruct run-id test-id))
			   (run-dir (db:test-get-rundir     tinfo))
			   (host    (db:test-get-host       tinfo))
			   (pid     (db:test-get-process_id tinfo))
			   (result (db:get-status-from-final-status-file run-dir)))
		      (if (and (list? result) (> (length result) 1) (equal? "PASS" (cadr result)) (equal? "COMPLETED" (car result))) 
			  (begin
			    (debug:print 0 *default-log-port* "INFO: test " test-id " actually passed, so marking PASS not DEAD")
			    (db:set-state-status-and-roll-up-items
			     dbstruct run-id test-id 'foo "COMPLETED" "PASS"
			     "Test stopped responding but it has PASSED; marking it PASS in the DB."))
			  (let ((is-alive (and (not (eq? pid 0))  ;; 0 is default in re-used field "attemptnum" where pid stored.
					       (launch:is-test-alive host pid))))
			    (if is-alive
				(debug:print 0 *default-log-port* "INFO: test " test-id " on host " host
					     " has a process on pid " pid ", NOT setting to DEAD.")
				(begin
				  (debug:print 0 *default-log-port* "INFO: test " test-id
					       " final state/status is not COMPLETED/PASS. It is " result)
				  (db:set-state-status-and-roll-up-items
				   dbstruct run-id test-id 'foo "COMPLETED" "DEAD"
				   "Test stopped responding while in RUNNING or REMOTEHOSTSTART; presumed dead.")))))))
		  ;; call end of eud of run detection for posthook - from merge, is it needed?
		  ;; (launch:end-of-run-check run-id)
		  all-ids)

		 ;; MOVE TO rmt:find-and-mark-incomplete - for now always call launch:end-of-run-check after
		 ;; calling rmt:find-and-mark-incompletes
		 
		 ;;ALWAYS CALL after rmt:find-and-mark-incompletes
		 ;; call end of eud of run detection for posthook
		 ;; (launch:end-of-run-check run-id)

		 )))))))

(define (db:test-set-state-status-process-triggers  dbstruct run-id test-id newstate newstatus newcomment)
  (db:test-set-state-status dbstruct run-id test-id newstate newstatus newcomment)
  (mt:process-triggers dbstruct run-id test-id newstate newstatus))

;; get a list of servers with all relevant data
;; ( mod-time host port start-time pid )
;;
(define (server:get-list areapath #!key (limit #f))
  (let ((fname-rx    (regexp "^(|.*/)server-(\\d+)-(\\S+).log$"))
	(day-seconds (* 24 60 60)))
    ;; if the directory exists continue to get the list
    ;; otherwise attempt to create the logs dir and then
    ;; continue
    (if (if (directory-exists? (conc areapath "/logs"))
	    '()
	    (if (file-write-access? areapath)
		(begin
		  (condition-case
		   (create-directory (conc areapath "/logs") #t)
		   (exn (i/o file)(debug:print 0 *default-log-port* "ERROR: Cannot create directory at " (conc areapath "/logs")))
		   (exn ()(debug:print 0 *default-log-port* "ERROR: Unknown error attemtping to get server list. exn=" exn)))
		  (directory-exists? (conc areapath "/logs")))
		'()))
	(let* ((server-logs   (glob (conc areapath "/logs/server-[0-9]*.log")))
	       (num-serv-logs (length server-logs)))
	  (if (null? server-logs)
	      '()
	      (let loop ((hed  (car server-logs))
			 (tal  (cdr server-logs))
			 (res '()))
		(let* ((mod-time  (handle-exceptions
				   exn
				   (begin
				     (debug:print 0 *default-log-port* "failed to get modification time on " hed ", exn=" exn)
				     (current-seconds)) ;; 0
				   (file-modification-time hed))) ;; default to *very* old so log gets ignored if deleted
		       (down-time (- (current-seconds) mod-time))
		       (serv-dat  (if (or (< num-serv-logs 10)
				  	  (< down-time 900)) ;; day-seconds))
				      (server:logf-get-start-info hed)
				      '())) ;; don't waste time processing server files not touched in the 15 minutes if there are more than ten servers to look at
		       (serv-rec (cons mod-time serv-dat))
		       (fmatch   (string-match fname-rx hed))
		       (pid      (if fmatch (string->number (list-ref fmatch 2)) #f))
		       (new-res  (if (null? serv-dat)
				     res
				     (cons (append serv-rec (list pid)) res)))) ;; any changes to number of elements in new-res will dirctly affect server:record->url,server:record->id,server:kill,server:get-num-alive which uses match let 
		  (if (null? tal)
		      (if (and limit
			       (> (length new-res) limit))
			  new-res ;; (take new-res limit)  <= need intelligent sorting before this will work
			  new-res)
		      (loop (car tal)(cdr tal) new-res)))))))))

;; options:
;;
;;  'killservers  - kills all servers
;;  'dejunk       - removes junk records
;;  'adj-testids  - move test-ids into correct ranges
;;  'old2new      - sync megatest.db to /tmp/.../megatest.db and /tmp/.../megatest_ref.db
;;  'new2old      - sync /tmp/.../megatest.db to megatest.db and /tmp/.../megatest_ref.db (and update data_synced)
;;  'closeall     - close all opened dbs
;;  'schema       - attempt to apply schema changes
;;  run-ids: '(1 2 3 ...) or #f (for all)
;;
(define (db:multi-db-sync dbstruct . options)
  ;; (if (not (launch:setup))
  ;;    (debug:print 0 *default-log-port* "ERROR: not able to setup up for megatest.")
  (let* ((mtdb     (dbr:dbstruct-mtdb dbstruct))
	 (tmpdb    (db:get-db dbstruct))
	 (refndb   (dbr:dbstruct-refndb dbstruct))
	 (allow-cleanup #t) ;; (if run-ids #f #t))
	 (servers (server:get-list *toppath*)) ;; (tasks:get-all-servers (db:delay-if-busy tdbdat)))
	 (data-synced 0)) ;; count of changed records (I hope)
    
    (for-each
     (lambda (option)
       
       (case option
	 ;; kill servers
	 ((killservers)
	  (for-each
	   (lambda (server)
             (handle-exceptions
             exn
             (begin 
               (debug:print-info 0 *default-log-port*  "Unable to get host and/or port from " server ", exn=" exn)     
               #f)
	     (match-let (((mod-time host port start-time server-id pid) server))
	       (if (and host pid)
		   (tasks:kill-server host pid)))))
	   servers)

          ;; /tmp/bjbarcla/megatest_localdb/fullrun/.nfs.pdx.disks.icf_env_disk001.bjbarcla.gwa.issues.mtdev.matt-bisect.megatest.ext-tests.runs.all.v1.65.1.6524.dev.bb-24-justrollup-f8.rollup.fullrun/megatest.db.lock
          (delete-file* (common:get-sync-lock-filepath))
          )
	 
	 ;; clear out junk records
	 ;;
	 ((dejunk)
	  ;; (db:delay-if-busy mtdb) ;; ok to delay on mtdb
	  (when (file-write-access? (db:dbdat-get-path mtdb)) (db:clean-up mtdb))
	  (db:clean-up tmpdb)
	  (db:clean-up refndb))

	 ;; sync runs, test_meta etc.
	 ;;
	 ((old2new)
	  (set! data-synced
	    (+ (db:sync-tables (db:sync-all-tables-list dbstruct) #f mtdb tmpdb refndb)
	       data-synced)))
	 
	 ;; now ensure all newdb data are synced to megatest.db
	 ;; do not use the run-ids list passed in to the function
	 ;;
	 ((new2old)
	  (set! data-synced
	    (+ (db:sync-tables (db:sync-all-tables-list dbstruct) #f tmpdb refndb mtdb)
	       data-synced)))

	 ((adj-target)
	  (db:adj-target (db:dbdat-get-db mtdb))
	  (db:adj-target (db:dbdat-get-db tmpdb))
	  (db:adj-target (db:dbdat-get-db refndb)))
	 
	 ((schema)
	  (db:patch-schema-maindb (db:dbdat-get-db mtdb))
	  (db:patch-schema-maindb (db:dbdat-get-db tmpdb))
	  (db:patch-schema-maindb (db:dbdat-get-db refndb))
	  (db:patch-schema-rundb  (db:dbdat-get-db mtdb))
	  (db:patch-schema-rundb  (db:dbdat-get-db tmpdb))
	  (db:patch-schema-rundb  (db:dbdat-get-db refndb))))
       
       (stack-push! (dbr:dbstruct-dbstack dbstruct) tmpdb))
     options)
    data-synced))

(define (db:get-access-mode)
  (if (args:get-arg "-use-db-cache") 'cached 'rmt))

;; given a path to a server log return: host port startseconds
;; any changes to number of elements returned by this fuction will dirctly affect server:record->url,server:record->id,server:kill,server:get-num-alive which uses match let 
;;
(define (server:logf-get-start-info logf)
  (let ((rx (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+) server-id: (\\S+)"))) ;; SERVER STARTED: host:port AT timesecs server id
    ;;(handle-exceptions
    ;; exn
    ;; (begin
    ;;   (debug:print-info 0 *default-log-port* "Unable to get server info from " logf ", exn=" exn)
    ;;   (list #f #f #f #f)) ;; no idea what went wrong, call it a bad server
    (if (and (file-exists? logf)
	     (file-read-access? logf))
	(with-input-from-file
	    logf
	  (lambda ()
	    (let loop ((inl  (read-line))
		       (lnum 0))
	      (if (not (eof-object? inl))
		  (let ((mlst (string-match rx inl)))
		    (if (not mlst)
			(if (< lnum 500) ;; give up if more than 500 lines of server log read
			    (loop (read-line)(+ lnum 1))
			    (begin 
			      (debug:print-info 0 *default-log-port* "Unable to get server info from first 500 lines of " logf )
			      (list #f #f #f #f)))
			(let ((dat  (cdr mlst)))
			  (list (car dat) ;; host
				(string->number (cadr dat)) ;; port
				(string->number (caddr dat))
				(cadr (cddr dat))))))
		  (begin 
		    (debug:print-info 0 *default-log-port* "Unable to get server info from " logf " at " (current-seconds))
		    (list #f #f #f #f))))))
	(begin
	  (debug:print-info 0 *default-log-port* "Unable to get server info from " logf ", file not found or not readable.")
	  (list #f #f #f #f)))))

;; cannot move to dbmod until lazy-read-testconfig is unravelled.
;;
(define (mt:process-triggers dbstruct run-id test-id newstate newstatus)
  (if test-id 
      (let* ((test-dat      (db:get-test-info-by-id dbstruct run-id test-id)))
	(if test-dat
	    (let* ((test-rundir   (db:test-get-rundir       test-dat)) ;; ) ;; )
		   (test-name     (db:test-get-testname     test-dat))
		   (item-path     (db:test-get-item-path    test-dat))
		   (duration      (db:test-get-run_duration test-dat))
		   (comment       (db:test-get-comment      test-dat))
		   (event-time    (db:test-get-event_time   test-dat))
		   (tconfig       #f)
		   (state         (if newstate  newstate  (db:test-get-state  test-dat)))
		   (status        (if newstatus newstatus (db:test-get-status test-dat))))
	      ;; (mutex-lock! *triggers-mutex*)
              (handle-exceptions
               exn
               (begin
                 (debug:print-error 0 *default-log-port* " Exception in mt:process-triggers for run-id="run-id" test-id="test-id" newstate="newstate" newstatus="newstatus
                                    "\n   error: " ((condition-property-accessor 'exn 'message) exn) ", exn=" exn
                                    "\n   test-rundir="test-rundir
                                    "\n   test-name="test-name
                                    "\n   item-path="item-path
                                    "\n   state="state
                                    "\n   status="status
                                    "\n")
                 (print-call-chain (current-error-port))
                 #f)
               (if (and test-name
                        test-rundir)   ;; #f means no dir set yet
                   ;; (common:file-exists? test-rundir)
                   ;; (directory? test-rundir))
                   (call-with-environment-variables
                    (list (cons "MT_TEST_NAME"    (or test-name "no such test"))
                          (cons "MT_TEST_RUN_DIR" (or test-rundir "no test directory yet"))
                          (cons "MT_ITEMPATH"     (or item-path "")))
                    (lambda ()
                      (if (directory-exists? test-rundir)
                          (push-directory test-rundir)
                          (push-directory *toppath*))
                      (set! tconfig (mt:lazy-read-test-config test-name))
                      (for-each (lambda (trigger)
                                  (let* ((munged-trigger (string-translate trigger "/ " "--"))
					(logname        (conc "last-trigger-" munged-trigger ".log")))
                                    ;; first any triggers from the testconfig
                                    (let ((cmd  (configf:lookup tconfig "triggers" trigger)))
                                      (if cmd (mt:run-trigger cmd test-id test-rundir trigger (conc "tconfig-" logname) test-name item-path event-time state status)))
                                    ;; next any triggers from megatest.config
                                    (let ((cmd  (configf:lookup *configdat* "triggers" trigger)))
                                      (if cmd (mt:run-trigger cmd test-id test-rundir trigger (conc "mtconfig-" logname) test-name item-path event-time state status)))))
                                (list
                                 (conc state "/" status)
                                 (conc state "/")
                                 (conc "/" status)))
		     (pop-directory))
                    )))
	      ;; (mutex-unlock! *triggers-mutex*)
	      )))))

(define (mt:lazy-read-test-config test-name)
  (let ((tconf (hash-table-ref/default *testconfigs* test-name #f)))
    (if tconf
	tconf
	(let ((test-dirs (tests:get-tests-search-path *configdat*)))
	  (let loop ((hed (car test-dirs))
		     (tal (cdr test-dirs)))
	    ;; Setting MT_LINKTREE here is almost certainly unnecessary. 
	    (let ((tconfig-file (conc hed "/" test-name "/testconfig")))
	      (if (and (common:file-exists? tconfig-file)
		       (file-read-access? tconfig-file))
		  (let ((link-tree-path (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree"))
			(old-link-tree  (get-environment-variable "MT_LINKTREE")))
		    (if link-tree-path (setenv "MT_LINKTREE" link-tree-path))
		    (let ((newtcfg (read-config tconfig-file #f #f))) ;; NOTE: Does NOT run [system ...]
		      (hash-table-set! *testconfigs* test-name newtcfg)
		      (if old-link-tree 
			  (setenv "MT_LINKTREE" old-link-tree)
			  (unsetenv "MT_LINKTREE"))
		      newtcfg))
		  (if (null? tal)
		      (begin
			(debug:print-error 0 *default-log-port* "No readable testconfig found for " test-name)
			#f)
		      (loop (car tal)(cdr tal))))))))))


;; MUST RESOLVE mt:process-triggers before these can move to dbmod.

;; set tests with state currstate and status currstatus to newstate and newstatus
;; use currstate = #f and or currstatus = #f to apply to any state or status respectively
;; WARNING: SQL injection risk. NB// See new but not yet used "faster" version below
;;
;;  AND NOT (item_path='' AND testname in (SELECT DISTINCT testname FROM tests WHERE testname=? AND item_path != ''));")))
;;  (debug:print 0 *default-log-port* "QRY: " qry)
;;  (db:delay-if-busy)
;;
;; NB// This call only operates on toplevel tests. Consider replacing it with more general call
;;
(define (db:set-tests-state-status dbstruct run-id testnames currstate currstatus newstate newstatus)
  (let ((test-ids '()))
    (for-each
     (lambda (testname)
       (let ((qry (conc "UPDATE tests SET state=?,status=? WHERE "
			(if currstate  (conc "state='" currstate "' AND ") "")
			(if currstatus (conc "status='" currstatus "' AND ") "")
			" run_id=? AND testname LIKE ?;"))
	     (test-id (db:get-test-id dbstruct run-id testname "")))
	 (db:with-db
	  dbstruct
	  run-id
	  #t
	  (lambda (db)
	    (sqlite3:execute db qry
			     (or newstate  currstate "NOT_STARTED")
			     (or newstatus currstate "UNKNOWN")
			     run-id testname)))
	 (if test-id
	     (begin
	       (set! test-ids (cons test-id test-ids))
	       (mt:process-triggers dbstruct run-id test-id newstate newstatus)))))
     testnames)
    test-ids))


;;======================================================================
;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(define (common:watchdog)
  (debug:print-info 13 *default-log-port* "common:watchdog entered.")
  (if (launch:setup)
      (if (common:on-homehost?)
	  (let ((dbstruct (db:setup #t)))
	    (debug:print-info 13 *default-log-port* "after db:setup with dbstruct=" dbstruct)
	    (cond
	     ((dbr:dbstruct-read-only dbstruct)
	      (debug:print-info 13 *default-log-port* "loading read-only watchdog")
	      (common:readonly-watchdog dbstruct))
	     (else
	      (debug:print-info 13 *default-log-port* "loading writable-watchdog.")
              (let* ((syncer (or (configf:lookup *configdat* "server" "sync-method") "brute-force-sync")))
                (cond
                 ((equal? syncer "brute-force-sync")
                  (server:writable-watchdog-bruteforce dbstruct))
                 ((equal? syncer "delta-sync")
                  (server:writable-watchdog-deltasync dbstruct))
                 (else
                  (debug:print-error 0 *default-log-port* "Unknown server/sync-method specified ("syncer") - valid values are brute-force-sync and delta-sync.")
                  (exit 1)))
                ;;(debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] Syncer started (method="syncer")")
                )))
	    (debug:print-info 13 *default-log-port* "watchdog done."))
	  (debug:print-info 13 *default-log-port* "no need for watchdog on non-homehost"))))

;;======================================================================
;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma
;;
(define (common:cleanup-db dbstruct #!key (full #f))
  (apply db:multi-db-sync 
   dbstruct
   'schema
   ;; 'new2old
   'killservers
   'adj-target
   ;; 'old2new
   'new2old
   ;; (if full
       '(dejunk)
       ;; '())
       )
  (if (common:api-changed?)
      (common:set-last-run-version)))

;;======================================================================
;; Force a megatest cleanup-db if version is changed and skip-version-check not specified
;; Do NOT check if not on homehost!
;;
(define (common:exit-on-version-changed)
  (if (common:on-homehost?)
      (if (common:api-changed?)
	  (let* ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config"))
                (dbfile  (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.db"))
                (read-only (not (file-write-access? dbfile)))
                (dbstruct (db:setup #t)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (common:get-last-run-version))
            (cond
             ((get-environment-variable "MT_SKIP_DB_MIGRATE") #t)
             ((and (common:file-exists? mtconf) (common:file-exists? dbfile) (not read-only)
                   (eq? (current-user-id)(file-owner mtconf))) ;; safe to run -cleanup-db
              (debug:print 0 *default-log-port* "   I see you are the owner of megatest.config, attempting to cleanup and reset to new version")
              (handle-exceptions
               exn
               (begin
                 (debug:print 0 *default-log-port* "Failed to switch versions. exn=" exn)
                 (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
                 (print-call-chain (current-error-port))
                 (exit 1))
               (common:cleanup-db dbstruct)))
             ((not (common:file-exists? mtconf))
              (debug:print 0 *default-log-port* "   megatest.config does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (common:file-exists? dbfile))
              (debug:print 0 *default-log-port* "   megatest.db does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (eq? (current-user-id)(file-owner mtconf)))
              (debug:print 0 *default-log-port* "   You do not own megatest.db in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (read-only
              (debug:print 0 *default-log-port* "   You have read-only access to this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (else
              (debug:print 0 *default-log-port* " to switch versions you can run: \"megatest -cleanup-db\"")
              (exit 1)))))))
;;======================================================================
;;      (begin
;;	(debug:print 0 *default-log-port* "ERROR: cannot migrate version unless on homehost. Exiting.")
;;	(exit 1))))

;;======================================================================
;; currently the primary job of the watchdog is to run the sync back to megatest.db from the db in /tmp
;; if we are on the homehost and we are a server (by definition we are on the homehost if we are a server)
;;
(define (common:readonly-watchdog dbstruct)
  (thread-sleep! 0.05) ;; delay for startup
  (debug:print-info 13 *default-log-port* "common:readonly-watchdog entered.")
  ;; sync megatest.db to /tmp/.../megatst.db
  (let* ((sync-cool-off-duration   3)
        (golden-mtdb     (dbr:dbstruct-mtdb dbstruct))
        (golden-mtpath   (db:dbdat-get-path golden-mtdb))
        (tmp-mtdb        (dbr:dbstruct-tmpdb dbstruct))
        (tmp-mtpath      (db:dbdat-get-path tmp-mtdb)))
    (debug:print-info 0 *default-log-port* "Read-only periodic sync thread started.")
    (let loop ((last-sync-time 0))
      (debug:print-info 13 *default-log-port* "loop top tmp-mtpath="tmp-mtpath" golden-mtpath="golden-mtpath)
      (let* ((duration-since-last-sync (- (current-seconds) last-sync-time)))
        (debug:print-info 13 *default-log-port* "duration-since-last-sync="duration-since-last-sync)
        (if (and (not *time-to-exit*)
                 (< duration-since-last-sync sync-cool-off-duration))
            (thread-sleep! (- sync-cool-off-duration duration-since-last-sync)))
        (if (not *time-to-exit*)
            (let ((golden-mtdb-mtime (file-modification-time golden-mtpath))
                  (tmp-mtdb-mtime    (file-modification-time tmp-mtpath)))
	      (if (> golden-mtdb-mtime tmp-mtdb-mtime)
		  (if (< golden-mtdb-mtime (- (current-seconds) 3)) ;; file has NOT been touched in past three seconds, this way multiple servers won't fight to sync back
		      (let ((res (db:multi-db-sync dbstruct 'old2new)))
			(debug:print-info 13 *default-log-port* "rosync called, " res " records transferred."))))
              (loop (current-seconds)))
            #t)))
    (debug:print-info 0 *default-log-port* "Exiting readonly-watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)" mtpath="golden-mtpath)))

;;======================================================================
;; first read ~/views.config if it exists, then read $MTRAH/views.config if it exists
;;
(define (common:load-views-config)
  (let* ((view-cfgdat    (make-hash-table))
	 (home-cfgfile   (conc (get-environment-variable "HOME") "/.mtviews.config"))
	 (mthome-cfgfile (conc *toppath* "/.mtviews.config")))
    (if (common:file-exists? mthome-cfgfile)
	(read-config mthome-cfgfile view-cfgdat #t))
    ;; we load the home dir file AFTER the MTRAH file so the user can clobber settings when running the dashboard in read-only areas
    (if (common:file-exists? home-cfgfile)
	(read-config home-cfgfile view-cfgdat #t))
    view-cfgdat))

;;======================================================================
;; T A R G E T S  ,   S T A T E ,   S T A T U S ,   
;;                    R U N N A M E    A N D   T E S T P A T T
;;======================================================================

;;======================================================================
;; (map print (map car (hash-table->alist (read-config "runconfigs.config" #f #t))))
;;
(define (common:get-runconfig-targets #!key (configf #f))
  (let ((targs       (sort (map car (hash-table->alist
				     (or configf ;; NOTE: There is no value in using runconfig:read here.
					 (read-config (conc *toppath* "/runconfigs.config")
						      #f #t)
					 (make-hash-table))))
			   string<?))
	(target-patt (args:get-arg "-target")))
    (if target-patt
	(filter (lambda (x)
		  (patt-list-match x target-patt))
		targs)
	targs)))

;;======================================================================
;; Lookup a value in runconfigs based on -reqtarg or -target
;; 
(define (runconfigs-get config var)
  (let ((targ (common:args-get-target))) ;; (or (args:get-arg "-reqtarg")(args:get-arg "-target")(getenv "MT_TARGET"))))
    (if targ
	(or (configf:lookup config targ var)
	    (configf:lookup config "default" var))
	(configf:lookup config "default" var))))

(define (common:args-get-state)
  (or (args:get-arg "-state")(args:get-arg ":state")))

(define (common:args-get-status)
  (or (args:get-arg "-status")(args:get-arg ":status")))

(define (common:args-get-testpatt rconf)
  (let* (;; (tagexpr       (args:get-arg "-tagexpr"))
         ;; (tags-testpatt (if tagexpr (string-join (runs:get-tests-matching-tags tagexpr) ",") #f))
         (testpatt-key  (or (args:get-arg "-modepatt") (args:get-arg "--modepatt") "TESTPATT"))
         (args-testpatt (or (args:get-arg "-testpatt") (args:get-arg "-runtests") "%"))
         (rtestpatt     (if rconf (runconfigs-get rconf testpatt-key) #f)))
    (cond
     ((or (args:get-arg "--modepatt") (args:get-arg "-modepatt")) ;; modepatt is a forced setting, when set it MUST refer to an existing PATT in the runconfig
      (if rconf
	  (let* ((patts-from-mode-patt	  (runconfigs-get rconf testpatt-key)))
	    (debug:print-info 0 *default-log-port* "modepatt defined is: "testpatt-key" runconfigs values for  " testpatt-key " "  patts-from-mode-patt)
	    patts-from-mode-patt)
	  (begin
	    (debug:print-info 0 *default-log-port* " modepatt defined is: "testpatt-key" runconfigs values for  " testpatt-key) ;;  " " patts-from-mode-patt)
	    #f)))     ;; We do NOT fall back to "%"
     ;; (tags-testpatt
     ;;  (debug:print-info 0 *default-log-port* "-tagexpr "tagexpr" selects testpatt "tags-testpatt)
     ;;  tags-testpatt)
     ((and (equal? args-testpatt "%") rtestpatt)
      (debug:print-info 0 *default-log-port* "testpatt defined in "testpatt-key" from runconfigs: " rtestpatt)
      rtestpatt)
     (else 
      (debug:print-info 0 *default-log-port* "using testpatt " args-testpatt " rtestpatt:" rtestpatt)
      args-testpatt))))

;;======================================================================
;; '(print (string-intersperse (map cadr (hash-table-ref/default (read-config "megatest.config" \#f \#t) "disks" '"'"'("none" ""))) "\n"))'
(define (common:get-disks #!key (configf #f))
  (hash-table-ref/default 
   (or configf (read-config "megatest.config" #f #t))
   "disks" '("none" "")))

;;======================================================================
;; postive number if megatest version > db version
;; negative number if megatest version < db version
(define (common:version-db-delta)
         (- megatest-version (common:get-last-run-version-number)))

(define (common:version-changed?)
  (not (equal? (common:get-last-run-version)
               (common:version-signature))))

(define (common:api-changed?)
  (not (equal? (substring (->string megatest-version) 0 4)
               (substring (conc (common:get-last-run-version)) 0 4))))

(define (server:writable-watchdog-bruteforce dbstruct)
  (thread-sleep! 1) ;; delay for startup
  (let* ((do-a-sync  (server:get-bruteforce-syncer dbstruct))
         (final-sync (server:get-bruteforce-syncer dbstruct fork-to-background: #t persist-until-sync: #t)))
    (when (and (not (args:get-arg "-sync-to-megatest.db")) ;; conditions under which we do not run the sync
	       (args:get-arg "-server"))
      
      (let loop ()
	(do-a-sync)
        (if (not *time-to-exit*) (loop))) ;; keep going unless time to exit

      ;; time to exit, close the no-sync db here
      (final-sync)

      (if (common:low-noise-print 30)
	  (debug:print-info 0 *default-log-port* "Exiting watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)
			    )))))

(define (server:writable-watchdog-deltasync dbstruct)
  (thread-sleep! 0.05) ;; delay for startup
  (let ((legacy-sync  (common:run-sync?))
        (sync-stale-seconds (configf:lookup-number *configdat* "server" "sync-stale-seconds" default: 300))
	(debug-mode   (debug:debug-mode 1))
	(last-time    (current-seconds))
	(no-sync-db   (db:open-no-sync-db))
	(stmt-cache   (dbr:dbstruct-stmt-cache dbstruct))
        (sync-duration 0) ;; run time of the sync in milliseconds
        )
    (set! *no-sync-db* no-sync-db) ;; make the no sync db available to api calls
    (debug:print-info 2 *default-log-port* "Periodic sync thread started.")
    (debug:print-info 3 *default-log-port* "watchdog starting. legacy-sync is " legacy-sync" pid="(current-process-id)  );;  " this-wd-num="this-wd-num)
    (if (and legacy-sync (not *time-to-exit*))
	(let* (;;(dbstruct (db:setup))
	       (mtdb       (dbr:dbstruct-mtdb dbstruct))
	       (mtpath     (db:dbdat-get-path mtdb))
	       (tmp-area   (common:get-db-tmp-area))
	       (start-file (conc tmp-area "/.start-sync"))
	       (end-file   (conc tmp-area "/.end-sync")))
	  (debug:print-info 0 *default-log-port* "Server running, periodic sync started.")
	  (let loop ()
	    ;; sync for filesystem local db writes
	    ;;
	    (mutex-lock! *db-multi-sync-mutex*)
	    (let* ((need-sync        (>= *db-last-access* *db-last-sync*)) ;; no sync since last write
		   (sync-in-progress *db-sync-in-progress*)
                   (min-intersync-delay (configf:lookup-number *configdat* "server" "minimum-intersync-delay" default: 5))
		   (should-sync      (and (not *time-to-exit*)
                                          (> (- (current-seconds) *db-last-sync*) min-intersync-delay))) ;; sync every five seconds minimum, deprecated logic, can probably be removed
		   (start-time       (current-seconds))
                   (cpu-load-adj     (alist-ref 'adj-proc-load (common:get-normalized-cpu-load #f)))
		   (mt-mod-time      (file-modification-time mtpath))
		   (last-sync-start  (if (common:file-exists? start-file)
					 (file-modification-time start-file)
					 0))
		   (last-sync-end    (if (common:file-exists? end-file)
					 (file-modification-time end-file)
					 10))
                   (sync-period      (+ 3 (* cpu-load-adj 30))) ;; as adjusted load increases increase the sync period
		   (recently-synced  (and (< (- start-time mt-mod-time) sync-period) ;; not useful if sync didn't modify megatest.db!
					  (< mt-mod-time last-sync-start)))
		   (sync-done        (<= last-sync-start last-sync-end))
                   (sync-stale       (> start-time (+ last-sync-start sync-stale-seconds)))
		   (will-sync        (and (not *time-to-exit*)       ;; do not start a sync if we are in the process of exiting
                                          (or need-sync should-sync)
					  (or sync-done sync-stale)
					  (not sync-in-progress)
					  (not recently-synced))))
              (debug:print-info 13 *default-log-port* "WD writable-watchdog top of loop.  need-sync="need-sync" sync-in-progress=" sync-in-progress
				" should-sync="should-sync" start-time="start-time" mt-mod-time="mt-mod-time" recently-synced="recently-synced" will-sync="will-sync
				" sync-done=" sync-done " sync-period=" sync-period)
              (if (and (> sync-period 5)
                       (common:low-noise-print 30 "sync-period"))
                  (debug:print-info 0 *default-log-port* "Increased sync period due to long sync times, sync took: " sync-period " seconds."))
	      ;; (if recently-synced (debug:print-info 0 *default-log-port* "Skipping sync due to recently-synced flag=" recently-synced))
	      ;; (debug:print-info 0 *default-log-port* "need-sync: " need-sync " sync-in-progress: " sync-in-progress " should-sync: " should-sync " will-sync: " will-sync)
	      (if will-sync (set! *db-sync-in-progress* #t))
	      (mutex-unlock! *db-multi-sync-mutex*)
	      (if will-sync
                  (let (;; (max-sync-duration  (configf:lookup-number *configdat* "server" "max-sync-duration")) ;; KEEPING THIS AVAILABLE BUT SHOULD NOT USE, I'M PRETTY SURE IT DOES NOT WORK!
                        (sync-start         (current-milliseconds)))
		    (with-output-to-file start-file (lambda ()(print (current-process-id))))
		    
		    ;; put lock here
		    
                    ;; (if (or (not max-sync-duration)
                    ;;        (< sync-duration max-sync-duration)) ;; NOTE: db:sync-to-megatest.db keeps track of time of last sync and syncs incrementally
                        (let ((res        (db:sync-to-megatest.db dbstruct no-sync-db: no-sync-db))) ;; did we sync any data? If so need to set the db touched flag to keep the server alive
                          (set! sync-duration (- (current-milliseconds) sync-start))
                          (if (> res 0) ;; some records were transferred, keep the db alive
                              (begin
                                (mutex-lock! *heartbeat-mutex*)
                                (set! *db-last-access* (current-seconds))
                                (mutex-unlock! *heartbeat-mutex*)
                                (debug:print-info 0 *default-log-port* "sync called, " res " records transferred."))
                              (debug:print-info 2 *default-log-port* "sync called but zero records transferred")))))
;;                         ;; TODO: factor this next routine out into a function
;;                         (with-input-from-pipe ;; this should not block other threads but need to verify this
;;                          (conc "megatest -sync-to-megatest.db -m testsuite:" (common:get-area-name) ":" *toppath*)
;;                          (lambda ()
;;                            (let loop ((inl (read-line))
;;                                       (res #f))
;;                              (if (eof-object? inl)
;;                                  (begin
;;                                    (set! sync-duration (- (current-milliseconds) sync-start))
;;                                    (cond
;;                                     ((not res)
;;                                      (debug:print 0 *default-log-port* "ERROR: sync from /tmp db to megatest.db appears to have failed. Recommended that you stop your runs and run \"megatest -cleanup-db\""))
;;                                     ((> res 0)
;;                                      (mutex-lock! *heartbeat-mutex*)
;;                                      (set! *db-last-access* (current-seconds))
;;                                      (mutex-unlock! *heartbeat-mutex*))))
;;                                  (let ((num-synced (let ((matches (string-match "^Synced (\\d+).*$" inl)))
;;                                                      (if matches
;;                                                          (string->number (cadr matches))
;;                                                          #f))))
;;                                    (loop (read-line)
;;                                          (or num-synced res))))))))))
	      (if will-sync
		  (begin
		    (mutex-lock! *db-multi-sync-mutex*)
		    (set! *db-sync-in-progress* #f)
		    (set! *db-last-sync* start-time)
		    (with-output-to-file end-file (lambda ()(print (current-process-id))))

		    ;; release lock here

		    (mutex-unlock! *db-multi-sync-mutex*)))
	      (if (and debug-mode
		       (> (- start-time last-time) 60))
		  (begin
		    (set! last-time start-time)
		    (debug:print-info 4 *default-log-port* "timestamp -> " (seconds->time-string (current-seconds)) ", time since start -> " (seconds->hr-min-sec (- (current-seconds) *time-zero*))))))
	    
	    ;; keep going unless time to exit
	    ;;
	    (if (not *time-to-exit*)
		(let delay-loop ((count 0))
                  ;;(debug:print-info 13 *default-log-port* "delay-loop top; count="count" pid="(current-process-id)" this-wd-num="this-wd-num" *time-to-exit*="*time-to-exit*)
                                                            
		  (if (and (not *time-to-exit*)
			   (< count 6)) ;; was 11, changing to 4. 
		      (begin
			(thread-sleep! 1)
			(delay-loop (+ count 1))))
		  (if (not *time-to-exit*) (loop))))
	    ;; time to exit, close the no-sync db here
	    (db:no-sync-close-db no-sync-db stmt-cache)
	    (if (common:low-noise-print 30)
		(debug:print-info 0 *default-log-port* "Exiting watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id) ))))))) ;;" this-wd-num="this-wd-num)))))))


)

  ((last-update   0)                 : number)    ;; last query to db got records from before last-update
  ((last-db-time  0)                 : number)    ;; last timestamp on megatest.db
  ((data-changed  #f)                : boolean)   
  ((run-data-offset  0)              : number)      ;; get only 100 items per call, set back to zero when received less than 100 items
  (db-path #f))

;;======================================================================the end

)

(declare (unit debugprint))
(declare (uses margsmod))

(module debugprint
	*
	
;;(import scheme chicken data-structures extras files ports)
(import scheme chicken)
(import margsmod)

(use data-structures extras files ports)
;; (use 
;;   (prefix base64 base64:)
;;   (prefix sqlite3 sqlite3:) 
;;   (srfi 18)
;;   (prefix dbi dbi:)
;;   directory-utils
;;   format
;;   matchable
;;   md5
;;   message-digest
;;   pkts
;;   posix 
;;   regex
;;   regex-case
;;   sparse-vectors
;;   srfi-1
;;   srfi-13
;;   srfi-69
;;   stack
;;   stml2
;;   typed-records 
;;   z3
;;  )

;;======================================================================
;; debug stuff
;;======================================================================

(define verbosity (make-parameter '()))
	 
;;======================================================================
;; (define (debug:print . params) #f)
;; (define (debug:print-info . params) #f)
;; 
;; (define (set-functions dbgp dbgpinfo)
;;   (set! debug:print dbgp)
;;   (set! debug:print-info dbgpinfo))

;;======================================================================
;; this was cached based on results from profiling but it turned out the profiling
;; somehow went wrong - perhaps too many processes writing to it. Leaving the caching
;; in for now but can probably take it out later.
;;
(define (debug:calc-verbosity vstr arg) ;; arg is 'v (verbose) or 'q (quiet)
  (let* ((res (cond
	       ((number? vstr) vstr)
	       ((not (string?  vstr))   1)
	       ;; ((string-match  "^\\s*$" vstr) 1)
	       (vstr           (let ((debugvals  (filter number? (map string->number (string-split vstr ",")))))
				 (cond
				  ((> (length debugvals) 1) debugvals)
				  ((> (length debugvals) 0)(car debugvals))
				  (else 1))))
	       ((eq? arg 'v)   2) ;; verbose
	       ((eq? arg 'q)   0) ;; quiet
	       (else                   1))))
    (verbosity res)
    res))

;;======================================================================
;; check verbosity, #t is ok
#;(define (debug-check-verbosity verbosity vstr)
  (if (not (or (number? verbosity)
	       (list?   verbosity)))
      (begin
	(print "ERROR: Invalid debug value \"" vstr "\"")
	#f)
      #t))

(define (debug:debug-mode n)
  (let* ((vb (verbosity)))
    (cond
   ((and (number? vb)   ;; number number
	 (number? n))
    (<= n vb))
   ((and (list? vb)     ;; list   number
	 (number? n))
    (member n vb))
   ((and (list? vb)     ;; list   list
	 (list? n))
    (not (null? (lset-intersection! eq? vb n))))
   ((and (number? vb)
	 (list? n))
    (member vb n)))))

(define (debug:print n e . params)
  (if (debug:debug-mode n)
      (with-output-to-port (or e (current-error-port))
	(lambda ()
	  ;; (if *logging*
	      ;; (db:log-event (apply conc params))
	  (apply print params)
	  )))) ;; )

(define (debug:print-error n e . params)
  ;; normal print
  (if (debug:debug-mode n)
      (with-output-to-port (if (port? e) e (current-error-port))
	(lambda ()
	  (apply print "ERROR: " params)
	  )))
  ;; pass important messages to stderr
  (if (and (eq? n 0)(not (eq? e (current-error-port)))) 
      (with-output-to-port (current-error-port)
	(lambda ()
	  (apply print "ERROR: " params)
	  ))))

(define (debug:print-info n e . params)
  (if (debug:debug-mode n)
      (with-output-to-port (if (port? e) e (current-error-port))
	(lambda ()
	  (apply print "INFO: (" n ") " params) ;; res)
	  ))))

)

;;======================================================================
;; S E R V E R   U T I L I T I E S 
;;======================================================================

;;======================================================================
;; C L I E N T S
;;======================================================================

  #t);)
	;;(mt:test-set-state-status-by-id run-id test-id new-state new-status new-comment)))

(define (mt:test-set-state-status-by-testname-unless-completed run-id test-name item-path new-state new-status new-comment)
  (let ((test-id (rmt:get-test-id run-id test-name item-path)))
    (mt:test-set-state-status-by-id-unless-completed run-id test-id new-state new-status new-comment)))
    

;;======================================================================

(declare (unit rmtmod))
(declare (uses commonmod))
(declare (uses apimod))
(declare (uses dbmod))
;; (declare (uses transport))
;; (declare (uses servermod))

(module rmtmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18)
(import commonmod)
(import dbmod)
(import apimod)
;; (import transport)
;; (import servermod)

(defstruct alldat
  (areapath #f)
  (ulexdat  #f)
  )

(define (rmtmod:calc-ro-mode runremote *toppath*)

      res)) ;; All good, return res

#;(set-functions rmt:send-receive                       remote-server-url-set!
	       http-transport:close-connections	      remote-conndat-set!
	       debug:print                            debug:print-info
	       remote-ro-mode                         remote-ro-mode-set!
	       remote-ro-mode-checked-set!            remote-ro-mode-checked)

;;======================================================================
;; EVERYTHING FROM TRANSPORT
;;======================================================================

(define (http-transport:make-server-url hostport)
  (if (not hostport)
      #f
      (conc "http://" (car hostport) ":" (cadr hostport))))

;;======================================================================
;; S E R V E R
;; ======================================================================

;; Call this to start the actual server
;;

;; (define *db:process-queue-mutex* (make-mutex))

(define (http-transport:run hostn)
  ;; Configurations for server
  (tcp-buffer-size 2048)
  (max-connections 2048) 
  (debug:print 2 *default-log-port* "Attempting to start the server ...")
  (let* ((db              #f) ;;        (open-db)) ;; we don't want the server to be opening and closing the db unnecesarily
	 (hostname        (get-host-name))
	 (ipaddrstr       (let ((ipstr (if (string=? "-" hostn)
					   ;; (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
					   (server:get-best-guess-address hostname)
					   #f)))
			    (if ipstr ipstr hostn))) ;; hostname))) 
	 (start-port      (portlogger:open-run-close
			   (lambda (db)
			     (portlogger:find-port db))))
	 (link-tree-path  (common:get-linktree))
	 (tmp-area        (common:get-db-tmp-area))
	 (start-file      (conc tmp-area "/.server-start")))
    (debug:print-info 0 *default-log-port* "portlogger recommended port: " start-port)
    ;; set some parameters for the server
    (root-path     (if link-tree-path 
		       link-tree-path
		       (current-directory))) ;; WARNING: SECURITY HOLE. FIX ASAP!
    (handle-directory spiffy-directory-listing)
    (handle-exception (lambda (exn chain)
			(signal (make-composite-condition
				 (make-property-condition 
				  'server
				  'message "server error")))))

    ;; http-transport:handle-directory) ;; simple-directory-handler)
    ;; Setup the web server and a /ctrl interface
    ;;
    (vhost-map `(((* any) . ,(lambda (continue)
			       ;; open the db on the first call 
				 ;; This is were we set up the database connections
			       (let* (($   (request-vars source: 'both))
				      (dat ($ 'dat))
				      (res #f))
				 (cond
				  ((equal? (uri-path (request-uri (current-request)))
					   '(/ "api"))
				   (send-response body:    (api:process-request *dbstruct-db* $) ;; the $ is the request vars proc
						  headers: '((content-type text/plain)))
				   (mutex-lock! *heartbeat-mutex*)
				   (set! *db-last-access* (current-seconds))
				   (mutex-unlock! *heartbeat-mutex*))
				  ;; ((equal? (uri-path (request-uri (current-request))) 
				  ;; 	   '(/ ""))
				  ;;  (send-response body: (http-transport:main-page)))
				;;((equal? (uri-path (request-uri (current-request))) 
				;;	   '(/ "json_api"))
				;; (send-response body: (http-transport:main-page)))
				;;((equal? (uri-path (request-uri (current-request))) 
				;;	   '(/ "runs"))
				;; (send-response body: (http-transport:main-page)))
				;;((equal? (uri-path (request-uri (current-request))) 
				;;	   '(/ any))
				;; (send-response body: "hey there!\n"
				;;		  headers: '((content-type text/plain))))
				;;((equal? (uri-path (request-uri (current-request))) 
				;;	   '(/ "hey"))
				;; (send-response body: "hey there!\n" 
				;;		  headers: '((content-type text/plain))))
                                ;;((equal? (uri-path (request-uri (current-request))) 
				;;	   '(/ "jquery3.1.0.js"))
				;; (send-response body: (http-transport:show-jquery) 
				;;		  headers: '((content-type application/javascript))))
                                ;;((equal? (uri-path (request-uri (current-request))) 
				;;	   '(/ "test_log"))
				;; (send-response body: (http-transport:html-test-log $) 
				;;		  headers: '((content-type text/HTML))))    
                                ;;((equal? (uri-path (request-uri (current-request))) 
				;;	   '(/ "dashboard"))
				;; (send-response body: (http-transport:html-dboard $) 
				;;		  headers: '((content-type text/HTML)))) 
				  (else (continue))))))))
    (handle-exceptions
	exn
      (debug:print 0 *default-log-port* "Failed to create file " start-file ", exn=" exn)
      (with-output-to-file start-file (lambda ()(print (current-process-id)))))
    (http-transport:try-start-server ipaddrstr start-port)))


;; This is recursively run by http-transport:run until sucessful
;;
(define (http-transport:try-start-server ipaddrstr portnum)
  (let ((config-hostname (configf:lookup *configdat* "server" "hostname"))
	(config-use-proxy (equal? (configf:lookup *configdat* "client" "use-http_proxy") "yes")))
    (if (not config-use-proxy)
	(determine-proxy (constantly #f)))
    (debug:print-info 0 *default-log-port* "http-transport:try-start-server time=" (seconds->time-string (current-seconds)) " ipaddrsstr=" ipaddrstr " portnum=" portnum " config-hostname=" config-hostname)
    (handle-exceptions
	exn
	(begin
	  (print-error-message exn)
	  (if (< portnum 64000)
	      (begin 
		(debug:print 0 *default-log-port* "WARNING: attempt to start server failed. Trying again ...")
		(debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
		(debug:print 5 *default-log-port* "exn=" (condition->list exn))
		(portlogger:open-run-close portlogger:set-failed portnum)
		(debug:print 0 *default-log-port* "WARNING: failed to start on portnum: " portnum ", trying next port")
		(thread-sleep! 0.1)
		
		;; get_next_port goes here
		(http-transport:try-start-server ipaddrstr
						 (portlogger:open-run-close portlogger:find-port)))
	      (begin
		(print "ERROR: Tried and tried but could not start the server"))))
      ;; any error in following steps will result in a retry
      (set! *server-info* (list ipaddrstr portnum))
      (debug:print 0 *default-log-port* "INFO: Trying to start server on " ipaddrstr ":" portnum)
      ;; This starts the spiffy server
      ;; NEED WAY TO SET IP TO #f TO BIND ALL
      ;; (start-server bind-address: ipaddrstr port: portnum)
      (if config-hostname ;; this is a hint to bind directly
	  (start-server port: portnum bind-address: (if (equal? config-hostname "-")
							ipaddrstr
							config-hostname))
	  (start-server port: portnum))
      (portlogger:open-run-close
       (lambda (db)
	 (portlogger:set-port db portnum "released")))
      (debug:print 1 *default-log-port* "INFO: server has been stopped"))))


;;======================================================================
;; EVERYTHING FROM SERVERMOD
;;======================================================================

(define (server:make-server-url hostport)
  (if (not hostport)
      #f
      (conc "http://" (car hostport) ":" (cadr hostport))))

;;======================================================================
;; S E R V E R   U T I L I T I E S 
;;======================================================================

;; Get the transport
#;(define (server:get-transport)
  (if *transport-type*
      *transport-type*
      (let ((ttype (string->symbol
		    (or (args:get-arg "-transport")
			(configf:lookup *configdat* "server" "transport")
			"rpc"))))
	(set! *transport-type* ttype)
	ttype)))
	    
;; Generate a unique signature for this server
(define (server:mk-signature)
  (message-digest-string (md5-primitive) 
			 (with-output-to-string
			   (lambda ()
			     (write (list (current-directory)
                                          (current-process-id)
					  (argv)))))))

;; When using zmq this would send the message back (two step process)
;; with spiffy or rpc this simply returns the return data to be returned
;; 
(define (server:reply return-addr query-sig success/fail result)
  (debug:print-info 11 *default-log-port* "server:reply return-addr=" return-addr ", result=" result)
  ;; (send-message pubsock target send-more: #t)
  ;; (send-message pubsock
  (db:obj->string (vector success/fail query-sig result)))
;;  (case (server:get-transport)
;;    ((rpc)  (db:obj->string (vector success/fail query-sig result)))
;;    ((http) (db:obj->string (vector success/fail query-sig result)))
;;    ((fs)   result)
;;    (else 
;;     (debug:print-error 0 *default-log-port* "unrecognised transport type: " *transport-type*)
;;     result)))

(define (server:kind-run areapath)
  ;; look for $MT_RUN_AREA_HOME/logs/server-start-last
  ;; and wait for it to be at least 3 seconds old
  (server:wait-for-server-start-last-flag areapath)
  (if (not (server:check-if-running areapath)) ;; why try if there is already a server running?
      (let* ((last-run-dat (hash-table-ref/default *server-kind-run* areapath '(0 0))) ;; callnum, whenrun
	     (call-num     (car last-run-dat))
	     (when-run     (cadr last-run-dat))
	     (run-delay    (+ (case call-num
				((0)    0)
				((1)   20)
				((2)  300)
				(else 600))
			      (random 5)))   ;; add a small random number just in case a lot of jobs hit the work hosts simultaneously
	     (lock-file    (conc areapath "/logs/server-start.lock")))
	(if	(> (- (current-seconds) when-run) run-delay)
		(let* ((start-flag (conc areapath "/logs/server-start-last")))
		  (common:simple-file-lock-and-wait lock-file expire-time: 15)
                  (debug:print-info  0 *default-log-port* "server:kind-run: touching " start-flag)
		  (system (conc "touch " start-flag)) ;; lazy but safe
		  (server:run areapath)
		  (thread-sleep! 2) ;; don't release the lock for at least a few seconds
		  (common:simple-file-release-lock lock-file)))
	(hash-table-set! *server-kind-run* areapath (list (+ call-num 1)(current-seconds))))))


(define (server:get-num-alive srvlst)
  (let ((num-alive 0))
    (for-each
     (lambda (server)
        (handle-exceptions
          exn
         (begin 
          (debug:print-info 0 *default-log-port*  "Unable to get server start-time and/or mod-time from " server ", exn=" exn))
       (match-let (((mod-time host port start-time server-id pid)
		    server))
	 (let* ((uptime  (- (current-seconds) mod-time))
		(runtime (if start-time
			     (- mod-time start-time)
			     0)))
	   (if (< uptime 5)(set! num-alive (+ num-alive 1)))))))
     srvlst)
    num-alive))

;; given a list of servers get a list of valid servers, i.e. at least
;; 10 seconds old, has started and is less than 1 hour old and is
;; active (i.e. mod-time < 10 seconds
;;
;; mod-time host port start-time pid
;;
;; sort by start-time descending. I.e. get the oldest first. Young servers will thus drop off
;; and servers should stick around for about two hours or so.
;;
(define (server:get-best srvlst)
  (let* ((nums (server:get-num-servers))
	 (now  (current-seconds))
	 (slst (sort
		(filter (lambda (rec)
			  (if (and (list? rec)
				   (> (length rec) 2))
			      (let ((start-time (list-ref rec 3))
				    (mod-time   (list-ref rec 0)))
				;; (print "start-time: " start-time " mod-time: " mod-time)
				(and start-time mod-time
				     (> (- now start-time) 0)    ;; been running at least 0 seconds
				     (< (- now mod-time)   16)   ;; still alive - file touched in last 16 seconds
				     (or (not (configf:lookup *configdat* "server" "runtime")) ;; skip if not set
					 (< (- now start-time)       
					    (+ (- (string->number (configf:lookup *configdat* "server" "runtime"))
					       180)
					    (random 360)))) ;; under one hour running time +/- 180
				     ))
			      #f))
			srvlst)
		(lambda (a b)
		  (< (list-ref a 3)
		     (list-ref b 3))))))
    (if (> (length slst) nums)
	(take slst nums)
	slst)))

(define (server:get-first-best areapath)
  (let ((srvrs (server:get-best (server:get-list areapath))))
    (if (and srvrs
	     (not (null? srvrs)))
	(car srvrs)
	#f)))

(define (server:get-rand-best areapath)
  (let ((srvrs (server:get-best (server:get-list areapath))))
    (if (and (list? srvrs)
	     (not (null? srvrs)))
	(let* ((len (length srvrs))
	       (idx (random len)))
	  (list-ref srvrs idx))
	#f)))

(define (server:record->id servr)
  (handle-exceptions
   exn
   (begin 
     (debug:print-info 0 *default-log-port*  "Unable to get server id from " servr ", exn=" exn)     
   #f)
  (match-let (((mod-time host port start-time server-id pid)
	       servr))
    (if server-id
	server-id
	#f))))

(define (server:record->url servr)
  (handle-exceptions
   exn
   (begin 
     (debug:print-info 0 *default-log-port*  "Unable to get server url from " servr ", exn=" exn)     
   #f)
  (match-let (((mod-time host port start-time server-id pid)
	       servr))
    (if (and host port)
	(conc host ":" port)
	#f))))

(define (server:get-client-signature) ;; BB> why is this proc named "get-"?  it returns nothing -- set! has not return value.
  (if *my-client-signature* *my-client-signature*
      (let ((sig (server:mk-signature)))
        (set! *my-client-signature* sig)
        *my-client-signature*)))

;; wait for server=start-last to be three seconds old
;;
(define (server:wait-for-server-start-last-flag areapath)
  (let* ((start-flag (conc areapath "/logs/server-start-last"))
	 ;;; THIS INTERACTS WITH [server] timeout. Suggest using 0.1 or above for timeout (6 seconds)
	 (reftime    (configf:lookup-number *configdat* "server" "idletime" default: 4))
	 (server-key (conc (get-host-name) "-" (current-process-id))))
    (if (file-exists? start-flag)
	(let* ((fmodtime (file-modification-time start-flag))
	       (delta    (- (current-seconds) fmodtime))
	       (all-go   (> delta reftime)))
	  (if (and all-go
		   (begin
		     (with-output-to-file start-flag
		       (lambda ()
			 (print server-key)))
		     (thread-sleep! 0.25)
		     (let ((res (with-input-from-file start-flag
				  (lambda ()
				    (read-line)))))
		       (equal? server-key res))))
	      #t ;; (system (conc "touch " start-flag)) ;; lazy but safe
	      (begin
		(debug:print-info 0 *default-log-port* "Gating server start, last start: "
				  fmodtime ", delta: " delta ", reftime: " reftime ", all-go=" all-go)
		(thread-sleep! reftime)
		(server:wait-for-server-start-last-flag areapath)))))))

(define (server:get-num-servers #!key (numservers 2))
  (let ((ns (string->number
	     (or (configf:lookup *configdat* "server" "numservers") "notanumber"))))
    (or ns numservers)))

(define (server:kill servr)
  (handle-exceptions
    exn
    (begin 
      (debug:print-info 0 *default-log-port*  "Unable to get host and/or port from " servr ", exn=" exn)     
    #f)
  (match-let (((mod-time hostname port start-time server-id pid)
	       servr))
    (tasks:kill-server hostname pid))))

;; called in megatest.scm, host-port is string hostname:port
;;
;; NOTE: This is NOT called directly from clients as not all transports support a client running
;;       in the same process as the server.
;;
;; run ping in separate process, safest way in some cases
;;
(define (server:ping-server ifaceport)
  (with-input-from-pipe 
   (conc (common:get-megatest-exe) " -ping " ifaceport)
   (lambda ()
     (let loop ((inl (read-line))
		(res "NOREPLY"))
       (if (eof-object? inl)
	   (case (string->symbol res)
	     ((NOREPLY)  #f)
	     ((LOGIN_OK) #t)
	     (else       #f))
	   (loop (read-line) inl))))))

;; NOT USED (well, ok, reference in rpc-transport but otherwise not used).
;;
(define (server:login toppath)
  (lambda (toppath)
    (set! *db-last-access* (current-seconds)) ;; might not be needed.
    (if (equal? *toppath* toppath)
	#t
	#f)))

;; timeout is hms string: 1h 5m 3s, default is 1 minute
;;
(define (server:expiration-timeout)
  (let ((tmo (configf:lookup *configdat* "server" "timeout")))
    (if (and (string? tmo)
	     (common:hms-string->seconds tmo)) ;; BUG: hms-string->seconds is broken, if given "10" returns 0. Also, it doesn't belong in this logic unless the string->number is changed below
        (* 3600 (string->number tmo))
	60)))

(define (server:get-best-guess-address hostname)
  (let ((res #f))
    (for-each 
     (lambda (adr)
       (if (not (eq? (u8vector-ref adr 0) 127))
	   (set! res adr)))
     ;; NOTE: This can fail when there is no mention of the host in /etc/hosts. FIXME
     (vector->list (hostinfo-addresses (hostname->hostinfo hostname))))
    (string-intersperse 
     (map number->string
	  (u8vector->list
	   (if res res (hostname->ip hostname)))) ".")))

;; (define server:sync-lock-token "SERVER_SYNC_LOCK")
;; (define (server:release-sync-lock)
;;   (db:no-sync-del! *no-sync-db* server:sync-lock-token))
;; (define (server:have-sync-lock?)
;;   (let* ((have-lock-pair (db:no-sync-get-lock *no-sync-db* server:sync-lock-token))
;;          (have-lock?     (car have-lock-pair))
;;          (lock-time      (cdr have-lock-pair))
;;          (lock-age       (- (current-seconds) lock-time)))
;;     (cond
;;      (have-lock? #t)
;;      ((>lock-age
;;        (* 3 (configf:lookup-number *configdat* "server" "minimum-intersync-delay" default: 180)))
;;       (server:release-sync-lock)
;;       (server:have-sync-lock?))
;;      (else #f))))

;; moving this here as it needs access to db and cannot be in common.
;;

(define (server:get-bruteforce-syncer dbstruct #!key (fork-to-background #f) (persist-until-sync #f))
  (let* ((sqlite-exe   (or (get-environment-variable "MT_SQLITE3_EXE"))) ;; defined in cfg.sh
         (sync-log     (or ;; (args:get-arg "-sync-log")
			(conc *toppath* "/logs/sync-" (current-process-id) "-" (get-host-name) ".log")))
	 (tmp-area     (common:get-db-tmp-area))
	 (tmp-db       (conc tmp-area "/megatest.db"))
	 (staging-file (conc *toppath* "/.megatest.db"))
	 (mtdbfile     (conc *toppath* "/megatest.db"))
	 (lockfile     (common:get-sync-lock-filepath))
         (sync-cmd-core     (conc sqlite-exe" " tmp-db " .dump | "sqlite-exe" " staging-file "&>"sync-log))
         (sync-cmd     (if fork-to-background 
                           (conc "/usr/bin/env NBFAKE_LOG="*toppath*"/logs/last-server-sync-"(current-process-id)".log nbfake \""sync-cmd-core" && /bin/mv -f " staging-file " " mtdbfile" \"")
                           sync-cmd-core))
         (default-min-intersync-delay 2)
	 (min-intersync-delay (configf:lookup-number *configdat* "server" "minimum-intersync-delay" default: default-min-intersync-delay))
         (default-duty-cycle 0.1)
         (duty-cycle   (configf:lookup-number *configdat* "server" "sync-duty-cycle" default: default-duty-cycle))
         (last-sync-seconds 10) ;; we will adjust this to a measurement and delay last-sync-seconds * (1 - duty-cycle)
         (calculate-off-time (lambda (work-duration duty-cycle)
                                  (* (/ (- 1 duty-cycle) duty-cycle) last-sync-seconds)))
         (off-time min-intersync-delay) ;; adjusted in closure below.
         (do-a-sync
          (lambda ()
            ;; (BB> "Start do-a-sync with fork-to-background="fork-to-background" persist-until-sync="persist-until-sync)
            (let* ((finalres
                    (let retry-loop ((num-tries 0))
                         (if (common:simple-file-lock lockfile)
	                     (begin
                               (cond
                                ((not (or fork-to-background persist-until-sync))
                                 (debug:print 0 *default-log-port* "INFO: syncer thread sleeping for max of (server.minimum-intersync-delay="min-intersync-delay
                                              " , off-time="off-time" seconds ]")
                                 (thread-sleep! (max off-time min-intersync-delay)))
                                (else
                                 (debug:print 0 *default-log-port* "INFO: syncer thread NOT sleeping ; maybe time-to-exit...")))

                               (if (not (configf:lookup *configdat* "server" "disable-db-snapshot"))
                                   (common:snapshot-file mtdbfile subdir: ".db-snapshot"))
		               (delete-file* staging-file)
		               (let* ((start-time (current-milliseconds))
                                      (res (system sync-cmd))
                                      (dbbackupfile (conc mtdbfile ".backup"))
                                      (res2 
                                       (cond
                                        ((eq? 0 res )
                                         (handle-exceptions
                                            exn
                                            #f
                                         (if (file-exists? dbbackupfile)
		                           (delete-file* dbbackupfile)
                                         )
                                         (if (eq? 0 (file-size sync-log))
                                             (delete-file* sync-log))
		                         (system (conc "/bin/mv " staging-file " " mtdbfile))
                                         
                                         (set! last-sync-seconds (/ (- (current-milliseconds) start-time) 1000))
                                         (set! off-time (calculate-off-time
                                                         last-sync-seconds
                                                         (cond
                                                          ((and (number? duty-cycle) (> duty-cycle 0) (< duty-cycle 1))
                                                           duty-cycle)
                                                          (else
                                                           (debug:print 0 *default-log-port* "WARNING: ["(common:human-time)"] server.sync-duty-cycle is invalid.  Should be a number between 0 and 1, but "duty-cycle" was specified.  Using default value: "default-duty-cycle)
                                                           default-duty-cycle))))
                                         
                                         (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" SYNC took "last-sync-seconds" sec")
                                         (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" SYNC took "last-sync-seconds" sec ; with duty-cycle of "duty-cycle" off time is now "off-time)
                                         'sync-completed))
                                        (else
                                         (system (conc "/bin/cp "sync-log" "sync-log".fail"))
                                         (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)"] Sync failed. See log at "sync-log".fail")
                                         (if (file-exists? (conc mtdbfile ".backup"))
                                             (system (conc "/bin/cp "mtdbfile ".backup " mtdbfile)))
                                         #f))))
                                 (common:simple-file-release-lock lockfile)
                                 ;; (BB> "released lockfile: " lockfile)
                                 #;(when (common:file-exists? lockfile)
                                   (BB> "DID NOT ACTUALLY RELEASE LOCKFILE"))
                                 res2) ;; end let
                               );; end begin
                             ;; else
                             (cond
                              (persist-until-sync
                               (thread-sleep! 1)
                               (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" other SYNC in progress; we're in a fork-to-background so we need to succeed.  Let's wait a jiffy and and try again. num-tries="num-tries" (waiting for lockfile="lockfile" to disappear)")
                               (retry-loop (add1 num-tries)))
                              (else
                               (thread-sleep! (max off-time (+ last-sync-seconds min-intersync-delay)))
                               (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" other SYNC in progress; not syncing.")
                               'parallel-sync-in-progress))
                             ) ;; end if got lockfile
                         )
                    ))
              ;; (BB> "End do-a-sync with fork-to-background="fork-to-background" persist-until-sync="persist-until-sync" and result="finalres)
              finalres)
            ) ;; end lambda
          ))
    do-a-sync))

;; (define (debug:print . params) #f)
;; (define (debug:print-info . params) #f)
;; 
;; (define (set-functions dbgp dbgpinfo)
;;   (set! debug:print dbgp)
;;   (set! debug:print-info dbgpinfo))

;; (define  *server-loop-heart-beat* (current-seconds))

;;======================================================================
;; P K T S   S T U F F 
;;======================================================================

;; ???

;;======================================================================
;; P K T S   S T U F F 
;;======================================================================

;; ???

;;======================================================================
;; S E R V E R
;;======================================================================

;; Call this to start the actual server
;;

;; all routes though here end in exit ...
;;
;; start_server
;;
(define (server:launch run-id transport-type)
  (case transport-type
    ((http)(http-transport:launch))
    ;;((nmsg)(nmsg-transport:launch run-id))
    ;;((rpc)  (rpc-transport:launch run-id))
    (else (debug:print-error 0 *default-log-port* "unknown server type " transport-type))))

;; Given a run id start a server process    ### NOTE ### > file 2>&1 
;; if the run-id is zero and the target-host is set 
;; try running on that host
;;   incidental: rotate logs in logs/ dir.
;;
(define  (server:run areapath) ;; areapath is *toppath* for a given testsuite area
  (let* ((curr-host   (get-host-name))
         ;; (attempt-in-progress (server:start-attempted? areapath))
         ;; (dot-server-url (server:check-if-running areapath))
	 (curr-ip     (server:get-best-guess-address curr-host))
	 (curr-pid    (current-process-id))
	 (homehost    (common:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" ))
	 (target-host (car homehost))
	 (testsuite   (common:get-testsuite-name))
	 (logfile     (conc areapath "/logs/server.log")) ;; -" curr-pid "-" target-host ".log"))
	 (profile-mode (or (configf:lookup *configdat* "misc" "profilesw")
			   ""))
	 (cmdln (conc (common:get-megatest-exe)
		      " -server " (or target-host "-") (if (equal? (configf:lookup *configdat* "server" "daemonize") "yes")
							   " -daemonize "
							   "")
		      ;; " -log " logfile
		      " -m testsuite:" testsuite
		      " " profile-mode
		      )) ;; (conc " >> " logfile " 2>&1 &")))))
	 (log-rotate  (make-thread common:rotate-logs  "server run, rotate logs thread")) ;; why are we rotating logs here? This is a sensitive location with a lot going on!?
         (load-limit  (configf:lookup-number *configdat* "jobtools" "max-server-start-load" default: 3.0)))
    ;; we want the remote server to start in *toppath* so push there
    (push-directory areapath)
    (debug:print 0 *default-log-port* "INFO: Trying to start server (" cmdln ") ...")
    (thread-start! log-rotate)
    
    ;; host.domain.tld match host?
    (if (and target-host 
	     ;; look at target host, is it host.domain.tld or ip address and does it 
	     ;; match current ip or hostname
	     (not (string-match (conc "("curr-host "|" curr-host"\\..*)") target-host))
	     (not (equal? curr-ip target-host)))
	(begin
	  (debug:print-info 0 *default-log-port* "Starting server on " target-host ", logfile is " logfile)
	  (setenv "TARGETHOST" target-host)))
      
    (setenv "TARGETHOST_LOGF" logfile)
    (thread-sleep! (/ (random 5000) 1000)) ;; add about a random (up to 5 seconds) initial delay. It seems pretty common that many running tests request a server at the same time
    ;; (common:wait-for-normalized-load load-limit " delaying server start due to load" target-host) ;; do not try starting servers on an already overloaded machine, just wait forever
    #;(common:wait-for-homehost-load load-limit (conc " delaying server start due to load on homehost. limit is " load-limit))
    (system (conc "nbfake " cmdln))
    (unsetenv "TARGETHOST_LOGF")
    (if (get-environment-variable "TARGETHOST")(unsetenv "TARGETHOST"))
    (thread-join! log-rotate)
    (pop-directory)))











(define (server:ping host-port-in server-id #!key (do-exit #f))
  (let ((host:port (if (not host-port-in) ;; use read-dotserver to find
		       #f ;; (server:check-if-running *toppath*)
		;; (if (number? host-port-in) ;; we were handed a server-id
		;; 	   (let ((srec (tasks:get-server-by-id (db:delay-if-busy (tasks:open-db)) host-port-in)))
		;; 	     ;; (print "srec: " srec " host-port-in: " host-port-in)
		;; 	     (if srec
		;; 		 (conc (vector-ref srec 3) ":" (vector-ref srec 4))
		;; 		 (conc "no such server-id " host-port-in)))
		       host-port-in))) ;; )
    (let* ((host-port (if host:port
			  (let ((slst (string-split   host:port ":")))
			    (if (eq? (length slst) 2)
				(list (car slst)(string->number (cadr slst)))
				#f))
			  #f)))
;;	   (toppath       (launch:setup)))
      ;; (print "host-port=" host-port)
      (if (not host-port)
	  (begin
	    (if host-port-in
		(debug:print 0 *default-log-port*  "ERROR: bad host:port"))
	    (if do-exit (exit 1))
	    #f)
	  (let* ((iface      (car host-port))
		 (port       (cadr host-port))
		 (server-dat (http-transport:client-connect iface port server-id))
		 (login-res  (rmt:login-no-auto-client-setup server-dat)))
	    (if (and (list? login-res)
		     (car login-res))
		(begin
		  ;; (print "LOGIN_OK")
		  (if do-exit (exit 0))
		  #t)
		(begin
		  ;; (print "LOGIN_FAILED")
		  (if do-exit (exit 1))
		  #f)))))))


;; kind start up of servers, wait 40 seconds before allowing another server for a given
;; run-id to be launched
;;
;; this one seems to be the general entry point
;;
(define (server:start-and-wait areapath #!key (timeout 60))
  (let ((give-up-time (+ (current-seconds) timeout)))
    (let loop ((server-info (server:check-if-running areapath))
	       (try-num    0))
      (if (or server-info
	      (> (current-seconds) give-up-time)) ;; server-url will be #f if no server available.
	  (server:record->url server-info)
	  (let ((num-ok (length (server:get-best (server:get-list areapath)))))
	    (if (and (> try-num 0)  ;; first time through simply wait a little while then try again
		     (< num-ok 1))  ;; if there are no decent candidates for servers then try starting a new one
		(server:kind-run areapath))
	    (thread-sleep! 5)
	    (loop (server:check-if-running areapath)
		  (+ try-num 1)))))))


;; no longer care if multiple servers are started by accident. older servers will drop off in time.
;;
(define (server:check-if-running areapath) ;;  #!key (numservers "2"))
  (let* ((ns            (server:get-num-servers))
	 (servers       (server:get-best (server:get-list areapath))))
    (if (or (and servers
		 (null? servers))
	    (not servers)
	    (and (list? servers)
		 (< (length servers) (random ns)))) ;; somewhere between 0 and numservers
        #f
        (let loop ((hed (car servers))
                   (tal (cdr servers)))
          (let ((res (server:check-server hed)))
            (if res
                hed
                (if (null? tal)
                    #f
                    (loop (car tal)(cdr tal)))))))))

;; ping the given server
;;
(define (server:check-server server-record)
  (let* ((server-url (server:record->url server-record))
         (server-id (server:record->id server-record)) 
         (res        (case *transport-type*
                       ((http)(server:ping server-url server-id))
                       ;; ((nmsg)(nmsg-transport:ping (tasks:hostinfo-get-interface server)
                       )))
    (if res
        server-url
	#f)))


(define server:try-running server:run) ;; there is no more per-run servers ;; REMOVE ME. BUG.















;;======================================================================
;; REMANANTS OF HTTP_TRANSPORT
;;======================================================================

(define *http-mutex* (make-mutex))

;; NOTE: Large block of code from 32436b426188080f72fceb6894af541fbad9921e removed here
;;       I'm pretty sure it is defunct.

;; This next block all imported en-mass from the api branch
(define *http-requests-in-progress* 0)
(define *http-connections-next-cleanup* (current-seconds))

(define (http-transport:get-time-to-cleanup)
  (let ((res #f))
    (mutex-lock! *http-mutex*)
    (set! res (> (current-seconds) *http-connections-next-cleanup*))
    (mutex-unlock! *http-mutex*)
    res))

(define (http-transport:inc-requests-count)
  (mutex-lock! *http-mutex*)
  (set! *http-requests-in-progress* (+ 1 *http-requests-in-progress*))
  ;; Use this opportunity to slow things down iff there are too many requests in flight
  (if (> *http-requests-in-progress* 5)
      (begin
	(debug:print-info 0 *default-log-port* "Whoa there buddy, ease up...")
	(thread-sleep! 1)))
  (mutex-unlock! *http-mutex*))

(define (http-transport:dec-requests-count proc) 
  (mutex-lock! *http-mutex*)
  (proc)
  (set! *http-requests-in-progress* (- *http-requests-in-progress* 1))
  (mutex-unlock! *http-mutex*))

(define (http-transport:dec-requests-count-and-close-all-connections)
  (set! *http-requests-in-progress* (- *http-requests-in-progress* 1))
  (let loop ((etime (+ (current-seconds) 5))) ;; give up in five seconds
    (if (> *http-requests-in-progress* 0)
	(if (> etime (current-seconds))
	    (begin
	      (thread-sleep! 0.05)
	      (loop etime))
	    (debug:print-error 0 *default-log-port* "requests still in progress after 5 seconds of waiting. I'm going to pass on cleaning up http connections"))
	(close-all-connections!)))
  (set! *http-connections-next-cleanup* (+ (current-seconds) 10))
  (mutex-unlock! *http-mutex*))

(define (http-transport:inc-requests-and-prep-to-close-all-connections)
  (mutex-lock! *http-mutex*)
  (set! *http-requests-in-progress* (+ 1 *http-requests-in-progress*)))

;; Send "cmd" with json payload "params" to serverdat and receive result
;;
(define (http-transport:client-api-send-receive run-id serverdat cmd params #!key (numretries 3)(area-dat #f))
  (let* ((fullurl    (if (vector? serverdat)
			 (http-transport:server-dat-get-api-req serverdat)
			 (begin
			   (debug:print 0 *default-log-port* "FATAL ERROR: http-transport:client-api-send-receive called with no server info")
			   (exit 1))))
	 (res        (vector #f "uninitialized"))
	 (success    #t)
	 (sparams    (db:obj->string params transport: 'http))
	 (runremote  (or area-dat *runremote*))
         (server-id   (if (vector? serverdat) 
                           (http-transport:server-dat-get-server-id serverdat)
                           (begin
			     (debug:print 0 *default-log-port* "FATAL ERROR: http-transport:client-api-send-receive called with no server info")
			     (exit 1)))))
       (debug:print-info 11 *default-log-port* "cmd=" cmd " fullurl=" fullurl " server-id=" server-id " current time:" (current-seconds)) 

       ;; set up the http-client here
       (max-retry-attempts 1)
       ;; consider all requests indempotent
       (retry-request? (lambda (request)
			 #f))
       ;; send the data and get the response
       ;; extract the needed info from the http data and 
       ;; process and return it.
       (let* ((send-recieve (lambda ()
			      (mutex-lock! *http-mutex*)
			      ;; (condition-case (with-input-from-request "http://localhost"; #f read-lines)
			      ;;					       ((exn http client-error) e (print e)))
			      (set! res (vector                ;;; DON'T FORGET - THIS IS THE CLIENT SIDE! NOTE: consider moving this to client.scm since we are only supporting http transport at this time.
					 success
					 (db:string->obj 
					  (handle-exceptions
					      exn
					      (let ((call-chain (get-call-chain))
						    (msg        ((condition-property-accessor 'exn 'message) exn)))
						(set! success #f)
                                                (if (debug:debug-mode 1)
                                                    (debug:print-info 0 *default-log-port* "couldn't talk to server, trying again ...")
                                                    (begin
                                                      (debug:print 0 *default-log-port* "WARNING: failure in with-input-from-request to " fullurl ".")
                                                      (debug:print 0 *default-log-port* " message: " msg ", exn=" exn)
                                                      (debug:print 0 *default-log-port* " cmd: " cmd " params: " params " key:" (or server-id "thekey"))
                                                      (debug:print 0 *default-log-port* " call-chain: " call-chain)))
                                                (if runremote
						    (remote-conndat-set! runremote #f))
						;; Killing associated server to allow clean retry.")
						;; (tasks:kill-server-run-id run-id)  ;; better to kill the server in the logic that called this routine?
						(mutex-unlock! *http-mutex*)
					     ;;; (signal (make-composite-condition
					     ;;;          (make-property-condition 'commfail 'message "failed to connect to server")))
					     ;;; "communications failed"
						(db:obj->string #f))
					    (with-input-from-request ;; was dat
					     fullurl 
					     (list (cons 'key (or server-id   "thekey"))
						   (cons 'cmd cmd)
						   (cons 'params sparams))
					     read-string))
					  transport: 'http)
                                         0)) ;; added this speculatively
			      ;; Shouldn't this be a call to the managed call-all-connections stuff above?
			      (close-all-connections!)
			      (mutex-unlock! *http-mutex*)
			      ))
	      (time-out     (lambda ()
			      (thread-sleep! 45)
			      (debug:print 0 *default-log-port* "WARNING: send-receive took more than 45 seconds!!")
			      #f))
	      (th1 (make-thread send-recieve "with-input-from-request"))
	      (th2 (make-thread time-out     "time out")))
	 (thread-start! th1)
	 (thread-start! th2)
	 (thread-join! th1)
          (vector-set! res 0 success)
	 (thread-terminate! th2)
	 (if (vector? res)
	     (if (vector-ref res 0) ;; this is the first flag or the second flag? 
                 (let* ((res-dat (vector-ref res 1)))
                    (if (and (string? res-dat) (string-contains res-dat "server-id mismatch"))
                     (signal (make-composite-condition
		          (make-property-condition 
		       'servermismatch
		       'message  (vector-ref res 1))))       
		      res)) ;; this is the *inner* vector? seriously? why?
                 (if (debug:debug-mode 11)
                     (let ((call-chain (get-call-chain))) ;; note: this code also called in nmsg-transport - consider consolidating it
                       (print-call-chain (current-error-port))
                       (debug:print-error 11 *default-log-port* "error above occured at server, res=" res) ;; " message: " ((condition-property-accessor 'exn 'message) exn))
                       (debug:print 11 *default-log-port* " server call chain:")
                       (pp (vector-ref res 1) (current-error-port))
                       (signal (vector-ref res 0)))
                     res))
	     (signal (make-composite-condition
		      (make-property-condition 
		       'timeout
		       'message "nmsg-transport:client-api-send-receive-raw timed out talking to server")))))))

;; careful closing of connections stored in *runremote*
;;
(define (http-transport:close-connections #!key (area-dat #f))
  (let* ((runremote  (or area-dat *runremote*))
	 (server-dat (if runremote
                         (remote-conndat runremote)
                         #f))) ;; (hash-table-ref/default *runremote* run-id #f)))
    (if (vector? server-dat)
	(let ((api-dat (http-transport:server-dat-get-api-uri server-dat)))
	  (handle-exceptions
	    exn
	    (begin
	      (print-call-chain *default-log-port*)
	      (debug:print-error 0 *default-log-port* " closing connection failed with error: " ((condition-property-accessor 'exn 'message) exn) ", exn=" exn))
	    (close-connection! api-dat)
            ;;(close-idle-connections!)
	    #t))
	#f)))


(define (make-http-transport:server-dat)(make-vector 6))
(define (http-transport:server-dat-get-iface         vec)    (vector-ref  vec 0))
(define (http-transport:server-dat-get-port          vec)    (vector-ref  vec 1))
(define (http-transport:server-dat-get-api-uri       vec)    (vector-ref  vec 2))
(define (http-transport:server-dat-get-api-url       vec)    (vector-ref  vec 3))
(define (http-transport:server-dat-get-api-req       vec)    (vector-ref  vec 4))
(define (http-transport:server-dat-get-last-access   vec)    (vector-ref  vec 5))
;(define (http-transport:server-dat-get-socket        vec)    (vector-ref  vec 6))
(define (http-transport:server-dat-get-server-id     vec)    (vector-ref  vec 6))

(define (http-transport:server-dat-make-url vec)
  (if (and (http-transport:server-dat-get-iface vec)
	   (http-transport:server-dat-get-port  vec))
      (conc "http://" 
	    (http-transport:server-dat-get-iface vec)
	    ":"
	    (http-transport:server-dat-get-port  vec))
      #f))

(define (http-transport:server-dat-update-last-access vec)
  (if (vector? vec)
      (vector-set! vec 5 (current-seconds))
      (begin
	(print-call-chain (current-error-port))
	(debug:print-error 0 *default-log-port* "call to http-transport:server-dat-update-last-access with non-vector!!"))))

;;
;; connect
;;
(define (http-transport:client-connect iface port server-id) 
  (let* ((api-url      (conc "http://" iface ":" port "/api"))
	 (api-uri      (uri-reference (conc "http://" iface ":" port "/api")))
	 (api-req      (make-request method: 'POST uri: api-uri))
	 (server-dat   (vector iface port api-uri api-url api-req (current-seconds) server-id)))
    server-dat))

;; run http-transport:keep-running in a parallel thread to monitor that the db is being 
;; used and to shutdown after sometime if it is not.
;;
(define (http-transport:keep-running) 
  ;; if none running or if > 20 seconds since 
  ;; server last used then start shutdown
  ;; This thread waits for the server to come alive
  (debug:print-info 0 *default-log-port* "Starting the sync-back, keep alive thread in server")
  (let* ((sdat              #f)
	 (tmp-area          (common:get-db-tmp-area))
	 (started-file      (conc tmp-area "/.server-started"))
	 (server-start-time (current-seconds))
	 (server-info (let loop ((start-time (current-seconds))
				 (changed    #t)
				 (last-sdat  "not this"))
                        (begin ;; let ((sdat #f))
			  (thread-sleep! 0.01)
			  (debug:print-info 0 *default-log-port* "Waiting for server alive signature")
                          (mutex-lock! *heartbeat-mutex*)
                          (set! sdat *server-info*)
                          (mutex-unlock! *heartbeat-mutex*)
                          (if (and sdat
				   (not changed)
				   (> (- (current-seconds) start-time) 2))
			      (begin
				(debug:print-info 0 *default-log-port* "Received server alive signature")
                                (common:save-pkt `((action . alive)
                                                   (T      . server)
                                                   (pid    . ,(current-process-id))
                                                   (ipaddr . ,(car sdat))
                                                   (port   . ,(cadr sdat)))
                                                 *configdat* #t)
				sdat)
                              (begin
				(debug:print-info 0 *default-log-port* "Still waiting, last-sdat=" last-sdat)
                                (sleep 4)
				(if (> (- (current-seconds) start-time) 120) ;; been waiting for two minutes
				    (begin
				      (debug:print-error 0 *default-log-port* "transport appears to have died, exiting server")
                                      (common:save-pkt `((action . died)
                                                         (T      . server)
                                                         (pid    . ,(current-process-id))
                                                         (ipaddr . ,(car sdat))
                                                         (port   . ,(cadr sdat))
                                                         (msg    . "Transport died?"))
						       *configdat* #t)
				      (exit))
				    (loop start-time
					  (equal? sdat last-sdat)
					  sdat)))))))
	 (iface       (car server-info))
         (port        (cadr server-info))
         (last-access 0)
	 (server-timeout (server:expiration-timeout))
	 (server-going  #f)
	 (server-log-file (args:get-arg "-log"))) ;; always set when we are a server

    (handle-exceptions
	exn
      (debug:print 0 *default-log-port* "Failed to create " started-file ", exn=" exn)
      (with-output-to-file started-file (lambda ()(print (current-process-id)))))

    (let loop ((count         0)
	       (server-state 'available)
	       (bad-sync-count 0)
	       (start-time     (current-milliseconds)))
      ;; Use this opportunity to sync the tmp db to megatest.db
      (if (not server-going) ;; *dbstruct-db* 
	  (begin
	    (debug:print 0 *default-log-port* "SERVER: dbprep")
	    (set! *dbstruct-db*  (db:setup #t)) ;;  run-id))
	    (set! server-going #t)
	    (debug:print 0 *default-log-port* "SERVER: running, megatest version: " (common:get-full-version)) ;; NOTE: the server is NOT yet marked as running in the log. We do that in the keep-running routine.
	    (thread-start! *watchdog*)))
      
      ;; when things go wrong we don't want to be doing the various queries too often
      ;; so we strive to run this stuff only every four seconds or so.
      (let* ((sync-time (- (current-milliseconds) start-time))
	    (rem-time  (quotient (- 4000 sync-time) 1000)))
	(if (and (<= rem-time 4)
		 (>  rem-time 0))
	    (thread-sleep! rem-time)))
      
      (if (< count 1) ;; 3x3 = 9 secs aprox
	  (loop (+ count 1) 'running bad-sync-count (current-milliseconds)))
      
      ;; Check that iface and port have not changed (can happen if server port collides)
      (mutex-lock! *heartbeat-mutex*)
      (set! sdat *server-info*)
      (mutex-unlock! *heartbeat-mutex*)
      
      (if (not (equal? sdat (list iface port)))
	  (let ((new-iface (car sdat))
		(new-port  (cadr sdat)))
	    (debug:print-info 0 *default-log-port* "WARNING: interface changed, refreshing iface and port info")
	    (set! iface new-iface)
	    (set! port  new-port)
             (if (not *server-id*)
              (set! *server-id* (server:mk-signature)))
            (debug:print 0 *default-log-port* (current-seconds) (current-directory) (current-process-id) (argv))
	    (debug:print 0 *default-log-port* "SERVER STARTED: " iface ":" port " AT " (current-seconds) " server-id: " *server-id*)
	    (flush-output *default-log-port*)))
      
      ;; Transfer *db-last-access* to last-access to use in checking that we are still alive
      (mutex-lock! *heartbeat-mutex*)
      (set! last-access *db-last-access*)
      (mutex-unlock! *heartbeat-mutex*)
      
      (if (common:low-noise-print 120 (conc "server running on " iface ":" port))
	  (begin
             (if (not *server-id*)
              (set! *server-id* (server:mk-signature)))
            (debug:print 0 *default-log-port* (current-seconds) (current-directory) (current-process-id) (argv))   
	    (debug:print 0 *default-log-port* "SERVER STARTED: " iface ":" port " AT " (current-seconds) " server-id: " *server-id*)
	    (flush-output *default-log-port*)))
      (if (common:low-noise-print 60 "dbstats")
	  (begin
	    (debug:print 0 *default-log-port* "Server stats:")
	    (db:print-current-query-stats)))
      (let* ((hrs-since-start  (/ (- (current-seconds) server-start-time) 3600)))
	(cond
         ((and *server-run*
	       (> (+ last-access server-timeout)
		  (current-seconds)))
          (if (common:low-noise-print 120 "server continuing")
              (debug:print-info 0 *default-log-port* "Server continuing, seconds since last db access: " (- (current-seconds) last-access))
	      (let ((curr-time (current-seconds)))
		(handle-exceptions
		    exn
		    (debug:print 0 *default-log-port* "ERROR: Failed to change timestamp on log file " server-log-file ". Are you out of space on that disk? exn=" exn)
		  (if (not *server-overloaded*)
		      (change-file-times server-log-file curr-time curr-time)))))
          (loop 0 server-state bad-sync-count (current-milliseconds)))
         (else
          (debug:print-info 0 *default-log-port* "Server timed out. seconds since last db access: " (- (current-seconds) last-access))
          (http-transport:server-shutdown port)))))))

(define (http-transport:server-shutdown port)
  (begin
    ;;(BB> "http-transport:server-shutdown called")
    (debug:print-info 0 *default-log-port* "Starting to shutdown the server. pid="(current-process-id))
    ;;
    ;; start_shutdown
    ;;
    (set! *time-to-exit* #t) ;; tell on-exit to be fast as we've already cleaned up
    (portlogger:open-run-close portlogger:set-port port "released")
    (thread-sleep! 1)

    ;; (debug:print-info 0 *default-log-port* "Max cached queries was    " *max-cache-size*)
    ;; (debug:print-info 0 *default-log-port* "Number of cached writes   " *number-of-writes*)
    ;; (debug:print-info 0 *default-log-port* "Average cached write time "
    ;; 		      (if (eq? *number-of-writes* 0)
    ;; 			  "n/a (no writes)"
    ;; 			  (/ *writes-total-delay*
    ;; 			     *number-of-writes*))
    ;; 		      " ms")
    ;; (debug:print-info 0 *default-log-port* "Number non-cached queries "  *number-non-write-queries*)
    ;; (debug:print-info 0 *default-log-port* "Average non-cached time   "
    ;; 		      (if (eq? *number-non-write-queries* 0)
    ;; 			  "n/a (no queries)"
    ;; 			  (/ *total-non-write-delay* 
    ;; 			     *number-non-write-queries*))
    ;; 		      " ms")
    
    (db:print-current-query-stats)
    (common:save-pkt `((action . exit)
                       (T      . server)
                       (pid    . ,(current-process-id)))
                     *configdat* #t)
    (debug:print-info 0 *default-log-port* "Server shutdown complete. Exiting")
    (exit)))

;; all routes though here end in exit ...
;;
;; start_server? 
;;
(define (http-transport:launch)
  ;; check that a server start is in progress, pause or exit if so
  (let* ((tmp-area            (common:get-db-tmp-area))
	 (server-start        (conc tmp-area "/.server-start"))
	 (server-started      (conc tmp-area "/.server-started"))
	 (start-time          (common:lazy-modification-time server-start))
	 (started-time        (common:lazy-modification-time server-started))
	 (server-starting     (< start-time started-time)) ;; if start-time is less than started-time then a server is still starting
	 (start-time-old      (> (- (current-seconds) start-time) 5))
         (cleanup-proc        (lambda (msg)
                                (let* ((serv-fname      (conc "server-" (current-process-id) "-" (get-host-name) ".log"))
                                       (full-serv-fname (conc *toppath* "/logs/" serv-fname))
                                       (new-serv-fname  (conc *toppath* "/logs/" "defunct-" serv-fname)))
                                  (debug:print 0 *default-log-port* msg)
                                  (if (common:file-exists? full-serv-fname)
                                      (system (conc "sleep 1;mv -f " full-serv-fname " " new-serv-fname))
                                      (debug:print 0 *default-log-port* "INFO: cannot move " full-serv-fname " to " new-serv-fname))
                                  (exit)))))
    #;(if (and (not start-time-old) ;; last server start try was less than five seconds ago
	     (not server-starting))
	(begin
	  (cleanup-proc "NOT starting server, there is either a recently started server or a server in process of starting")
	  (exit)))
    ;; lets not even bother to start if there are already three or more server files ready to go
    #;(let* ((num-alive   (server:get-num-alive (server:get-list *toppath*))))
      (if (> num-alive 3)
          (begin
            (cleanup-proc (conc "ERROR: Aborting server start because there are already " num-alive " possible servers either running or starting up"))
            (exit))))
  (common:save-pkt `((action . start)
		     (T      . server)
		     (pid    . ,(current-process-id)))
		   *configdat* #t)
    (let* ((th2 (make-thread (lambda ()
                               (debug:print-info 0 *default-log-port* "Server run thread started")
                               (http-transport:run 
                                (if (args:get-arg "-server")
                                    (args:get-arg "-server")
                                    "-")
                                )) "Server run"))
           (th3 (make-thread (lambda ()
                               (debug:print-info 0 *default-log-port* "Server monitor thread started")
                               (http-transport:keep-running)
                               "Keep running"))))
      (thread-start! th2)
      (thread-sleep! 0.25) ;; give the server time to settle before starting the keep-running monitor.
      (thread-start! th3)
      (set! *didsomething* #t)
      (thread-join! th2)
      (exit))))

;; (define (http-transport:server-signal-handler signum)
;;   (signal-mask! signum)
;;   (handle-exceptions
;;    exn
;;    (debug:print 0 *default-log-port* " ... exiting ...")
;;    (let ((th1 (make-thread (lambda ()
;; 			     (thread-sleep! 1))
;; 			   "eat response"))
;; 	 (th2 (make-thread (lambda ()
;; 			     (debug:print-error 0 *default-log-port* "Received ^C, attempting clean exit. Please be patient and wait a few seconds before hitting ^C again.")
;; 			     (thread-sleep! 3) ;; give the flush three seconds to do it's stuff
;; 			     (debug:print 0 *default-log-port* "       Done.")
;; 			     (exit 4))
;; 			   "exit on ^C timer")))
;;      (thread-start! th2)
;;      (thread-start! th1)
;;      (thread-join! th2))))

;;======================================================================
;; faux-lock is deprecated. Please use simple-lock below
;;======================================================================
;;======================================================================
;;
(define (common:faux-lock keyname #!key (wait-time 8)(allow-lock-steal #t))
  (if (rmt:no-sync-get/default keyname #f) ;; do not be tempted to compare to pid. locking is a one-shot action, if already locked for this pid it doesn't actually count
      (if (> wait-time 0)
	  (begin
	    (thread-sleep! 1)
	    (if (eq? wait-time 1) ;; only one second left, steal the lock
		(begin
		  (debug:print-info 0 *default-log-port* "stealing lock for " keyname)
		  (common:faux-unlock keyname force: #t)))
	    (common:faux-lock keyname wait-time: (- wait-time 1)))
	  #f)
      (begin
        (rmt:no-sync-set keyname (conc (current-process-id)))
        (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))))

(define (common:faux-unlock keyname #!key (force #f))
  (if (or force (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))
      (begin
        (if (rmt:no-sync-get/default keyname #f) (rmt:no-sync-del! keyname))
        #t)
      #f))
;;======================================================================

;;======================================================================
;; simple lock. improve and converge on this one.
;;
(define (common:simple-lock keyname)
  (rmt:no-sync-get-lock keyname))

(define (common:simple-unlock keyname #!key (force #f))
  (rmt:no-sync-del! keyname))

;;======================================================================
;; ideally put all this info into the db, no need to preserve it across moving homehost
;;
;; return list of
;;  ( reachable? cpuload update-time )
(define (common:get-host-info hostname)
  (let* ((loadinfo                         (rmt:get-latest-host-load hostname)) ;; if this host happens to have been recently used by a test reuse the load data
         (load                             (car loadinfo))
         (load-sample-time                 (cdr loadinfo))
         (load-sample-age                  (- (current-seconds) load-sample-time))
         (loadinfo-timeout-seconds         6) ;; this was 20 seconds, seems way too lax. Switch to 6 seconds
         (host-last-update-timeout-seconds 4)
         (host-rec (hash-table-ref/default *host-loads* hostname #f))
         )
    (cond
     ((< load-sample-age loadinfo-timeout-seconds)
      (list #t
            load-sample-time
            load))
     ((and host-rec
           (< (current-seconds) (+ (host-last-update host-rec) host-last-update-timeout-seconds)))
      (list #t
            (host-last-update host-rec)
            (host-last-cpuload host-rec )))
     ((common:unix-ping hostname)
      (list #t
            (current-seconds)
            (alist-ref 'adj-core-load (common:get-normalized-cpu-load hostname)))) ;; this is cheaper than you might think. get-normalized-cpu-load is cached for up to 5 seconds
     (else
      (list #f 0 -1) ;; bad host, don't use!
      ))))

(define (std-exit-procedure)
  ;;(common:telemetry-log-close)
  (on-exit (lambda () 0))
  ;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
  (let ((no-hurry  (if *time-to-exit* ;; hurry up
		       #f
		       (begin
			 (set! *time-to-exit* #t)
			 #t))))
    (debug:print-info 4 *default-log-port* "starting exit process, finalizing databases.")
    (if (and no-hurry (debug:debug-mode 18))
	(rmt:print-db-stats))
    (let ((th1 (make-thread (lambda () ;; thread for cleaning up, give it five seconds
                              (if *dbstruct-db* (db:close-all *dbstruct-db*)) ;; one second allocated
			      (if *task-db*    
				  (let ((db (cdr *task-db*)))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! *task-db* 0 #f)
					  (set! *task-db* #f)))))
                              (http-client#close-all-connections!)
                              ;; (if (and *runremote*
                              ;;          (remote-conndat *runremote*))
                              ;;     (begin
                              ;;       (http-client#close-all-connections!))) ;; for http-client
                              (if (not (eq? *default-log-port* (current-error-port)))
                                  (close-output-port *default-log-port*))
			      (set! *default-log-port* (current-error-port))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 *default-log-port* "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
                                  (begin
                                    (thread-sleep! 5)) ;; give the clean up few seconds to do it's stuff
                                  (begin
      				  (thread-sleep! 2)))
      			      (debug:print 4 *default-log-port* " ... done")
      			      )
			    "clean exit")))
      (thread-start! th1)
      (thread-start! th2)
      (thread-join! th1)
      )
    )

  0)

;;======================================================================
;; from metadat lookup MEGATEST_VERSION
;;
(define (common:get-last-run-version) ;; RADT => How does this work in send-receive function??; assume it is the value saved in some DB
  (rmt:get-var "MEGATEST_VERSION"))

(define (common:set-last-run-version)
  (rmt:set-var "MEGATEST_VERSION" (common:version-signature)))

(define (common:get-last-run-version-number)
  (string->number 
   (substring (common:get-last-run-version) 0 6)))

;;======================================================================
;; see defstruct host at top of file.
;;    host: reachable last-update last-used last-cpuload
;;
(define (common:update-host-loads-table hosts-raw)
  (let* ((hosts (filter (lambda (x)
                          (string-match (regexp "^\\S+$") x))
                        hosts-raw)))
    (for-each
     (lambda (hostname)
       (let* ((rec       (let ((h (hash-table-ref/default *host-loads* hostname #f)))
                          (if h
                              h
                              (let ((h (make-host)))
                                (hash-table-set! *host-loads* hostname h)
                                h))))
              (host-info         (common:get-host-info hostname))
              (is-reachable      (car host-info))
              (last-reached-time (cadr host-info))
              (load              (caddr host-info)))
         (host-reachable-set!    rec is-reachable)
         (host-last-update-set!  rec last-reached-time)
         (host-last-cpuload-set! rec load)))
     hosts)))

;;======================================================================
;; go through the hosts from least recently used to most recently used, pick the first that meets the load criteral from the
;; [host-rules] section.
;;
(define (common:get-least-loaded-host hosts-raw host-type configdat)
  (let* ((rdat       (configf:lookup configdat "host-rules" host-type))
	 (rules      (common:val->alist (or rdat "") convert: #t))   ;; maxnload, maxnjobs, maxjobrate
	 (maxnload   (common:alist-ref/default 'maxnload rules 1.5)) ;; max normalized load
	 (maxnjobs   (common:alist-ref/default 'maxnjobs rules 1.5)) ;; max normalized number of jobs
	 (maxjobrate (common:alist-ref/default 'maxjobrate rules (/ 1 6))) ;; max rate of submitting jobs to a given host in jobs/second
	 (hosts      (filter (lambda (x)
			       (string-match (regexp "^\\S+$") x))
			     hosts-raw))
         ;; (best-host #f)
	 (get-rec    (lambda (hostname)
		       ;; (print "get-rec hostname=" hostname)
		       (let ((h (hash-table-ref/default *host-loads* hostname #f)))
			 (if h
			     h
			     (let ((h (make-host)))
			       (hash-table-set! *host-loads* hostname h)
			       h)))))
         (best-load 99999)
         (curr-time (current-seconds))
	 (get-hosts-sorted (lambda (hosts)
			     (sort hosts (lambda (a b)
					   (let ((a-rec (get-rec a))
						 (b-rec (get-rec b)))
					     ;; (print "a=" a " a-rec=" a-rec " host-last-used=" (host-last-used a-rec))
					     ;; (print "b=" b " b-rec=" b-rec " host-last-used=" (host-last-used b-rec))
					     (< (host-last-used a-rec)
						(host-last-used b-rec))))))))
    (debug:print 0 *default-log-port* "INFO: hosts-sorted=" (get-hosts-sorted hosts))
    (if (null? hosts)
	#f ;; no hosts to select from. All done and giving up now.
	(let ((hosts-sorted (get-hosts-sorted hosts)))
	  (common:update-host-loads-table hosts)
	  (let loop ((hostname  (car hosts-sorted))
		     (tal       (cdr hosts-sorted))
		     (best-host #f))
	    (let* ((rec       (get-rec hostname))
		   (reachable (host-reachable     rec))
		   (load      (host-last-cpuload  rec))
		   (last-used (host-last-used     rec))
		   (delta     (- curr-time last-used))
		   (job-rate  (if (> delta 0)
				  (/ 1 delta)
				  999)) ;; jobs per second
		   (new-best  
		    (cond
		     ((not reachable)
		      (debug:print 0 *default-log-port* "Skipping host " hostname " as it cannot be reached.")
		      best-host)
		     ((and (< load maxnload)        ;; load is acceptable
			   (< job-rate maxjobrate)) ;; job rate is acceptable
		      (set! best-load load)
		      hostname)
		     (else best-host))))
	      (debug:print 0 *default-log-port* "INFO: Trying host " hostname " with load " load ", last used " delta " seconds ago, with job-rate " job-rate " for running a test." )
	      (if new-best
		  (begin ;; found a host, return it
		    (debug:print 0 *default-log-port* "INFO: Found host: " new-best " load: " load " last-used: " delta " seconds ago, with job-rate: " job-rate)
		    (host-last-used-set! rec curr-time)
		    new-best)
		  (if (null? tal) #f (loop (car tal)(cdr tal) best-host)))))))))

;;======================================================================
;;  T E S T   L A U N C H I N G   P E R   I T E M   W I T H   H O S T   T Y P E S
;;======================================================================
;; 
;; [hosts]
;; arm cubie01 cubie02
;; x86_64 zeus xena myth01
;; allhosts #{g hosts arm} #{g hosts x86_64}
;; 
;; [host-types]
;; general #MTLOWESTLOAD #{g hosts allhosts}
;; arm     #MTLOWESTLOAD #{g hosts arm}
;; nbgeneral nbjob run JOBCOMMAND -log $MT_LINKTREE/$MT_TARGET/$MT_RUNNAME.$MT_TESTNAME-$MT_ITEM_PATH.lgo
;;
;; [host-rules]
;; # maxnload   => max normalized load
;; # maxnjobs   => max jobs per cpu
;; # maxjobrate => max jobs per second
;; general maxnload=1.1; maxnjobs=1.2; maxjobrate=0.1 
;; 
;; [launchers]
;; envsetup general
;; xor/%/n 4C16G
;; % nbgeneral
;; 
;; [jobtools]
;; # if defined and not "no" flexi-launcher will bypass "launcher" unless no match.
;; flexi-launcher yes  
;; launcher nbfake
;;
(define (common:get-launcher configdat testname itempath)
  (let ((fallback-launcher (configf:lookup configdat "jobtools" "launcher")))
    (if (and (configf:lookup configdat "jobtools" "flexi-launcher") ;; overrides launcher
	     (not (equal? (configf:lookup configdat "jobtools" "flexi-launcher") "no")))
	(let* ((launchers         (hash-table-ref/default configdat "launchers" '())))
	  (if (null? launchers)
	      fallback-launcher
	      (let loop ((hed (car launchers))
			 (tal (cdr launchers)))
		(let ((patt      (car hed))
		      (host-type (cadr hed)))
		  (if (tests:match patt testname itempath)
		      (begin
			(debug:print-info 2 *default-log-port* "Have flexi-launcher match for " testname "/" itempath " = " host-type)
			(let ((launcher (configf:lookup configdat "host-types" host-type)))
			  (if launcher
			      (let* ((launcher-parts (string-split launcher))
				     (launcher-exe   (car launcher-parts)))
				(if (equal? launcher-exe "#MTLOWESTLOAD") ;; this is our special case, we will find the lowest load and craft a nbfake commandline
				    (let host-loop ((targ-host (common:get-least-loaded-host (cdr launcher-parts) host-type configdat))
						    (count     100))
				      (if targ-host
					  (conc "remrun " targ-host)
					  (if (> count 0)
					      (begin
						(debug:print 0 *default-log-port* "INFO: Waiting for a host for host-type " host-type)
						(thread-sleep! (- 101 count))
						(host-loop (common:get-least-loaded-host (cdr launcher-parts) host-type configdat)
							   (- count 1)))
					      (begin
						(debug:print 0 *default-log-port* "FATAL: Failed to find a host from #MTLOWESTLOAD for host-type " host-type)
						(exit)))))
				    launcher))
			      (begin
				(debug:print-info 0 *default-log-port* "WARNING: no launcher found for host-type " host-type)
				(if (null? tal)
				    fallback-launcher
				    (loop (car tal)(cdr tal)))))))
		      ;; no match, try again
		      (if (null? tal)
			  fallback-launcher
			  (loop (car tal)(cdr tal))))))))
	fallback-launcher)))



)