Megatest

           diff-report.scm cgisetup/models/pgdb.scm

# module source files
MSRCFILES = dbfile.scm debugprint.scm mtargs.scm commonmod.scm dbmod.scm \
            tcp-transportmod.scm rmtmod.scm portlogger.scm

transport-mode.scm : transport-mode.scm.template
	@if [[ -e transport-mode.scm ]];then \
	echo "WARNING: transport-mode.scm.template is newer than transport-mode.scm"; else \
	cp transport-mode.scm.template transport-mode.scm; fi
	cp transport-mode.scm.template transport-mode.scm

dashboard-transport-mode.scm : dashboard-transport-mode.scm.template
	@if [[ -e dashboard-transport-mode.scm ]];then \
	echo "WARNING: dashboard-transport-mode.scm.template is newer than dashboard-transport-mode.scm"; else \
	cp dashboard-transport-mode.scm.template dashboard-transport-mode.scm; fi
	cp dashboard-transport-mode.scm.template dashboard-transport-mode.scm

megatest.scm : transport-mode.scm
dashboard.scm : dashboard-transport-mode.scm
mtest : transport-mode.scm
dboard : dashboard-transport-mode.scm

# dbmod.import.o is just a hack here
mofiles/portlogger.o : mofiles/dbmod.o

mofiles/dbfile.o     : \
       mofiles/debugprint.o mofiles/commonmod.o
       mofiles/debugprint.o mofiles/commonmod.o 

mofiles/dbmod.o : mofiles/dbfile.o

mofiles/commonmod.o : mofiles/debugprint.o
configf.o : commonmod.import.o
mofiles/dbfile.o : mofiles/debugprint.o
mofiles/rmtmod.o mofiles/dbmod.o : mofiles/dbfile.o mofiles/commonmod.o mofiles/debugprint.o
db.o : mofiles/dbmod.o mofiles/dbfile.o
mofiles/debugprint.o : mofiles/mtargs.o
mofiles/tcp-transportmod.o : mofiles/portlogger.o

	$(INSTALL) $< $@
	chmod a+x $@

$(PREFIX)/bin/mt-new-to-old.sh : utils/mt-new-to-old.sh
	$(INSTALL) $< $@
	chmod a+x $@

$(PREFIX)/bin/convert-db.sh : utils/convert-db.sh
	$(INSTALL) $< $@
	chmod a+x $@

deploytarg/nbfake : utils/nbfake
	$(INSTALL) $< $@
	chmod a+x $@

deploytarg/viewscreen : utils/viewscreen
	$(INSTALL) $< $@

	$(INSTALL) lib/libxcb-xlib.so.0 $(PREFIX)/bin/.$(ARCHSTR)/lib/libxcb-xlib.so.0; \
        fi

install : $(PREFIX)/bin/.$(ARCHSTR) $(PREFIX)/bin/.$(ARCHSTR)/mtest $(PREFIX)/bin/megatest \
          $(PREFIX)/bin/.$(ARCHSTR)/dboard $(PREFIX)/bin/dashboard $(HELPERS) $(PREFIX)/bin/nbfake \
	  $(PREFIX)/bin/nbfind $(PREFIX)/bin/mtrunner $(PREFIX)/bin/viewscreen $(PREFIX)/bin/mt_xterm \
          $(PREFIX)/bin/mt-old-to-new.sh $(PREFIX)/bin/mt-new-to-old.sh \
          $(PREFIX)/bin/convert-db.sh $(PREFIX)/bin/convert-db.sh \
	  $(PREFIX)/share/docs/megatest_manual.html $(PREFIX)/bin/remrun \
	  $(PREFIX)/share/docs/megatest_manual.html $(PREFIX)/bin/remrun $(PREFIX)/bin/mtutil \
	  $(PREFIX)/share/db/mt-pg.sql \
          $(PREFIX)/share/js/jquery-3.1.0.slim.min.js \
          $(PREFIX)/bin/.$(ARCHSTR)/lib/libpangox-1.0.so \
          $(PREFIX)/bin/.$(ARCHSTR)/lib/libpangox-1.0.so.0 \
          $(PREFIX)/bin/.$(ARCHSTR)/lib/libxcb-xlib.so.0 $(PREFIX)/bin/serialize-env

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

TODO
====

23WW48
. Add calls-per-minute to db access stats
. Find out why start-server calls are taking 250ms and fix
. Allow two or three servers to run for any given db
. Update avg call count/sec every 30 sec in no-sync
. get server uses no-sync process info to decide which server to suggest
. Use process table to decide who will do sync back
. Fix metadat being synced over and over

23WW47
. Finding server
.. look at .servinfo for likely prime main
.. ask the .servinfo prime main for real prime main
.. save prime main (for how long, 10 seconds or 10 minutes?)

. Starting prime main
.. get servinfo files - START
.. no files? create my servinfo file, goto START
.. have files? am I the prime main according to servinfo files?
.. no, I'm not the prime main, ping prime main
.. ping is good, prime main exists, register self as server if on same host as prime main DONE
.. no pirng response, remove the .servinfo file - goto START
.. if I am prime main according to .servinfo files, register directly in no-sync

. Starting non-main
.. get servinfo files
.. no files? launch server for main.db
.. have files? pick out prime main
.. register self as server with prime main

23WW46 - v1.80 branch
. Use file semaphore to kill tests, eliminate db load of the KILLREQ query
. Merge this change to revolution branch
23WW45 - the revolution branch
. Add "fast" db start option (no handshaking over NFS)
. Add server-ro to server types (just "server" is fine for read/write).
. [DONE] Create pause-server and resume-server calls
. Create rsync or cp sync to MTRAH function
. Change rmt:send-receive to divert calls to read-only server when possible
. [DONE] Change start server to call main.db server for 1..N.db servers, block until server is read for use.

23WW21
. Dashboard needs its own cache db in /tmp

23WW07
. Remove use of *dbstruct-dbs*

WW15

(import debugprint)
(import tcp-transportmod)

(use srfi-69
     srfi-18
     posix
     matchable
     s11n)
     s11n
     typed-records)

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

    ;; TASKS
    tasks-add
    tasks-set-state-given-param-key
    ))

(define *db-write-mutexes* (make-hash-table))
(define *server-signature* #f)
;; ;; These are called by the server on recipt of /api calls
;; ;;    - keep it simple, only return the actual result of the call, i.e. no meta info here
;; ;;
;; ;;    - returns #( flag result )
;; ;;
;; (define (api:execute-requests dbstruct dat)
;;   (if (> *api-process-request-count* 50)
;;       (begin
;; 	(if (common:low-noise-print 30 "too many threads")
;; 	    (debug:print 0 *default-log-port* "WARNING: "*api-process-request-count*" threads, potential overload, adding 0.5 sec delay."))
;; 	;; (thread-sleep! 0.5) ;; take a nap - no, the napping is moved to the clients via tt:backoff-incr
;; 	))
;;   (cond
;;    ((not (vector? dat))                    ;; it is an error to not receive a vector
;;     (vector #f (vector #f "remote must be called with a vector")))
;;    (else  
;;     (let* ((cmd-in            (vector-ref dat 0))
;;            (cmd               (if (symbol? cmd-in)
;; 				  cmd-in
;; 				  (string->symbol cmd-in)))
;;            (params            (vector-ref dat 1))
;; 	   (run-id            (if (null? params)
;; 				  0
;; 				  (car params)))
;; 	   (write-mutex       (if (hash-table-exists? *db-write-mutexes* run-id)
;; 				  (hash-table-ref *db-write-mutexes* run-id)
;; 				  (let* ((newmutex (make-mutex)))
;; 				    (hash-table-set! *db-write-mutexes* run-id newmutex)
;; 				    newmutex)))
;;            (start-t           (current-milliseconds))
;;            (readonly-mode     (dbr:dbstruct-read-only dbstruct))
;;            (readonly-command  (member cmd api:read-only-queries))
;;            (writecmd-in-readonly-mode (and readonly-mode (not readonly-command))))
;;       (if (not readonly-command)
;; 	  (mutex-lock! write-mutex))
;;       (let* ((tmppath    (dbr:dbstruct-tmppath  dbstruct))
;; 	     (clean-run-id (cond
;; 			    ((number? run-id)   run-id)
;; 			    ((equal? run-id #f) "main")
;; 			    (else               "other")))
;; 	     (crumbfile  (dbfile:wait-for-qif tmppath clean-run-id (cons cmd params)))
;; 	     (res    
;;               (if writecmd-in-readonly-mode
;;                   (conc "attempt to run write command "cmd" on a read-only database")
;; 		  (api:dispatch-request dbstruct cmd run-id params))))
;; 	(delete-file* crumbfile)
;; 	(if (not readonly-command)
;; 	    (mutex-unlock! write-mutex))
;; 	
;; 	;; save all stats
;; 	(let ((delta-t (- (current-milliseconds)
;; 			  start-t))
;; 	      (modified-cmd (if (eq? cmd 'general-call)
;; 				(string->symbol (conc "general-call-" (car params)))
;; 				cmd)))
;; 	  (hash-table-set! *db-api-call-time* modified-cmd
;; 			   (cons delta-t (hash-table-ref/default *db-api-call-time* modified-cmd '()))))
;; 	(if writecmd-in-readonly-mode
;;             (begin
;;               #;(common:telemetry-log (conc "api-out:"(->string cmd))
;;               payload: `((params . ,params)
;;               (ok-res . #t)))
;; 	      (vector #f res))
;;             (begin
;;               #;(common:telemetry-log (conc "api-out:"(->string cmd))
;;               payload: `((params . ,params)
;;               (ok-res . #f)))
;;               (vector #t res))))))))

(define *api-threads* '())
(define (api:register-thread th-in)
  (set! *api-threads* (cons (cons th-in (current-seconds)) *api-threads*)))

(define (api:unregister-thread th-in)
  (set! *api-threads* (filter (lambda (thdat)
				(not (eq? th-in (car thdat))))
			      *api-threads*)))

(define (api:remove-dead-or-terminated)
  (set! *api-threads* (filter (lambda (thdat)
				(not (member (thread-state (car thdat)) '(terminated dead))))
			      *api-threads*)))

(define (api:get-count-threads-alive)
  (length *api-threads*))
  

(define *api:last-stats-print* 0)
(define *api-print-db-stats-mutex* (make-mutex))
(define (api:print-db-stats)
  (debug:print-info 0 *default-log-port* "Started periodic db stats printer")
  (let loop ()
    (mutex-lock! *api-print-db-stats-mutex*)
    (if (> (- (current-seconds) *api:last-stats-print*) 15)
	(begin
	  (rmt:print-db-stats)
	  (set! *api:last-stats-print* (current-seconds))))
    (mutex-unlock! *api-print-db-stats-mutex*)
    (thread-sleep! 5)
    (loop)))

;; ==========================================================================================================================================
;; api:tcp-dispatch-request-make-handler is a complex TCP request handler that manages server load, dispatches requests, and ensures that the 
;; server's state is consistent with the incoming requests. It is designed to be used in a concurrent environment where multiple requests are 
;; being handled simultaneously.
;;
;; It processes incoming requests and dispatches them accordingly. 
;; The function takes a dbstruct argument, which is a structure representing the database.

(define (api:tcp-dispatch-request-make-handler dbstruct) ;; cmd run-id params)
  (assert *toppath* "FATAL: api:tcp-dispatch-request-make-handler called but *toppath* not set.")
  (if (not *server-signature*)
      (set! *server-signature* (tt:mk-signature *toppath*)))
  (lambda (indat)
    (api:register-thread (current-thread))
    (let* (;; (indat      (deserialize))
    (let* ((result 
	   (newcount   (+ *api-process-request-count* 1))
	   (numthreads (api:get-count-threads-alive))
	   (delay-wait (if (> newcount 10)
			   (- newcount 10)
			   0))
	   (normal-proc (lambda (cmd run-id params)
			  (case cmd
			    ((ping) *server-signature*)
			    (else
			     (api:dispatch-request dbstruct cmd run-id params))))))
      (set! *api-process-request-count* newcount)
      (set! *db-last-access* (current-seconds))
      (if (not (eq? newcount numthreads))
	  (begin
	    (api:remove-dead-or-terminated)
	    (let ((threads-now (api:get-count-threads-alive)))
	      (debug:print 0 *default-log-port* "WARNING: newcount="newcount", numthreads="numthreads", remaining="threads-now)
	      (set! newcount threads-now))))
      (match indat
	((cmd run-id params meta)
	 (let* ((db-ok  (let* ((dbfname (dbmod:run-id->dbfname run-id))
			       (ok      (equal? dbfname (dbr:dbstruct-dbfname dbstruct))))
			  (case cmd
			    ((ping) #t) ;; we are fine
			    (else
	    (let* ((numthreads (api:get-count-threads-alive))
		   (delay-wait (if (> numthreads 10)
				   (- numthreads 10)
				   0))
		   (normal-proc (lambda (cmd run-id params)
				  (case cmd
				    ((ping) *server-signature*)
				    (else
				     (api:dispatch-request dbstruct cmd run-id params))))))
	      (set! *api-process-request-count* numthreads)
	      (set! *db-last-access* (current-seconds))
;; 	      (if (not (eq? numthreads numthreads))
;; 	      (begin
;; 	      (api:remove-dead-or-terminated)
;; 	      (let ((threads-now (api:get-count-threads-alive)))
;; 	      (debug:print 0 *default-log-port* "WARNING: numthreads="numthreads", numthreads="numthreads", remaining="threads-now)
;; 	      (set! numthreads threads-now))))
	      (match indat
		     ((cmd run-id params meta)
		      (let* ((start-t (current-milliseconds))
			     (db-ok  (let* ((dbfname (dbmod:run-id->dbfname run-id))
					    (ok      (equal? dbfname (dbr:dbstruct-dbfname dbstruct))))
				       (case cmd
					 ((ping) #t) ;; we are fine
					 (else
			     (if (not ok)(debug:print 0 *default-log-port* "ERROR: "cmd", run-id "run-id", not correct for dbfname "(dbr:dbstruct-dbfname dbstruct)))
			     (assert ok "FATAL: database file and run-id not aligned.")))))
		(ttdat   *server-info*)
		(server-state (tt-state ttdat))
		(status  (cond
			  ((> newcount 3) 'busy)
			  ;; ((> newcount 5) 'loaded) ;; this gets transmitted to the client which calls tt:backoff-incr to slow stuff down.
			  (else 'ok)))
		(errmsg  (case status
			   ((busy)   (conc "Server overloaded, "newcount" threads in flight"))
			   ((loaded) (conc "Server loaded, "newcount" threads in flight"))
			   (else     #f)))
		(result  (case status
			   ((busy)
			    (if (eq? cmd 'ping)
				(normal-proc cmd run-id params)
				;; newcount must be greater than 5 for busy
				(* 1 (- newcount 3)) ;; was 15
				)) ;; (- newcount 29)) ;; call back in as many seconds
			   ((loaded)
;; 			    (if (eq? (rmt:transport-mode) 'tcp)
;; 				(thread-sleep! 0.5))
			    (normal-proc cmd run-id params))
			   (else
			    (normal-proc cmd run-id params))))
		(meta   (case cmd
			  ((ping) `((sstate . ,server-state)))
			  (else   `((wait . ,delay-wait)))))
		(payload (list status errmsg result meta)))
	   (set! *api-process-request-count* (- *api-process-request-count* 1))
	   ;; (serialize payload)
	   (api:unregister-thread (current-thread))
	   payload))
	(else
	 (assert #f "FATAL: failed to deserialize indat "indat))))))
       
					  (assert ok "FATAL: database file and run-id not aligned.")))))
			     (ttdat   *server-info*)
			     (server-state (tt-state ttdat))
			     (maxthreads   20) ;; make this a parameter?
			     (status  (cond
				       ((and (> numthreads maxthreads)
					     (> (random 100) 70)) ;; allow a 30% probability to go through so we can figure out what is going wrong in main.db server.
					'busy)
				       ;; ((> numthreads 5) 'loaded) ;; this gets transmitted to the client which calls tt:backoff-incr to slow stuff down.
				       (else 'ok)))
			     (errmsg  (case status
					((busy)   (conc "Server overloaded, "numthreads" threads in flight"))
					((loaded) (conc "Server loaded, "numthreads" threads in flight"))
					(else     #f)))
			     (result  (case status
					((busy)
					 (if (eq? cmd 'ping)
					     (normal-proc cmd run-id params)
					     ;; numthreads must be greater than 5 for busy
					     (* 0.1 (- numthreads maxthreads)) ;; was 15
					     )) ;; (- numthreads 29)) ;; call back in as many seconds
					((loaded)
					 ;; 			    (if (eq? (rmt:transport-mode) 'tcp)
					 ;; 				(thread-sleep! 0.5))
					 (normal-proc cmd run-id params))
					(else
					 (normal-proc cmd run-id params))))
			     (meta   (case cmd
				       ((ping) `((sstate . ,server-state)))
				       (else   `((wait . ,delay-wait)))))
			     (payload (list status errmsg result meta)))
			;; (cmd run-id params meta)
			(db:add-stats cmd run-id params (- (current-milliseconds) start-t))
			payload))
		     (else
		      (assert #f "FATAL: failed to deserialize indat "indat))))))
      ;; (set! *api-process-request-count* (- *api-process-request-count* 1))
      ;; (serialize payload)
     
      (api:unregister-thread (current-thread))
      result)))



(define *api-halt-writes* #f)

(define (api:dispatch-request dbstruct cmd run-id params)
  (if (not *no-sync-db*)
      (db:open-no-sync-db))
  (let* ((start-time (current-milliseconds)))
    (if (member cmd api:write-queries)
	(let loop ()
	  (if *api-halt-writes*
	      (begin
		(thread-sleep! 0.2)
		(if (< (- (current-milliseconds) start-time)
		       5000) ;; hope it don't take more than five seconds to sync
		    (loop-time)
		    #;(debug:print 0 *default-log-port* "ERROR: writes halted for more than 5 seconds, sync might be taking too long"))))))
    (db:add-stats 'api-write-blocking-for-sync run-id params (- (current-milliseconds) start-time)))
  (case cmd
    ;;===============================================
    ;; 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))
    ((start-server)                    (apply tt:server-process-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)

    ;; TASKS 
    ((find-task-queue-records)   (apply tasks:find-task-queue-records dbstruct params))
    (else
     (debug:print 0 *default-log-port* "ERROR: bad api call " cmd)
     (conc "ERROR: BAD api call " cmd))))

;; http-server  send-response
;;                 api:process-request
;;                    db:*
;;
;; NB// Runs on the server as part of the server loop
;;
(define (api:process-request dbstruct $) ;; the $ is the request vars proc
  (debug:print 4 *default-log-port* "server-id:"  *server-id*)
  (let* ((cmd     ($ 'cmd))
	 (paramsj ($ 'params))
         (key     ($ 'key))   
	 (params  (db:string->obj paramsj transport: 'http))) ;; incoming data from the POST (or is it a GET?)
    (debug:print 4 *default-log-port* "cmd:" cmd " with params " params "key " key)
    (if (equal? key *server-id*)
      (begin
        (set! *api-process-request-count* (+ *api-process-request-count* 1))
 	(let* ((resdat  (api:execute-requests dbstruct (vector cmd params))) ;; process the request, resdat = #( flag result )
	       (success (vector-ref resdat 0))
	       (res     (vector-ref resdat 1))) ;; (vector flag payload), get the payload, ignore the flag (why?)
	  (debug:print 4 *default-log-port* "res:" res)
	  (if (not success)
	      (debug:print 0 *default-log-port* "ERROR: success flag is #f for " cmd " with params " params))
	  (if (> *api-process-request-count* *max-api-process-requests*)
	      (set! *max-api-process-requests* *api-process-request-count*))
	  (set! *api-process-request-count* (- *api-process-request-count* 1))
	  ;; This can be here but needs controls to ensure it doesn't run more than every 4 seconds
	  ;; (rmt:dat->json-str
	  ;;  (if (or (string? res)
	  ;;          (list?   res)
	  ;;          (number? res)
	  ;;          (boolean? res))
	  ;;      res 
	  ;;      (list "ERROR, not string, list, number or boolean" 1 cmd params res)))))
	  (db:obj->string res transport: 'http)))
	(begin
	  (debug:print 0 *default-log-port*   "Server refused to process request. Server id mismatch. recived " key " expected:  " *server-id* ".\nOther arguments recived: cmd=" cmd " params = " params) 
	  (db:obj->string (conc "Server refused to process request server-id mismatch: " key ", " *server-id*) transport: 'http)))))

   (let* ((bup-exe               (or (configf:lookup *configdat* "archive" "bup") "bup"))
         (archive-internal-path (conc (common:get-testsuite-name) "-megatest-db/" ts "/megatest.db" ))
         (bup-restore-params  (list "-d" archive-path "restore" "-C" *toppath* archive-internal-path)))
		 (debug:print-info 2 *default-log-port* "Restoring archived data to " *toppath* " from archive in " archive-path " ... " archive-internal-path)
		 (run-n-wait bup-exe params: bup-restore-params print-cmd: "Running:"))
      (sleep 2)
      (db:multi-db-sync 
       (db:setup #t) ;; (db:setup-db *dbstruct-dbs* *toppath* #f)
       (db:setup) ;; (db:setup-db *dbstruct-dbs* *toppath* #f)
       'killservers
       ;'dejunk
       ;'adj-testids
       'old2new
       )
      (debug:print-info 1 *default-log-port* "dropping triggers to update linktree") 
      (rmt:drop-all-triggers)

(use posix-extras pathname-expand files)


(import commonmod
	debugprint
	rmtmod
	(prefix mtargs args:))

	
(define (remove-server-files directory-path)
  (let ((files (glob (string-append directory-path "/server*"))))
    (for-each delete-file* files)))
(include "common_records.scm")

(define (remove-files filespec)
  (let ((files (glob filespec)))
    (for-each delete-file* files)))

(define (stop-the-train)
  (thread-start! (make-thread (lambda ()
				(let loop ()
				  (if (and *toppath*
					   (file-exists? (conc *toppath*"/stop-the-train")))
				      (let* ((msg (conc "ERROR: found file "*toppath*"/stop-the-train, exiting immediately")))
					;; yes, print to current-output-port AND *default-log-port*, annoying but necessary I think
					(print msg)
                                        (remove-server-files (conc *toppath* "/logs"))
					(debug:print 0 *default-log-port* msg)
                                        (remove-files (conc *toppath* "/logs/server*"))
                                        (remove-files (conc *toppath* "/.servinfo/*"))
                                        (remove-files (conc *toppath* "/.mtdb/*lock"))
					(exit 1)))
				  (thread-sleep! 5)
				  (loop))))))

;; (define *common:denoise*    (make-hash-table)) ;; for low noise printing
(define *default-log-port*  (current-error-port))
(define *time-zero* (current-seconds)) ;; for the watchdog
(define *on-exit-procs* '()) ;; add procs to this list to be executed on exit
(define *default-area-tag* "local")

;; DATABASE
;; (define *dbstruct-dbs*         #f) ;; used to cache the dbstruct in db:setup. Goal is to remove this.
;; db stats
(define *db-stats*            (make-hash-table)) ;; hash of vectors < count duration-total >
(define *db-stats-mutex*      (make-mutex))
;; db access
(define *db-last-access*      (current-seconds)) ;; last db access, used in server
;; (define *db-write-access*     #t)
;; db sync
;; (define *db-last-sync*        0)                 ;; last time the sync to megatest.db happened
(define *db-sync-in-progress* #f)                ;; if there is a sync in progress do not try to start another
;; (define *db-multi-sync-mutex* (make-mutex))      ;; protect access to *db-sync-in-progress*, *db-last-sync*

(define *transport-type*    'http)             ;; override with [server] transport http|rpc|nmsg
(define *runremote*         #f)                ;; if set up for server communication this will hold <host port>
;; (define *max-cache-size*    0)
(define *logged-in-clients* (make-hash-table))
(define *server-id*         #f)
(define *server-info*       #f)  ;; good candidate for easily convert to non-global
(define *time-to-exit*      #f)
(define *server-run*        #t)
(define *run-id*            #f)
(define *server-kind-run*   (make-hash-table))
(define *home-host*         #f)
;; (define *total-non-write-delay* 0)
(define *heartbeat-mutex*   (make-mutex))
;; (define *api-process-request-count* 0)
;; (define *max-api-process-requests* 0)

	    'schema
	    'killservers
	    'adj-target
	    'new2old
	    '(dejunk)
	    ))
    ((tcp nfs)
     (debug:print 0 *default-log-port* "WARNING: cleanup-db NOT implemented yet for tcp and nfs.")
     #;(apply db:multi-db-sync 
     (apply db:multi-db-sync 
	    dbstruct
	    'schema
	    'killservers
	    'adj-target
	    'new2old
	    '(dejunk)
	    )))

(define (common:exit-on-version-changed)
  (if (and *toppath*              ;; do nothing if *toppath* not yet provided
	   (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") ".mtdb/main.db"))
                 (read-only (not (file-write-access? dbfile)))
                 (dbstruct (db:setup #t))) ;; (db:setup-db *dbstruct-dbs* *toppath* #f))) ;;  #t)))
                 (dbstruct (db:setup))) ;; (db:setup-db *dbstruct-dbs* *toppath* #f))) ;;  #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)

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

(define (get-cpu-load #!key (remote-host #f))
  (car (common:get-cpu-load remote-host)))
;; (define (get-cpu-load #!key (remote-host #f))
;;    (car (common:get-cpu-load remote-host)))

;;======================================================================
;;   (let* ((load-res (process:cmd-run->list "uptime"))
;; 	 (load-rx  (regexp "load average:\\s+(\\d+)"))
;; 	 (cpu-load #f))
;;     (for-each (lambda (l)
;; 		(let ((match (string-search load-rx l)))
;; 		  (if match
;; 		      (let ((newval (string->number (cadr match))))
;; 			(if (number? newval)
;; 			    (set! cpu-load newval))))))
;; 	      (car load-res))
;;     cpu-load))

;;======================================================================
;; get values from cached info from dropping file in logs dir
;; get values from cached info from dropping file in .sysdata dir
;;  e.g. key is host and dtype is normalized-load
;;
(define (common:get-cached-info key dtype #!key (age 10))
  (if *toppath*
      (let* ((fullpath (conc *toppath* "/.sysdata/" key "-" dtype ".log"))
	     (delfile  (lambda (exn)
			 (debug:print-info 1 *default-log-port* " removing bad file " fullpath ", exn=" exn)
			 (debug:print-info 2 *default-log-port* " removing bad file " fullpath ", exn=" exn)
			 (delete-file* fullpath)
			 #f)))
	(if (and (file-exists? fullpath)
		 (file-read-access? fullpath))
	    (handle-exceptions
		exn
	      (begin

	  ;; posix-extras
	  chicken.base
	  chicken.condition
	  chicken.file
	  chicken.file.posix
	  chicken.io
	  chicken.pathname
	  chicken.port
	  chicken.process
	  chicken.process-context
	  chicken.process-context.posix
	  chicken.sort
	  chicken.string
	  chicken.time
	  chicken.time.posix
	  
	  matchable
	  md5
	  message-digest
	  pathname-expand
	  regex
	  regex-case
	  srfi-1
	  srfi-18
	  srfi-69
	  typed-records
	  system-information

	  debugprint
  )))

;;======================================================================
;; CONTENTS
;;
;;  config file utils
;;  misc conversion, data manipulation functions

			     (if convert (lazy-convert inval) inval))))
		   (else f))))
	     (filter (lambda (x)
		       (not (string-match "^\\s*" x)))
		     val-list))
	'())))

(define (get-cpu-load)
  (let* ((load-info (with-input-from-file "/proc/loadavg" read-lines)))
    (map string->number (string-split load-info))))
(define (commonmod:get-cpu-load)
  (let* ((load-info (with-input-from-file "/proc/loadavg" read-lines))
	 (res       (map string->number (string-split (car load-info)))))
    (if (null? res)
	#f ;; something is wrong
	(car res))))

(define *current-host-cores* #f)

(define (get-current-host-cores)
  (or *current-host-cores*
      (let ((cpu-info (with-input-from-file "/proc/cpuinfo" read-lines)))
	(let loop ((lines cpu-info))

   (conc "ps -def | egrep \""processname"\" |wc -l")
   (lambda ()
     (string->number (read-line)))))

;; get the normalized (i.e. load / numcpus) for *this* host
;;
(define (get-normalized-cpu-load)
  (/ (get-cpu-load)(get-current-host-cores)))
  (/ (commonmod:get-cpu-load)(get-current-host-cores)))

;;======================================================================
;; testsuite and area utilites
;;======================================================================

(define (get-testsuite-name toppath configdat)
  (or (lookup configdat "setup" "area-name")

							      ((m) 60) ;; minutes
							      ((h) 3600)
							      ((d) 86400)
							      ((w) 604800)
							      ((M) 2628000) ;; aproximately one month
							      ((y) 31536000)
							      (else
							       0)))))))
		      ;; (print "ERROR: can't parse timestring "tstr", component "part)
							       0)))))
			    (debug:print 0 *default-log-port* "ERROR: can't parse timestring "tstr", component "part", string: "(cadr match))))
		      (debug:print 0 *default-log-port* "ERROR: can't parse timestring "tstr", component "part))))
		      ;; can't (yet) use debugprint. rely on -show-config for user to find errors
		      )))
	      parts)
    time-secs))
		       
(define (seconds->hr-min-sec secs)
  (let* ((hrs (quotient secs 3600))
	 (min (quotient (- secs (* hrs 3600)) 60))
	 (sec (- secs (* hrs 3600)(* min 60))))

(declare (uses keys))
(declare (uses debugprint))
(declare (uses mtargs))
(declare (uses mtargs.import))
(declare (uses common))
(declare (uses commonmod))
(declare (uses commonmod.import))
(declare (uses dbfile))
(declare (uses dbfile.import))
(declare (uses dbmod))
(declare (uses dbmod.import))

(import commonmod
	(prefix mtargs args:)
	debugprint)

(include "common_records.scm")

;; return list (path fullpath configname)

;;======================================================================
;; Test info panel
;;======================================================================

(declare (unit dashboard-tests))
(declare (uses common))
(declare (uses commonmod))
(declare (uses dcommon))
(declare (uses db))
(declare (uses gutils))
(declare (uses rmt))
(declare (uses ezsteps))
(declare (uses subrun))
(declare (uses debugprint))
(declare (uses rmtmod))

;;======================================================================
;; set up transport, db cache and sync methods
;;
;; sync-method:        'original, 'attach or 'none
;; cache-method:       'tmp or 'none
;; rmt:transport-mode: 'http, 'tcp, 'nfs
;;
;; NOTE: NOT ALL COMBINATIONS WORK
;;
;;======================================================================

;; uncomment this block to test without tcp or cachedb
;; (dbfile:sync-method 'none)
;; (dbfile:cache-method 'none)
;; (rmt:transport-mode 'nfs)

;; uncomment this block to test with tcp and cachedb
(dbfile:sync-method 'none) ;; original was causing crash on start. 
(dbfile:cache-method 'none)
(rmt:transport-mode 'nfs)

;; uncomment this block to test without tcp or cachedb
;; (dbfile:sync-method 'none)
;; (dbfile:cache-method 'none)
;; (rmt:transport-mode 'nfs)

;; uncomment this block to test with tcp and cachedb
(dbfile:sync-method 'attach) ;; original was causing crash on start. 
(dbfile:sync-method 'none) ;; original was causing crash on start. 
(dbfile:cache-method 'none)
(rmt:transport-mode 'nfs)

(declare (uses tree))
(declare (uses dcommon))
(declare (uses dashboard-context-menu))
(declare (uses vg))
(declare (uses subrun))
(declare (uses mt))
(declare (uses dbmod))
(declare (uses dbfile))
(declare (uses dbfile.import))
(declare (uses rmtmod))
(declare (uses dbfile))
(declare (uses rmtmod.import))
(declare (uses commonmod))
(declare (uses commonmod.import))

(use format)

(require-library iup)
(import (prefix iup iup:))

;; set some parameters here - these need to be put in something that can be loaded from other
;; executables such as dashboard and mtutil
;;
(include "dashboard-transport-mode.scm")

(dbfile:db-init-proc db:initialize-main-db)
(set! rmtmod:send-receive rmt:send-receive)

(debug:print-info 0 *default-log-port* "transport-mode="(rmt:transport-mode))

(define help (conc 
	      "Megatest Dashboard, documentation at http://www.kiatoa.com/fossils/megatest version " megatest-version 
              " license GPL, Copyright (C) Matt Welland 2012-2017

Usage: dashboard [options]
  -h                    : this help

			"-:p"     ;; ignore the built in chicken profiling switch
			)
		 args:arg-hash
		 0))

(if (args:get-arg "-mode")
    (let* ((mode (string->symbol (args:get-arg "-mode"))))
      (rmt:transport-mode mode))
      (rmt:transport-mode mode)))
)

(if (args:get-arg "-test") ;; need to use tcp for test control panel -- Why? - Martin
    (rmt:transport-mode 'tcp))

;; RA => Might require revert for filters 
;; create a watch dog to move changes from lt/.db/*.db to megatest.db
;;
;;;(if (file-write-access? (conc *toppath* "/megatest.db"))
;;(debug:print-info 13 *default-log-port* "Before common:watchdog spawn")

;; (thread-start! (make-thread common:watchdog "Watchdog thread"))
;;(debug:print-info 13 *default-log-port* "After common:watchdog spawn")
;; (if (not (args:get-arg "-use-db-cache"))
;;     (begin
;;       (debug:print-info 0 *default-log-port* "Forcing db-cache mode due to read-only access to megatest.db")
;;       (hash-table-set! args:arg-hash "-use-db-cache" #t)));;;)
;;)

;; data common to all tabs goes here
;;
;; Moved to dcommon.scm
;;
(defstruct dboard:commondat
  ((curr-tab-num 0) : number)
  please-update  
  tabdats
  update-mutex
  updaters 
  updating
  uidat ;; needs to move to tabdat at some time
  hide-not-hide-tabs
  target
;; (defstruct dboard:commondat
;;   ((curr-tab-num 0) : number)
;;   please-update  
;;   tabdats
;;   update-mutex
;;   updaters 
;;   updating
;;   uidat ;; needs to move to tabdat at some time
;;   hide-not-hide-tabs
;;   target
  )

(define (dboard:commondat-make)
  (make-dboard:commondat
   curr-tab-num:         0
   tabdats:              (make-hash-table)
   please-update:        #t
   update-mutex:         (make-mutex)
   updaters:             (make-hash-table)
   updating:             #f
   hide-not-hide-tabs:   #f
   target:               ""
   ))
;;   )
;;
;; (define (dboard:commondat-make)
;;   (make-dboard:commondat
;;    curr-tab-num:         0
;;    tabdats:              (make-hash-table)
;;    please-update:        #t
;;    update-mutex:         (make-mutex)
;;    updaters:             (make-hash-table)
;;    updating:             #f
;;    hide-not-hide-tabs:   #f
;;    target:               ""
;;    ))

;;======================================================================
;; buttons color using image
;;======================================================================

(define *images* (make-hash-table))

	;; (iup:attribute-set! img1 "0" "0 0 0")
	(iup:attribute-set! img1 "1" color) ;; "BGCOLOR")
	;; (iup:attribute-set! img1 "2" "255 0 0")
	(hash-table-set! images name img1)
	name)))


;; RA => returns the tabdat stored at hashkey passed in commondat-tabdats table (e.g. 0 gives summary)
;;
(define (dboard:common-get-tabdat commondat #!key (tab-num #f))
  (let* ((tnum (or tab-num
		   (dboard:commondat-curr-tab-num commondat)
		   0)) ;; tab-num value is curr-tab-num value in passed commondat
	 (ht   (dboard:commondat-tabdats commondat))
	 (res  (hash-table-ref/default ht tnum #f)))
    (or res
	(let ((new-tabdat  (dboard:tabdat-make-data)))
	  (hash-table-set! ht tnum new-tabdat)
	  new-tabdat))))

;; RA => sets the tabdat passed to the hashkey at commondat:tabdats hash table
;;
(define (dboard:common-set-tabdat! commondat tabnum tabdat)
  (hash-table-set!
   (dboard:commondat-tabdats commondat)
   tabnum
   tabdat))

;; gets and calls updater list based on curr-tab-num
;;
(define (dboard:common-run-curr-updaters commondat #!key (tab-num #f))
  ;; (sync-db-to-tmp (dboard:common-get-tabdat commondat tab-num: tab-num)) ;; no longer applies

  ;; maybe need sleep here?

  
  (if (dboard:common-get-tabdat commondat tab-num: tab-num) ;; only update if there is a tabdat
      (let* ((tnum     (or tab-num (dboard:commondat-curr-tab-num commondat)))
	     (updaters (hash-table-ref/default (dboard:commondat-updaters commondat)
					       tnum
					       '())))
	(debug:print 4 *default-log-port* "Found these updaters: " updaters " for tab-num: " tnum)
	(for-each ;; perform the function calls for the complete updaters list

;;
;;    NOTE: Yes, this is used
;;
(define (dboard:get-tests-for-run-duplicate tabdat run-id run testnamepatt key-vals)
  (let* ((start-time   (current-seconds))
	 (access-mode  (dboard:tabdat-access-mode tabdat))
         (num-to-get   (string->number (or (configf:lookup *configdat* "setup" "num-tests-to-get")
                                           "200")))
                                           "1000")))
	 (states       (hash-table-keys (dboard:tabdat-state-ignore-hash tabdat)))
	 (statuses     (hash-table-keys (dboard:tabdat-status-ignore-hash tabdat)))
         (do-not-use-db-file-timestamps #f) ;; (configf:lookup *configdat* "setup" "do-not-use-db-file-timestamps")) ;; this still hosts runs-summary-tab
         (do-not-use-query-timestamps   #t) ;; (configf:lookup *configdat* "setup" "do-not-use-query-timestamps")) ;; this no longer troubles runs-summary-tab
	 (sort-info    (get-curr-sort))
	 (sort-by      (vector-ref sort-info 1))
	 (sort-order   (vector-ref sort-info 2))

	  (dboard:rundat-run-data-offset-set! run-dat
					      (+ num-to-get (dboard:rundat-run-data-offset run-dat)))))

    (for-each 
     (lambda (tdat)
       (let ((test-id (db:test-get-id tdat))
	     (state   (db:test-get-state tdat)))
	 (dboard:rundat-data-changed-set! run-dat #t)
	 (if (equal? state "DELETED")
	     (hash-table-delete! tests-ht test-id)
	     (hash-table-set! tests-ht test-id tdat))))
     tmptests)
    
        (dboard:rundat-data-changed-set! run-dat #t)
        (if (equal? state "DELETED")
            (hash-table-delete! tests-ht test-id)
	    (hash-table-set! tests-ht test-id tdat))))
      tmptests)

    tests-ht))

;; tmptests   - new tests data
;; prev-tests - old tests data
;;
;; (define (dashboard:merge-changed-tests tabdat tests tmptests) ;;  use-new prev-tests) 
;;   (let* ((newdat     (filter

		    maxtests)
		  (if (> (dboard:rundat-run-data-offset run-struct) 0)
		      (loop run tal new-res newmaxtests) ;; not done getting data for this run
		      (loop (car tal)(cdr tal) new-res newmaxtests)))))))
    (dboard:tabdat-filters-changed-set! tabdat #f)
    (dboard:update-tree tabdat runs-hash header tb)))


(define *dashboard-last-run-id-update* (make-hash-table)) ;; id => seconds

(define (dboard:clear-run-id-update-hash)
  (hash-table-clear! *dashboard-last-run-id-update*))

;; this calls dboard:get-tests-for-run-duplicate for each run
;;
;; create a virtual table of all the tests
;; keypatts: ( (KEY1 "abc%def")(KEY2 "%") )
;;
(define (dboard:update-rundat tabdat runnamepatt numruns testnamepatt keypatts)
  (let* ((access-mode      (dboard:tabdat-access-mode tabdat))

	  (dboard:tabdat-allruns-set! tabdat '())
	  (dboard:tabdat-all-test-names-set! tabdat '())
	  (dboard:tabdat-item-test-names-set! tabdat '())
	  (hash-table-clear! (dboard:tabdat-allruns-by-id tabdat)))
	(let loop ((run      (car runs))
		   (tal      (cdr runs))
		   (res     '())
		   (maxtests 0))
		   (maxtests 0)
		   (cont-run #f))
	  (let* ((run-id       (db:get-value-by-header run header "id"))
		 (recently-done  (< (- (current-seconds)
				       (hash-table-ref/default *dashboard-last-run-id-update* run-id 0)) 1))
		 (run-struct   (hash-table-ref/default (dboard:tabdat-allruns-by-id tabdat) run-id #f))
		 ;; (last-update  (if run-struct (dboard:rundat-last-update run-struct) 0))
		 (key-vals     (rmt:get-key-vals run-id))
		 (tests-ht     (let* ((tht (if (and recently-done run-struct)
					       (let ((rht (dboard:rundat-tests run-struct))) ;; (dboard:tabdat-allruns-by-id tabdat)))
						 (or rht
		 (tests-ht     (dboard:get-tests-for-run-duplicate tabdat run-id run testnamepatt key-vals))
						     (dboard:get-tests-for-run-duplicate tabdat run-id run testnamepatt key-vals)))
					       (dboard:get-tests-for-run-duplicate tabdat run-id run testnamepatt key-vals))))
                                (assert (hash-table? tht) "FATAL: But here tht should be a hash-table")
				 tht))
		 ;; GET RID OF dboard:get-tests-dat - it is superceded by dboard:get-tests-for-run-duplicate
		 ;;  dboard:get-tests-for-run-duplicate - returns a hash table
		 ;;  (dboard:get-tests-dat tabdat run-id last-update))
		 (all-test-ids (hash-table-keys tests-ht))
		 (num-tests    (length all-test-ids)))
	    ;; (print "run-struct: " run-struct)
	    ;; NOTE: bubble-up also sets the global (dboard:tabdat-item-test-names tabdat)
	    ;; (tests       (bubble-up tmptests priority: bubble-type))
	    ;; NOTE: 11/01/2013 This routine is *NOT* getting called excessively.
	    ;; (debug:print 0 *default-log-port* "Getting data for run " run-id " with key-vals=" key-vals)
	    ;; Not sure this is needed?
	    (let* ((newmaxtests (max num-tests maxtests))
		   ;; (last-update (- (current-seconds) 10))
		   (run-struct  (or run-struct
				    (dboard:rundat-make-init
				     run:         run 
				     tests:       tests-ht
				     key-vals:    key-vals)))
		   (new-res     (if (null? all-test-ids)
                                    res
                                    (delete-duplicates
                                     (cons run-struct res)
                                     (lambda (a b)
                                       (eq? (db:get-value-by-header (dboard:rundat-run a) header "id")
                                            (db:get-value-by-header (dboard:rundat-run b) header "id"))))))
		   (elapsed-time (- (current-seconds) start-time)))
	      (if (null? all-test-ids)
		 (num-tests    (length all-test-ids))
		 ;; (print "run-struct: " run-struct)
		 ;; NOTE: bubble-up also sets the global (dboard:tabdat-item-test-names tabdat)
		 ;; (tests       (bubble-up tmptests priority: bubble-type))
		 ;; NOTE: 11/01/2013 This routine is *NOT* getting called excessively.
		 ;; (debug:print 0 *default-log-port* "Getting data for run " run-id " with key-vals=" key-vals)
		 ;; Not sure this is needed?
		 (newmaxtests (max num-tests maxtests))
		 ;; (last-update (- (current-seconds) 10))
		 (run-struct  (or run-struct
				  (dboard:rundat-make-init
				   run:         run 
				   tests:       tests-ht
				   key-vals:    key-vals)))
		 (new-res     (if (null? all-test-ids)
				  res
				  (delete-duplicates
				   (cons run-struct res)
				   (lambda (a b)
				     (eq? (db:get-value-by-header (dboard:rundat-run a) header "id")
					  (db:get-value-by-header (dboard:rundat-run b) header "id"))))))
		 (elapsed-time (- (current-seconds) start-time)))
	    (if (null? all-test-ids)
		  (hash-table-delete! (dboard:tabdat-allruns-by-id tabdat) run-id)
		  (hash-table-set!    (dboard:tabdat-allruns-by-id tabdat) run-id run-struct))
	   
	      (if (or (null? tal)
		      (> elapsed-time 2)) ;; stop loading data after 5 seconds, on the next call more data *should* be loaded since get-tests-for-run uses last update
		  (begin
		    (when (> elapsed-time 2)   
	    (if (or (null? tal)
		    (> elapsed-time 2)) ;; stop loading data after 5
					;; seconds, on the next call
					;; more data *should* be
					;; loaded since
					;; get-tests-for-run uses last
					;; update
		(begin
		  (when (> elapsed-time 2)   
                      (debug:print 2 *default-log-port* "NOTE: updates are taking a long time, " elapsed-time "s elapsed.")
                      (let* ((old-val (iup:attribute *tim* "TIME"))
                             (new-val (number->string (inexact->exact (floor (* 2  (string->number old-val)))))))
                        (if (< (string->number new-val) 5000)
                            (begin
		    (let* ((old-val (iup:attribute *tim* "TIME"))
			   (new-val (number->string (inexact->exact (floor (* 2  (string->number old-val)))))))
		      (if (< (string->number new-val) 5000)
			  (begin
			      (debug:print 2 *default-log-port* "NOTE: increasing poll interval from "old-val" to "new-val)
			      (iup:attribute-set! *tim* "TIME" new-val)))))
		    (dboard:tabdat-allruns-set! tabdat new-res)
		    maxtests)
		  (if (> (dboard:rundat-run-data-offset run-struct) 0)
		      (loop run tal new-res newmaxtests) ;; not done getting data for this run
		      (loop (car tal)(cdr tal) new-res newmaxtests)))))))
    (dboard:tabdat-filters-changed-set! tabdat #f)
    (dboard:update-tree tabdat runs-hash header tb)))
			    (iup:attribute-set! *tim* "TIME" new-val)))))
		  (dboard:tabdat-allruns-set! tabdat new-res)
		  maxtests)
		(if (> (dboard:rundat-run-data-offset run-struct) 0)
		      (begin
			(thread-sleep! 0.2) ;; let the gui re-draw
			(loop run tal new-res newmaxtests #t)) ;; not done getting data for this run
		      (begin
			(hash-table-set! *dashboard-last-run-id-update* run-id (current-seconds))
			(loop (car tal)(cdr tal) new-res newmaxtests #f)))))))
	(dboard:tabdat-filters-changed-set! tabdat #f)
	(dboard:update-tree tabdat runs-hash header tb)))

(define *collapsed* (make-hash-table))

(define (toggle-hide lnum uidat) ; fulltestname)
  (let* ((btn (vector-ref (dboard:uidat-get-lftcol uidat) lnum))
	 (fulltestname (iup:attribute btn "TITLE"))
	 (parts        (string-split fulltestname "("))

     tabdat
     (let ((xl (if (> (length (dboard:tabdat-all-test-names tabdat)) (dboard:tabdat-start-test-offset tabdat))
		   (drop (dboard:tabdat-all-test-names tabdat)
			 (dboard:tabdat-start-test-offset tabdat))
		   '())))
       (append xl (make-list (- (dboard:tabdat-num-tests tabdat) (length xl)) ""))))
    (update-labels uidat (dboard:tabdat-all-test-names tabdat))
    (for-each
    (for-each ;;run
     (lambda (rundat)
       ;; if rundat is junk clobber it with a decent placeholder
       (if (or (not rundat) ;; handle padded runs
	       (not (dboard:rundat-run rundat)))
           ;; Need to put an empty column in to erase previous contents.
	   (set! rundat (dboard:rundat-make-init
			 key-vals: (map (lambda (x) "")(dboard:tabdat-keys tabdat)))))
       (let* ((run              (dboard:rundat-run rundat))
	      (testsdat-by-name (dboard:rundat-tests-by-name rundat))
	      (key-val-dat      (dboard:rundat-key-vals rundat))
	      (run-id           (db:get-value-by-header run (dboard:tabdat-header tabdat) "id"))
	      (key-vals         (append key-val-dat
					(list (let ((x (db:get-value-by-header run (dboard:tabdat-header tabdat) "runname")))
						(if (string? x) x "")))))
	      (run-key          (string-intersperse key-vals "\n")))
	 

	 ;; fill in the run header key values
	 ;;
	 (let ((rown      0)
        (let ((rown      0)
	       (headercol (vector-ref tableheader coln)))
	   (for-each (lambda (kval)
		       (let* ((labl      (vector-ref headercol rown)))
			 (if (not (equal? kval (iup:attribute labl "TITLE")))
			     (iup:attribute-set! (vector-ref headercol rown) "TITLE" kval))
			 (set! rown (+ rown 1))))
		     key-vals))
	 
	 ;; For this run now fill in the buttons for each test
        ;; For this run now fill in the buttons for each test	 
	 ;;
	 (let ((rown 0)
	       (columndat  (vector-ref table coln)))
	   (for-each
	    (lambda (testname)
	      (let ((buttondat  (hash-table-ref/default (dboard:tabdat-buttondat tabdat) (mkstr coln rown) #f)))
		(if (and buttondat
			 (hash-table? testsdat-by-name))
		    (let* ((testdat      (let ((matching (hash-table-ref/default testsdat-by-name testname #f)))
					   ;; (filter 
					   ;;   (lambda (x)(equal? (test:test-get-fullname x) testname))
					   ;;     testsdat)))
					   (if (not matching)
					       (vector -1 -1 "" "" "" 0 "" "" 0 "" "" "" 0 "" "")
					       ;; (car matching))))
					       matching)))
			   (testname     (db:test-get-testname   testdat))
			   (itempath     (db:test-get-item-path  testdat))
			   (testfullname (test:test-get-fullname testdat))
			   (teststatus   (db:test-get-status     testdat))
			   (teststate    (db:test-get-state      testdat))
			   ;;(teststart  (db:test-get-event_time test))
			   ;;(runtime    (db:test-get-run_duration test))
			   (buttontxt    (cond
					  ((member teststate '("COMPLETED" "ARCHIVED")) teststatus)
					  ((and (equal? teststate "NOT_STARTED")
						(member teststatus '("ZERO_ITEMS" "BLOCKED" "PREQ_FAIL" "PREQ_DISCARDED" "TIMED_OUT" "KEEP_TRYING" "TEN_STRIKES")))
					   teststatus)
					  (else
					   teststate)))

                                 (equal? tp ""))
                             "%"
                             tp)))
	 (states       (dboard:tabdat-states     tabdat))
	 (statuses     (dboard:tabdat-statuses   tabdat))
	 (target       (let ((targ-list (dboard:tabdat-target     tabdat)))
			 (if targ-list (string-intersperse targ-list "/") "no-target-selected")))
	 (run-name     (dboard:tabdat-run-name   tabdat))
	 (run-name     (let ((run-input (dboard:tabdat-run-name   tabdat))
                             )
                         (if (equal? run-input "")
                           "no-runname-specified"
                           run-input)))
	 (states-str   (if (or (not states)
			       (null? states))
			   ""
			   (conc " -state "  (string-intersperse states ","))))
	 (statuses-str (if (or (not statuses)
			       (null? statuses))
			   ""

                                         #:modal? "NO")
                               )
                              )
                            
                             )) "runs-summary-click-callback"))))
	 (runs-summary-updater  
          (lambda ()
	    (mutex-lock! update-mutex)
	    ;; (mutex-lock! update-mutex)
            (if  (or (dashboard:database-changed? commondat tabdat context-key: 'runs-summary-updater)
                     (dboard:tabdat-view-changed tabdat))
                 (debug:catch-and-dump
                  (lambda () ;; check that run-matrix is initialized before calling the updater
		    (if run-matrix 
			(dashboard:runs-summary-updater commondat tabdat tb cell-lookup run-matrix)))
                  "dashboard:runs-summary-updater")
                 )
	    (mutex-unlock! update-mutex)))
	    #;(mutex-unlock! update-mutex)
	    ))
         (runs-summary-control-panel (dashboard:runs-summary-control-panel tabdat))
         )
    (dboard:commondat-add-updater commondat runs-summary-updater tab-num: tab-num)
    (dboard:tabdat-runs-tree-set! tabdat tb)
    (iup:vbox
     (iup:split
      #:value 200

       (iup:vbox
        (iup:hbox
	(iup:vbox
	 (iup:textbox #:size "120x15" #:fontsize "10" #:value "%"
		      #:expand "NO"
		      #:action (lambda (obj unk val)
				 (debug:catch-and-dump
				  (lambda ()
				  (lambda ()57
				    (mark-for-update tabdat)
				    (update-search commondat tabdat "test-name" val))
				  "make-controls")))
	 (iup:hbox
	  (iup:button "Quit"      #:action (lambda (obj)
					     (exit))
		      #:expand "NO" #:size "40x15")
	  (iup:button "Refresh"   #:action (lambda (obj)
                                             (dboard:tabdat-last-data-update-set! tabdat 0)
                                             (dboard:tabdat-last-runs-update-set! tabdat 0)
                                             (dboard:tabdat-run-update-times-set! tabdat (make-hash-table))
                                             (dboard:tabdat-last-test-dat-set!    tabdat (make-hash-table))
                                             (dboard:tabdat-allruns-set!          tabdat '())
                                             (dboard:tabdat-allruns-by-id-set!    tabdat (make-hash-table))
                                             (dboard:tabdat-done-runs-set!        tabdat '())
                                             (dboard:tabdat-not-done-runs-set!    tabdat '())
                                             (dboard:tabdat-view-changed-set!     tabdat #t)
                                             (dboard:commondat-please-update-set! commondat #t)
					     (dboard:clear-run-id-update-hash)
					     (mark-for-update tabdat))
		      #:expand "NO" #:size "40x15")
	  (iup:button "Collapse"  #:action (lambda (obj)
					     (debug:catch-and-dump 
					      (lambda ()
						(let ((myname (iup:attribute obj "TITLE")))
						  (if (equal? myname "Collapse")

(define (dboard:set-last-db-update! tabdat context newtime)
  (hash-table-set! (dboard:tabdat-last-db-update tabdat) context newtime))

;;
(define (dashboard:database-changed? commondat tabdat #!key (context-key 'default))
  (let* ((run-update-time (current-seconds))
         (dbdir           (conc *toppath* "/.mtdb"`))
         (dbdir           (conc *toppath* "/.mtdb"))
	 (modtime         (dashboard:get-youngest-run-db-mod-time dbdir))
	 (recalc          (dashboard:recalc modtime 
					    (dboard:commondat-please-update commondat) 
					    (dboard:get-last-db-update tabdat context-key))))
    (if recalc 
	(dboard:set-last-db-update! tabdat context-key run-update-time))
    (dboard:commondat-please-update-set! commondat #f)

	(dwg (dboard:tabdat-drawing tabdat))
	(mtx (dboard:tabdat-runs-mutex tabdat))
	(vch (dboard:tabdat-view-changed tabdat)))
    (if (and cnv dwg vch)
	(begin
	  (vg:drawing-xoff-set! dwg (dboard:tabdat-xadj tabdat))
	  (vg:drawing-yoff-set! dwg (dboard:tabdat-yadj tabdat))
	  (mutex-lock! mtx)
	  ;; (mutex-lock! mtx)
	  (canvas-clear! cnv)
	  (vg:draw dwg tabdat)
	  (mutex-unlock! mtx)
	  ;; (mutex-unlock! mtx)
	  (dboard:tabdat-view-changed-set! tabdat #f)))))
  
;; doesn't work.
;;
;;(define (gotoescape tabdat escape)
;;  (or (dboard:tabdat-layout-update-ok tabdat)
;;      (escape #t)))

			       (width      (* timescale run-duration))
			       (graph-lly  (calc-y (/ -50 row-height)))
			       (graph-uly  (- (calc-y 0) canvas-margin))
			       (sec-per-50pt (/ 50 timescale))
			       )
			  ;; (print "timeoffset: " timeoffset " timescale: " timescale " run-duration: " (seconds->hr-min-sec run-duration) " width: " width " sec-per-50pt: " sec-per-50pt)
			  ;; (print "timescale: " timescale " timeoffset: " timeoffset " sizex: " sizex " originx: " originx)
			  (mutex-lock! mtx)
			  ;; (mutex-lock! mtx)
			  (vg:add-comp-to-lib runslib run-full-name runcomp)
			  ;; Have to keep moving the instantiated box as it is anchored at the lower left
			  ;; this should have worked for x in next statement? (maptime run-start)
			  ;; add 60 to make room for the graph
			  (vg:instantiate drawing "runslib" run-full-name run-full-name 8 (- (calc-y curr-run-start-row) (+ 5 graph-height run-to-run-margin)))
			  (mutex-unlock! mtx)
			  ;; (mutex-unlock! mtx)
			  ;; (set! run-start-row (+ max-row 2))
			  ;; (dboard:tabdat-start-row-set! tabdat (+ new-run-start-row 1))
			  ;; get tests in list sorted by event time ascending
			  (let testsloop ((test-ids  (car hierdat))              ;; loop on tests (NOTE: not items!)
					  (tests-tal (cdr hierdat))
					  (test-num  1))
			    (let ((iterated     (> (length test-ids) 1))

				 (ulx       (list-ref new-xtnts 2))
				 (uly       (list-ref new-xtnts 3))
				 (outln     (vg:make-rect-obj -5 lly ulx uly 
							      text: run-full-name
							      line-color:  (vg:rgb->number  255 0 255 a: 128))))
					;  (vg:components-get-extents d1 c1)))
			    ;; this is the box around the run
			    (mutex-lock! mtx)
			    ;; (mutex-lock! mtx)
			    (vg:add-obj-to-comp runcomp outln)
			    (mutex-unlock! mtx)
			    ;; (mutex-unlock! mtx)
			    ;; this is where we have enough info to place the graph
			    (dboard:graph commondat tabdat tab-num -5 (+ uly 10) ulx (+ uly graph-height 3) run-start run-end timescale maptime run-full-name canvas-margin)
			    (dboard:tabdat-max-row-set! tabdat (+ (dboard:tabdat-max-row tabdat)(quotient (+ graph-height 40 3) row-height)))
			    ;; (vg:instance-move drawing run-full-name 0 (dboard:tabdat-max-row tabdat))
			    ))
			;; end of the run handling loop 
			(if (not (dboard:tabdat-layout-update-ok tabdat))

	;;  (lambda ()
	   ;;  (dashboard:runs-tab-updater commondat 1))
	;; tab-num: 2)
	(iup:callback-set! *tim*
			   "ACTION_CB"
			   (lambda (time-obj)
			     (let ((update-is-running #f))
			     (mutex-lock! (dboard:commondat-update-mutex commondat))
			     (set! update-is-running (dboard:commondat-updating commondat))
			     (if (not update-is-running)
			     (dboard:commondat-updating-set! commondat #t))
			     (mutex-unlock! (dboard:commondat-update-mutex commondat))
			     (if (not update-is-running) ;; we know that the update was not running and we now have a lock on doing an update
			       ;; (mutex-lock! (dboard:commondat-update-mutex commondat))
			       (set! update-is-running (dboard:commondat-updating commondat))
			       (if (not update-is-running)
				   (dboard:commondat-updating-set! commondat #t))
			       ;; (mutex-unlock! (dboard:commondat-update-mutex commondat))
			       (if (not update-is-running) ;; we know that the update was not running and we now have a lock on doing an update
			     (begin
			     (dboard:common-run-curr-updaters commondat) ;; (dashboard:run-update commondat)
			     (mutex-lock! (dboard:commondat-update-mutex commondat))
			     ;; (mutex-lock! (dboard:commondat-update-mutex commondat))
			     (dboard:commondat-updating-set! commondat #f)
			     (mutex-unlock! (dboard:commondat-update-mutex commondat)))
				   ))
			     ;; (mutex-unlock! (dboard:commondat-update-mutex commondat))
			     )))
			     1))))
      ;; (debug:print 0 *default-log-port* "Starting updaters")
      (let ((th1 (make-thread (lambda ()
				(thread-sleep! 1)
				(dboard:common-run-curr-updaters commondat 0) ;; force update of summary tab 
				) "update buttons once"))
	    (th2 (make-thread iup:main-loop "Main loop")))

;; Sync to tmp only if in read-only mode.

(define (sync-db-to-tmp tabdat)
  (let* ((db-file "./.mtdb/main.db"))
    (if (and (not (file-write-access? db-file)) ( > (current-seconds) (+ last-copy-time 5)))
      (begin
        (db:multi-db-sync (db:setup #f) 'old2new)
        (db:multi-db-sync (db:setup) 'old2new)
        (set! last-copy-time (current-seconds))
      )
    )
  )
)

;; ########################### top level code ########################
;; check for MT_* environment variables and exit if found
(if (not (args:get-arg "-test"))
    (begin
      (for-each (lambda (var)
		  ;; (display " ")(display var)
		  (if (get-environment-variable var)
		      (begin
			(debug:print 0 *default-log-port* "ERROR: environment variable " var " is set in this terminal, this will cause you problems. Exiting now.")
			(exit 1))))
		'("MT_RUN_AREA_HOME" "MT_MEGATEST" "MT_CMDINFO" "MT_TEST_RUN_DIR" "MT_LINKTREE" "MT_TESTSUITENAME"))
    )
)

;; This is NOT good
(setenv "MT_RUN_AREA_HOME" (get-environment-variable "PWD"))
;; (setenv "MT_RUN_AREA_HOME" (get-environment-variable "PWD"))
;; This should be OK but it really should not be necessary
(setenv "MT_RUN_AREA_HOME" (current-directory))

(if (not (null? remargs))
  (if remargs
    (begin
      (debug:print 0 *default-log-port* "Unrecognised arguments: " (string-intersperse remargs " "))
      (exit)
    )

	 default
	 (begin
	   (debug:print-error 0 *default-log-port* " query " stmt " failed, params: " params ", error: " ((condition-property-accessor 'exn 'message) exn) ", exn=" exn)
	   (print-call-chain (current-error-port))
	   default)))
   (apply sqlite3:first-result db stmt params)))

(define (db:setup do-sync)
(define (db:setup)
  (assert *toppath* "FATAL: db:setup called before launch:setup has been run.")
  (let* ((tmpdir (common:make-tmpdir-name *toppath* "")))
    (if (not *dbstruct-dbs*)
	(dbfile:setup do-sync *toppath* tmpdir)
	(dbfile:setup (conc *toppath* "/.mtdb") tmpdir)
	*dbstruct-dbs*)))

;; moved from dbfile
;;
;; ADD run-id SUPPORT
;;
(define (db:create-all-triggers dbstruct)

;;      dbfiles)
;;     ;; WHY does the dbdat need to be added back?
;;     (if dbdat (dbfile:add-dbdat dbstruct #f dbdat))
;;     )
;;   #t)

(define (db:kill-servers)
  (let* ((tl (launch:setup)) ;; need this to initialize *toppath*
        (servdir (conc *toppath* "/.servinfo"))
        (servfiles (glob (conc servdir "/*:*.db")))
        (fmtstr  "~10a~22a~10a~25a~25a~8a\n")
        (dbfiles (append (glob (conc *toppath* "/.mtdb/main.db")) (glob (conc *toppath* "/.mtdb/?.db"))(glob (conc *toppath* "/.mtdb/??.db"))))
  (let* ((servers (server:choose-server *toppath* 'all-valid))) ;; (server:get-list *toppath*))
    (for-each
     (lambda (server)
       (handle-exceptions
           exn
         (begin 
        (ttdat (make-tt areapath: *toppath*))
     )
     (format #t fmtstr "DB" "host:port" "PID" "age" "last mod" "state")
     (for-each
        (lambda (dbfile)
          (let* (
            (dbfname (conc (pathname-file dbfile) ".db"))
            (sfiles   (tt:find-server *toppath* dbfname))
            )
            (for-each 
              (lambda (sfile)
                (let (
                  (sinfos (tt:get-server-info-sorted ttdat dbfname))
                  )
                  (for-each 
           (debug:print-info 0 *default-log-port*  "Unable to get host and/or port from " server ", exn=" exn)     
           #f)
                     (lambda (sinfo)
	 (match-let (((mod-time host port start-time server-id pid) server))
		    (if (and host pid)
			(tasks:kill-server host pid)))))
     servers)
    (delete-file* (common:get-sync-lock-filepath))))
                       (let* (
                         (db (list-ref sinfo 5))
                         (pid (list-ref sinfo 4))
                         (host (list-ref sinfo 0))
                         (port (list-ref sinfo 1))
                         (server-id (list-ref sinfo 3))
                         (age (seconds->hr-min-sec (- (current-seconds) (list-ref sinfo 2))))
                         (last-mod (seconds->string (list-ref sinfo 2)))
                         (killed (system (conc "ssh " host " kill " pid " > /dev/null")))
                         (dummy2 (sleep 1))
                         (state (if (> (system (conc "ssh " host " ps " pid " > /dev/null")) 0) "dead" "alive"))
                            )
                         (format #t fmtstr db (conc host ":" port) pid age last-mod state)
                         (system (conc "rm " sfile))
                       )
                     )
                     sinfos
                  )
                ) 
              )
              sfiles
            )
          )
       )
       dbfiles
     )
     ;; remove this db, because otherwise metadata contains records for old servers, and this causes a problem with db:no-sync-get-lock-with-id.
     (if (file-exists? (conc *toppath* "/.mtdb/no-sync.db"))
       (delete-file (conc *toppath* "/.mtdb/no-sync.db"))
     )
  )
)

;; 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)
  (let* (;; (dbdat       (db:open-db dbstruct #f dbfile:db-init-proc))
	 (data-synced 0) ;; count of changed records
	 (tmp-area       (common:make-tmpdir-name *toppath* ""))
	 (old2new (member 'old2new options))
	 (dejunk (member 'dejunk options))
	 (killservers (member 'killservers options))
	 (src-area (if old2new *toppath* tmp-area))
	 (dest-area (if old2new tmp-area *toppath*))
	 (dest-area (if old2new tmp-area (conc *toppath* "/.mtdb")))
	 (dbfiles        (if old2new (glob (conc *toppath* "/.mtdb/*.db"))
			     (glob (conc tmp-area "/.mtdb/*.db"))))
			     (glob (conc tmp-area "/*.db"))))
	 (keys (db:get-keys dbstruct))
	 (sync-durations (make-hash-table)))

    ;; kill servers
    (if killservers (db:kill-servers))
    ;; (if killservers (db:kill-servers))
    
    (if (not dbfiles)
	(debug:print-error 0 *default-log-port* "no dbfiles found in " (conc *toppath* "/.mtdb"))
	(for-each
	 (lambda (srcfile)
	   (debug:print-info 3 *default-log-port* "file: " srcfile)
	   (let* ((fname    (conc (pathname-file srcfile) ".db"))
		  (basename (pathname-file srcfile))
		  (run-id   (if (string= basename "main") #f (string->number basename)))
		  (destfile (conc dest-area "/.mtdb/" fname))
		  (dest-directory  (conc dest-area "/.mtdb/"))
		  (destfile (conc dest-area "/" fname))
		  (dest-directory  dest-area)
		  (time1    (file-modification-time srcfile))
		  (time2    (if (file-exists? destfile)
				(begin
				  (debug:print-info 2 *default-log-port* "destfile " destfile " exists")
				  (file-modification-time destfile))
				(begin
				  (debug:print-info 0 *default-log-port* "Sync - I do not see file " destfile)

			       #t)
			      (changed ;; (and changed
			       #t)
			      ((and changed *time-to-exit*) ;; last sync
			       #t)
			      (else
			       #f))))
             
             (if (or dejunk do-cp)
               (if (or dejunk do-cp)
		 (let* ((start-time (current-milliseconds))
			;; subdb is misnamed - should be dbdat (I think...)
			(subdb    (dbfile:open-db dbstruct run-id dbfile:db-init-proc))
			;;        (or (dbfile:get-subdb dbstruct run-id)
			;;            (dbfile:init-subdb dbstruct run-id dbfile:db-init-proc)))
			(subdb    (or (dbfile:get-subdb dbstruct run-id) (dbfile:init-subdb dbstruct run-id dbfile:db-init-proc)))
			(dbdat    (or (dbfile:get-dbdat dbstruct run-id) (dbfile:open-db dbstruct run-id dbfile:db-init-proc)))
			(mtdb      (dbr:subdb-mtdbdat subdb))
			;;
			;; BUG: -mrw- I think this next line is wrong. run-id should be the path to .mtdb/<runid>.db
			;; 
			(tmpdb     (dbfile:open-db dbstruct run-id dbfile:db-init-proc)))
		   
                   (if dejunk
                     (begin
                       (debug:print 0 *default-log-port* "Cleaning tmp DB")
                       (db:clean-up run-id tmpdb)
                       (debug:print 0 *default-log-port* "Cleaning nfs DB")
                       (db:clean-up run-id mtdb)
                     )
                   )
		   (debug:print-info 2 *default-log-port* "delta syncing file: " srcfile ", time diff: " (- time1 time2) " seconds")
		   (if old2new
                       (begin
			 (if dejunk (db:clean-up run-id mtdb))
			 (db:sync-tables (db:sync-all-tables-list
					  dbstruct
					  (db:get-keys dbstruct))
					 #f mtdb tmpdb))
                       (begin
			 (if dejunk (db:clean-up run-id tmpdb))
			 (db:sync-tables (db:sync-all-tables-list dbstruct (db:get-keys dbstruct)) #f tmpdb mtdb)))
			 (db:sync-tables (db:sync-all-tables-list (db:get-keys dbstruct)) #f tmpdb mtdb)))
		   (hash-table-set! sync-durations (conc srcfile ".db") (- (current-milliseconds) start-time)))
		 (debug:print-info 2 *default-log-port* "skipping delta sync. " srcfile " is up to date"))))
	 dbfiles))
    data-synced))

;; Sync all changed db's
;;
(define (db:tmp->megatest.db-sync dbstruct run-id last-update)
  (let* ((subdbs (hash-table-values (dbr:dbstruct-subdbs dbstruct)))
	 (res    '()))
    (for-each
     (lambda (subdb)
       (let* ((mtdb   (dbr:subdb-mtdb subdb))
	      (tmpdb  (db:get-subdb dbstruct run-id))
	      (refndb (dbr:subdb-refndb subdb))
	      (newres (db:sync-tables (db:sync-all-tables-list dbstruct (db:get-keys dbstruct)) last-update tmpdb refndb mtdb)))
	      (newres (db:sync-tables (db:sync-all-tables-list (db:get-keys dbstruct)) last-update tmpdb refndb mtdb)))
	 ;; (stack-push! (dbr:subdb-dbstack subdb) tmpdb)
	 ;; BUG: verify this is really needed
	 (dbfile:add-dbdat dbstruct run-id tmpdb)
	 (set! res (cons newres res))))
     subdbs)
    res))

;;    a. If test dir exists, set the the test to state='UNKNOWN', Set the run to 'unknown'
;;    b. If test dir gone, delete the test record
;; 2. Look at run records
;;    a. If have tests that are not deleted, set state='unknown'
;;    b. ....
;;
(define (db:clean-up run-id dbdat)
  (debug:print 2 *default-log-port* "db:clean-up")


  (if run-id
    (begin
      (debug:print 0 *default-log-port* "Cleaning run DB " run-id)
    (db:clean-up-rundb dbdat)
    (db:clean-up-maindb dbdat)
      (db:clean-up-rundb dbdat run-id)
    )
    (begin
      (debug:print 0 *default-log-port* "Cleaning main DB ")
      (db:clean-up-maindb dbdat)
    )
  )
)


;; Clean out old junk and vacuum the database
;;
;; Ultimately do something like this:
;;
;; 1. Look at test records either deleted or part of deleted run:
;;    a. If test dir exists, set the the test to state='UNKNOWN', Set the run to 'unknown'
;;    b. If test dir gone, delete the test record
;; 2. Look at run records
;;    a. If have tests that are not deleted, set state='unknown'
;;    b. ....
;;
(define (db:clean-up-rundb dbdat)
(define (db:clean-up-rundb dbdat run-id)
  ;; (debug:print 0 *default-log-port* "WARNING: db clean up not fully ported to v1.60, cleanup action will be on megatest.db")
  (let* ((db         (dbr:dbdat-dbh dbdat))
	 (count-stmt (sqlite3:prepare db "SELECT (SELECT count(id) FROM tests);"))
	 (test-count-stmt (sqlite3:prepare db "SELECT (SELECT count(id) FROM tests);"))
       	 (step-count-stmt (sqlite3:prepare db "SELECT (SELECT count(id) FROM test_steps);"))
	(statements
	 (map (lambda (stmt)
		(sqlite3:prepare db stmt))
	      (list
	       ;; delete all tests that belong to runs that are 'deleted'
	       ;; (conc "DELETE FROM tests WHERE run_id NOT IN (" (string-intersperse (map conc valid-runs) ",") ");")
	       ;; delete all tests that are 'DELETED'
	       "DELETE FROM tests WHERE state='DELETED';"
	       "DELETE FROM tests WHERE state='DELETED';"
               "DELETE FROM test_steps WHERE status = 'DELETED';"
               "DELETE FROM tests WHERE run_id IN (SELECT id FROM runs WHERE state = 'deleted');"
	       ))))
    ;; (db:delay-if-busy dbdat)
    (sqlite3:with-transaction 
     db
     (lambda ()
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 *default-log-port* "Records count before clean: " tot))
			     count-stmt)
			       (debug:print-info 0 *default-log-port* "Test records count before clean: " tot))
			     test-count-stmt)
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 *default-log-port* "Test_step records count before clean: " tot))
			     step-count-stmt)
       (map sqlite3:execute statements)
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 *default-log-port* "Records count after  clean: " tot))
			     count-stmt)))
			       (debug:print-info 0 *default-log-port* "Test records count after clean: " tot))
			     test-count-stmt)
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 *default-log-port* "Test_step records count after clean: " tot))
			     step-count-stmt)))
    (map sqlite3:finalize! statements)
    (sqlite3:finalize! count-stmt)
    ;; (db:find-and-mark-incomplete db)
    (sqlite3:finalize! test-count-stmt)
    (sqlite3:finalize! step-count-stmt)
    ;; (db:delay-if-busy dbdat)
    (sqlite3:execute db "VACUUM;")))

;; Clean out old junk and vacuum the database
;;
;; Ultimately do something like this:
;;
;; 1. Look at test records either deleted or part of deleted run:

       db
       "SELECT id FROM runs WHERE state='deleted';")
    ;; (db:delay-if-busy dbdat)
    (sqlite3:with-transaction 
     db
     (lambda ()
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 *default-log-port* "Records count before clean: " tot))
			       (debug:print-info 0 *default-log-port* "Run records count before clean: " tot))
			     count-stmt)
       (map sqlite3:execute statements)
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 *default-log-port* "Records count after  clean: " tot))
			       (debug:print-info 0 *default-log-port* "Run records count after clean: " tot))
			     count-stmt)))
    (map sqlite3:finalize! statements)
    (sqlite3:finalize! count-stmt)
    ;; (db:find-and-mark-incomplete db)
    ;; (db:delay-if-busy dbdat)
    (sqlite3:execute db "VACUUM;")
    dead-runs))

  (let ((runs (db:simple-get-runs dbstruct runname #f #f target #f))) ;; runpatt count offset target last-update
    (if (null? runs)
	#f
	(simple-run-id (car runs)))))

;; called with run-id=#f so will operate on main.db
;;
(define (db:insert-run dbstruct target runname run-meta)
(define (db:insert-run dbstruct run-id target runname run-meta)
  (let* ((keys (db:get-keys dbstruct))
     	 (runs (db:simple-get-runs dbstruct runname #f #f target #f))) ;; runpatt count offset target last-update
    ;; need to insert run based on target and runname
    (let* ((targvals (string-split target "/"))
	   (keystr   (string-intersperse keys ","))
	   (key?str  (string-intersperse (make-list (length targvals) "?") ","))
	   (qrystr   (conc "INSERT INTO runs (runname,"keystr") VALUES (?,"key?str")"))
	   (qrystr   (conc "INSERT INTO runs (id,runname,"keystr") VALUES (?,?,"key?str")"))
	   (get-var  (lambda (db qrystr)
		       (let* ((res #f))
			 (sqlite3:for-each-row
			  (lambda row
			    (set res (car row)))
			  db qrystr runname)
			  db qrystr run-id runname)
			 res))))
      (if (null? runs)
        (begin
	  (db:create-initial-run-record dbstruct runname target))
      (let* ((run-id (db:get-run-id dbstruct runname target)))
	(db:with-db
	  (db:create-initial-run-record dbstruct run-id runname target)
        )
      )
      (let* ()
        ;;(debug:print 0 *default-log-port* "db:insert-run: Calling db:with-db to update the run record")
        (debug:print 0 *default-log-port* "db:insert-run: runid = " run-id)
#; (db:with-db
	 dbstruct
	 #f #t
	 (lambda (dbdat db)
           (debug:print 0 *default-log-port* "In the lambda proc for " dbdat " " db)
	   (for-each
	    (lambda (keyval)
              (debug:print 0 *default-log-port* "In the lambda proc for " keyval)
	      (let* ((fieldname (car keyval))
		     (getqry    (conc "SELECT "fieldname" FROM runs WHERE id=?;"))
		     (setqry    (conc "UPDATE runs SET "fieldname"=? WHERE id=?;"))
		     (val       (cdr keyval))
		     (valnum    (if (number? val)
				    val
				    (if (string? val)
					(string->number val)
					#f))))
                (debug:print 0 *default-log-port* "fieldname " fieldname " val " val " valnum " valnum)
		(if (not (member fieldname (cons "runname" keys))) ;; don't attempt to tweak these
		    (let* ((curr-val (get-var db getqry))
			   (have-it  (or (equal? curr-val val)
					 (equal? curr-val valnum))))
                      (debug:print 0 *default-log-port* "have-it = " have-it)
		      (if (not have-it)
                        (begin
                          (debug:print 0 *default-log-port* "Do sqlite3:execute")
			  (sqlite3:execute db setqry (or valnum val) run-id))))))
			  ;; (sqlite3:execute db setqry (or valnum val) run-id)
                        )
                      )
                    )
                 )
                (debug:print 0 *default-log-port* "Done with update")
              )
              (debug:print 0 *default-log-port* "next keyval")
            )
	    run-meta)))
	run-id))))
  
(define (db:create-initial-run-record dbstruct runname target)	  
(define (db:create-initial-run-record dbstruct run-id runname target)	  
  (let* ((keys     (db:get-keys dbstruct))
     	 (targvals (string-split target "/"))
	 (keystr   (string-intersperse keys ","))
	 (key?str  (string-intersperse (make-list (length targvals) "?") ","))
	 (qrystr   (conc "INSERT INTO runs (runname,"keystr") VALUES (?,"key?str")")))
	 (key?str  (string-intersperse (make-list (length targvals) "?") ",")) ;; a string with the same length as targvals, where each element is "?" and interspersed with commas.
	 (qrystr   (conc "INSERT INTO runs (id,runname,"keystr") VALUES (?,?,"key?str")")))
    (debug:print 0 *default-log-port* "db:create-initial-run-record")
    (debug:print 0 *default-log-port* "qrystr = " qrystr)

    (db:with-db
     dbstruct #f #t
     dbstruct #f #t ;; run-id writable
     (lambda (dbdat db)
       (debug:print 0 *default-log-port* "lambda proc: dbdat: " dbdat " db: " db) 
       (apply sqlite3:execute db qrystr runname targvals)))))
       (apply sqlite3:execute db qrystr run-id runname targvals)))))

(define (db:insert-test dbstruct run-id test-rec)
  (let* ((testname  (alist-ref "testname" test-rec equal?))
	 (item-path (alist-ref "item_path" test-rec equal?))
	 (id        (db:get-test-id dbstruct run-id testname item-path))
	 (fieldvals (filter (lambda (x)(not (member (car x) '("id" "last_update")))) test-rec))
	 (setqry    (conc "UPDATE tests SET "(string-intersperse
					      (map (lambda (dat)
						     (conc (car dat)"=?"))
						   fieldvals)
					      ",")" WHERE id=?;"))
	 (insqry   (conc "INSERT INTO tests ("(string-intersperse (map (lambda (x) (car x)) fieldvals) ",")
			 ") VALUES ("(string-intersperse (make-list (length fieldvals) "?") ",")");")))
    (debug:print 0 *default-log-port* "id: "id"\nset: "setqry"\ninsqry: "insqry)
    ;; (debug:print 0 *default-log-port* "id: "id"\nset: "setqry"\ninsqry: "insqry)
    (db:with-db
     dbstruct
     run-id #t
     (lambda (dbdat db)
       (if id
	   (apply sqlite3:execute db setqry (append (map cdr fieldvals) (list id)))
	   (apply sqlite3:execute db insqry (map cdr fieldvals)))))))

		   '()
		   db 
		   qry
		   run-id
		   (or last-update 0))))))

(define (db:get-testinfo-state-status dbstruct run-id test-id)
  (let ((res            #f))
    (db:with-db dbstruct run-id #f
		(lambda (dbdat db)
		  (sqlite3:for-each-row
		   (lambda (run-id testname item-path state status)
		     ;; id,run_id,testname,state,status,event_time,host,cpuload,diskfree,uname,rundir,item_path,run_duration,final_logf,comment
		     (set! res (vector test-id run-id testname state status -1 "" -1 -1 "" "-" item-path -1 "-" "-")))
		   db 
		   "SELECT run_id,testname,item_path,state,status FROM tests WHERE id=? and run_id=?;" 
		   test-id run-id)))
    res))
  (db:with-db
   dbstruct run-id #f
   (lambda (dbdat db)
     (let* ((res   #f)
	    (stmth (db:get-cache-stmth dbdat db "SELECT run_id,testname,item_path,state,status FROM tests WHERE id=? and run_id=?;")))
       (sqlite3:for-each-row
	(lambda (run-id testname item-path state status)
	  ;; id,run_id,testname,state,status,event_time,host,cpuload,diskfree,uname,rundir,item_path,run_duration,final_logf,comment
	  (set! res (vector test-id run-id testname state status -1 "" -1 -1 "" "-" item-path -1 "-" "-")))
	;; db 
	;; "SELECT run_id,testname,item_path,state,status FROM tests WHERE id=? and run_id=?;"
	stmth
	test-id run-id)
       res))))

;; get a useful subset of the tests data (used in dashboard
;; use db:mintest-get-{id ,run_id,testname ...}
;;
(define (db:get-tests-for-run-mindata dbstruct run-id testpatt states statuses not-in)
  (db:get-tests-for-run dbstruct run-id testpatt states statuses #f #f not-in #f #f "id,run_id,testname,state,status,event_time,item_path" 0 #f))

  (db:general-call dbstruct run-id 'delete-test-data-records (list test-id))
  (db:with-db
   dbstruct run-id #t
   (lambda (dbdat db)
     (sqlite3:execute db "UPDATE tests SET state='DELETED',status='n/a',comment='' WHERE id=?;" test-id))))

;; 
(define (db:delete-old-deleted-test-records dbstruct)
  (let ((targtime (- (current-seconds)
		     (or (configf:lookup-number *configdat* "setup" "keep-deleted-records")
			 (* 30 24 60 60))))) ;; one month in the past
(define (db:delete-old-deleted-test-records dbstruct run-id)
  (let* ((targtime (- (current-seconds)
		      (or (configf:lookup-number *configdat* "setup" "keep-deleted-records")
			  (* 7 24 60 60)))) ;; cleanup if over one week old
    (db:with-db
     dbstruct
	 (mtdbfile (dbmod:run-id->full-dbfname dbstruct run-id))
     0
     #t
     (lambda (dbdat db)
       (sqlite3:with-transaction
	db
	(lambda ()
	  (sqlite3:execute db "DELETE FROM test_steps WHERE test_id IN (SELECT id FROM tests WHERE state='DELETED' AND event_time<?);" targtime)
	  (sqlite3:execute db "DELETE FROM test_data WHERE test_id IN (SELECT id FROM tests WHERE state='DELETED' AND event_time<?);" targtime)
	  (sqlite3:execute db "DELETE FROM tests WHERE state='DELETED' AND event_time<?;" targtime)))))))
	 (qry1     "DELETE FROM test_steps WHERE test_id IN (SELECT id FROM tests WHERE state='DELETED' AND event_time<?);")
	 (qry2     "DELETE FROM test_data WHERE test_id IN (SELECT id FROM tests WHERE state='DELETED' AND event_time<?);")
	 (qry3     "DELETE FROM tests WHERE state='DELETED' AND event_time<?;")
	 (delproc  (lambda (db)
		     (sqlite3:with-transaction
		      db
		      (lambda ()
			(sqlite3:execute db qry1 targtime)
			(sqlite3:execute db qry2 targtime)
			(sqlite3:execute db qry3 targtime))))))
    ;; first the /tmp db
    (db:with-db
     dbstruct
     run-id
     #t
     (lambda (dbdat db)
       (delproc db)))
    (if (and (file-exists? mtdbfile)
	     (file-write-access? mtdbfile))
	(let* ((db (sqlite3:open-database mtdbfile)))
	  (delproc db)
	  (sqlite3:finalize! db)))))

;; 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)

;; 
(define (db:get-test-state-status-by-id dbstruct run-id test-id)
  (db:with-db
   dbstruct
   run-id
   #f
   (lambda (dbdat db)
     (let ((res   (cons #f #f)))
;;	   (stmth (db:get-cache-stmth dbdat db "SELECT state,status FROM tests WHERE id=?;")))
       (sqlite3:for-each-row ;; attemptnum added to hold pid of top process (not Megatest) controlling a test
	(lambda (state status)
	  (cons state status))
	db
	"SELECT state,status FROM tests WHERE id=? AND run_id=?;" ;; stmth try not compiling this one - yes, this fixed the bind issue
	test-id run-id)
     (let ((res   (cons #f #f))
	   (stmth (db:get-cache-stmth dbdat db "SELECT state,status FROM tests WHERE id=? AND run_id=?;")))
	  (sqlite3:for-each-row ;; attemptnum added to hold pid of top process (not Megatest) controlling a test
	   (lambda (state status)
	     (cons state status))
	   ;; db
	   stmth ;;"SELECT state,status FROM tests WHERE id=? AND run_id=?;" ;; stmth try not compiling this one - yes, this fixed the bind issue
	   test-id run-id)
       res))))

;; Use db:test-get* to access
;; Get test data using test_ids. NB// Only works within a single run!!
;;
(define (db:get-test-info-by-ids dbstruct run-id test-ids)
  (db:with-db

				(delete-duplicates
				 (cons testname (hash-table-ref/default res tag '())))))
	     tags)))
	db
	"SELECT testname,tags FROM test_meta")
       (hash-table->alist res)))))

;; testmeta doesn't change, we can cache it for up too an hour

(define *db:testmeta-cache* (make-hash-table))
(define *db:testmeta-last-update* 0)

;; read the record given a testname
(define (db:testmeta-get-record dbstruct testname)
  (if (and (< (- (current-seconds) *db:testmeta-last-update*) 600)
	   (hash-table-exists? *db:testmeta-cache* testname))
      (hash-table-ref *db:testmeta-cache* testname)
  (let ((res   #f))
    (db:with-db
     dbstruct
     #f
     #f
     (lambda (dbdat db)
       (sqlite3:for-each-row
	(lambda (id testname author owner description reviewed iterated avg_runtime avg_disk tags jobgroup)
	  (set! res (vector id testname author owner description reviewed iterated avg_runtime avg_disk tags jobgroup)))
	db
	"SELECT id,testname,author,owner,description,reviewed,iterated,avg_runtime,avg_disk,tags,jobgroup FROM test_meta WHERE testname=?;"
	testname)
       res))))
      (let ((res   #f))
	(db:with-db
	 dbstruct
	 #f
	 #f
	 (lambda (dbdat db)
	   (sqlite3:for-each-row
	    (lambda (id testname author owner description reviewed iterated avg_runtime avg_disk tags jobgroup)
	      (set! res (vector id testname author owner description reviewed iterated avg_runtime avg_disk tags jobgroup)))
	    db
	    "SELECT id,testname,author,owner,description,reviewed,iterated,avg_runtime,avg_disk,tags,jobgroup FROM test_meta WHERE testname=?;"
	    testname)))
	(hash-table-set! *db:testmeta-cache* testname res)
	(set! *db:testmeta-last-update* (current-seconds))
	res)))

;; create a new record for a given testname
(define (db:testmeta-add-record dbstruct testname)
  (db:with-db dbstruct #f #t
	      (lambda (dbdat db)
		(sqlite3:execute 
		 db

		     (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) )))) 
))


(define (std-exit-procedure)
  ;;(common:telemetry-log-close)
  (on-exit (lambda () 0)) ;; why is this here?
  ;;(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

(declare (unit dbfile))
(declare (uses debugprint))
(declare (uses commonmod))

(module dbfile
	*
(import scheme)
	
  (import scheme
	  chicken
	  	
(cond-expand
 (chicken-4
  
  (import chicken
	  data-structures
	  extras
	  matchable
  
	  (prefix sqlite3 sqlite3:)
	  posix posix-extras typed-records

	  srfi-18
	  srfi-1
	  srfi-69
	  stack
	  files
	  ports
	  hostinfo
	  
	  commonmod
	  debugprint
	  )

  )
 (chicken-5
  (import (prefix sqlite3 sqlite3:)
	  ;; data-structures
	  ;; extras
	  ;; files
	  ;; posix
	  ;; posix-extras
	  chicken.base
	  chicken.condition
	  chicken.file
	  chicken.file.posix
	  chicken.format
	  chicken.io
	  chicken.pathname
	  chicken.port
	  chicken.process
	  chicken.process-context
	  chicken.process-context.posix
	  chicken.sort
	  chicken.string
	  chicken.time
	  chicken.time.posix
	  
	  matchable
	  md5
	  message-digest
	  pathname-expand
	  regex
	  regex-case
	  srfi-1
	  srfi-18
	  srfi-69
	  typed-records
	  stack
	  system-information

	  commonmod
	  debugprint
	  )
  (define file-write-access? file-writable?)
  (define file-move move-file)
  ))
  
;; parameters
;;
(define dbfile:testsuite-name (make-parameter #f))

(define keep-age-param        (make-parameter 10))      ;; qif file age, if over move to attic
(define num-run-dbs           (make-parameter 10))      ;; number of db's in .mtdb
(define dbfile:sync-method    (make-parameter 'attach)) ;; 'attach or 'original

;; just the filename
(define (dbfile:run-id->dbfname run-id)
  (conc (dbfile:run-id->dbnum run-id)".db"))

;; the path in MTRAH with the filename
(define (dbfile:run-id->dbname run-id)
  (conc ".mtdb/"(dbfile:run-id->dbfname run-id)))
  (conc (dbfile:run-id->dbfname run-id)))


;; Make the dbstruct, setup up auxillary db's and call for main db at least once
;;
;; called in http-transport and replicated in rmt.scm for *local* access. 
;;
(define (dbfile:setup do-sync areapath tmppath)
(define (dbfile:setup areapath tmppath)
  (cond
   (*dbstruct-dbs*
    (dbfile:print-err "WARNING: dbfile:setup called when *dbstruct-dbs* is already initialized")
    *dbstruct-dbs*) ;; TODO: when multiple areas are supported, this optimization will be a hazard
   (else
    (let* ((dbstruct (make-dbr:dbstruct areapath: areapath tmppath: tmppath)))
      (set! *dbstruct-dbs* dbstruct)

	  (dbfile:open-db dbstruct run-id init-proc))
	(let* ((dbdat (dbfile:get-dbdat dbstruct run-id)))
	  (if dbdat
	      dbdat
	      (let* ((tmppath   (dbr:dbstruct-tmppath  dbstruct))
		     (tmpdbpath (dbfile:run-id->path tmppath run-id))
		     (dbdat     (dbfile:open-sqlite3-db tmpdbpath init-proc sync-mode: 0 journal-mode: "WAL")))
		
		;; the following line short-circuits the "one db handle per thread" model
               ;; the following line short-circuits the "one db handle per thread" model
		;; 
		;; (dbfile:add-dbdat dbstruct run-id dbdat)
		;;
		dbdat))))))
    
;; COMBINE dbfile:open-sqlite-db and dbfile:lock-create-open
;;

                            (begin
			      (if (file-exists? fname )
                                  (let ((db (sqlite3:open-database fname)))
				    ;; pragmas synchronous not needed because this db is used read-only
				    ;; (sqlite3:execute db (conc "PRAGMA synchronous = "mode";")
				    (sqlite3:set-busy-handler! db (sqlite3:make-busy-timeout 30000)) ;; read-only but still need timeout
				    db )
                                  (print "file doesn't exist: " fname))))
                                  (print "cautious-open-database: file doesn't exist: " fname))))
			(exn (io-error)
			     (dbfile:print-err exn "ERROR: i/o error with " fname ". Check permissions, disk space etc. and try again.")
			     (retry))
			(exn (corrupt)
			     (dbfile:print-err exn "ERROR: database " fname " is corrupt. Repair it to proceed.")
			     (retry))
			(exn (busy)

                                 dbname TEXT,
                                 mtversion TEXT,
                                 reason TEXT DEFAULT 'none',
                                   CONSTRAINT no_sync_processes UNIQUE (host,pid));"
			    ))))))
	 (on-tmp      (equal? (car (string-split dbpath "/")) "tmp"))
	 (db        (if on-tmp
			(dbfile:cautious-open-database dbname init-proc 0 "WAL" force-init: #t)
			(dbfile:cautious-open-database dbname init-proc 0 #f    force-init: #t)
			(dbfile:cautious-open-database dbname init-proc 1 "WAL" force-init: #t) ;; WAL MODE should use syncronous=1
			;; (dbfile:cautious-open-database dbname init-proc 0 #f    force-init: #t)
			(dbfile:cautious-open-database dbname init-proc 0 "MEMORY" force-init: #t) ;; Journal mode = memory is fastest?
			;; (sqlite3:open-database dbname)
			)))
    (if on-tmp	      ;; done in cautious-open-database
	(begin
	  (sqlite3:execute db "PRAGMA synchronous = 0;")
	  (sqlite3:set-busy-handler! db (sqlite3:make-busy-timeout 136000))))
    ;; (if on-tmp	      ;; done in cautious-open-database
    ;;    (begin
    ;; (sqlite3:execute db "PRAGMA synchronous = 0;") ;; why was this here when is is handled by cautious-open-database?
    (sqlite3:set-busy-handler! db (sqlite3:make-busy-timeout 136000)) ;; ))
    db))

;; mtest processes registry calls

(define (dbfile:insert-or-update-process nsdb dat)
  (let* ((host      (procinf-host dat))
	 (pid       (procinf-pid  dat))

(define (dbfile:register-process nsdb host port pid starttime endtime status purpose dbname mtversion)
  (sqlite3:execute nsdb "INSERT INTO processes (host,port,pid,starttime,endtime,status,purpose,dbname,mtversion) VALUES (?,?,?,?,?,?,?,?,?);"
		   host port pid starttime endtime status purpose dbname mtversion))

(define (dbfile:set-process-status nsdb host pid newstatus)
  (sqlite3:execute nsdb "UPDATE processes SET status=? WHERE host=? AND pid=?;" newstatus host pid))

;; as sorted should be stable. can use to choose "winner"
;;
(define (dbfile:get-process-options nsdb purpose dbname)
  (sqlite3:fold-row
   ;; host port pid starttime status mtversion
   (lambda (res . row)
     (cons row res))
   '()
   nsdb
   "SELECT host,port,pid,starttime,endtime,status,mtversion FROM processes WHERE purpose=? AND dbname LIKE ? AND status='alive';"
   "SELECT host,port,pid,starttime,endtime,status,mtversion FROM processes WHERE purpose=? AND dbname LIKE ? AND status IN ('running','alive') ORDER BY starttime ASC,host,port;"
   purpose dbname))

(define (dbfile:get-process-info nsdb host pid)
  (let ((res (sqlite3:fold-row
	      ;; host port pid starttime status mtversion
	      (lambda (res . row)
		(cons row res))
	      '()
	      nsdb
	      "SELECT host,port,pid,starttime,endtime,status,purpose,dbname,mtversion FROM processes WHERE host=? AND pid=?;"
	      host pid)))
    (if (null? res)
	#f
	(car res))))

(define (dbfile:row->procinf row)
  (match row
   ((host port pid starttime endtime status mtversion)
    (make-procinf host: host port: port pid: pid starttime: starttime endtime: endtime status: status mtversion: mtversion))
   (else
    (debug:print 0 *default-log-port* "ERROR: row "row" did not match host,port,pid,starttime,endtime,status,mtversion")
    #f)))

(define (dbfile:set-process-done nsdb host pid reason)
  (sqlite3:execute nsdb "UPDATE processes SET status='ended',endtime=?,reason=? WHERE host=? AND pid=?;" (current-seconds) reason host pid)
  (sqlite3:execute nsdb "UPDATE processes SET status='done',endtime=?,reason=? WHERE host=? AND pid=?;" (current-seconds) reason host pid)
  (dbfile:cleanup-old-entries nsdb))

(define (dbfile:cleanup-old-entries nsdb)
  (sqlite3:execute nsdb "DELETE FROM process WHERE status='ended' AND endtime<?;" (- (current-seconds) (* 3600 48))))
  (sqlite3:execute nsdb "DELETE FROM process WHERE status='done' AND endtime<?;" (- (current-seconds) (* 3600 48))))

;; other no-sync functions

(define (dbfile:with-no-sync-db dbpath proc)
  (mutex-lock! *no-sync-db-mutex*)
  (let* ((already-open *no-sync-db*)
	 (db  (or already-open (dbfile:raw-open-no-sync-db dbpath)))

(declare (uses dbfile))
(declare (uses commonmod))
(declare (uses debugprint))

(module dbmod
	*
	
(import scheme
	chicken
	data-structures
	extras
	files
(import scheme)
	
(cond-expand
 (chicken-4
  (import chicken
	  data-structures
	  extras
	  files

	  posix

	  ))
 (chicken-5
  (import chicken.base
	  chicken.condition
	  chicken.file
	  chicken.pathname
	  chicken.process
	  chicken.sort
	  chicken.string
	  chicken.time
	  
	  )
  (define file-read-access? file-readable?)
  (define file-copy copy-file)
  ))

(import	format
	(prefix sqlite3 sqlite3:)
	matchable
	posix
	typed-records
	srfi-1
	srfi-18
	srfi-69

	commonmod
	dbfile
	debugprint
	debugprint)
	)

;; NOTE: This returns only the name "1.db", "main.db", not the path
;;
(define (dbmod:run-id->dbfname run-id)
  (conc (dbfile:run-id->dbnum run-id)".db"))

(define (dbmod:get-dbdir dbstruct)
  (let* ((areapath (dbr:dbstruct-areapath dbstruct))
	 (dbdir    (conc areapath"/.mtdb")))
    (if (and (file-write-access? areapath)
	     (not (file-exists? dbdir)))
	(create-directory dbdir))
    dbdir))

(define (dbmod:run-id->full-dbfname dbstruct run-id)
  (conc (dbmod:get-dbdir dbstruct
  (conc (dbmod:get-dbdir dbstruct)

			 run-id

			 )"/"(dbmod:run-id->dbfname run-id)))
	"/"(dbmod:run-id->dbfname run-id)))

;;======================================================================
;; Read-only cachedb cached direct from disk method
;;======================================================================

(define *dbmod:nfs-db-handles* (make-hash-table)) ;; dbfname -> dbstruct

;;======================================================================
;; The cachedb one-db file per server method goes in here
;;======================================================================

;; NOTE: the r/w is now w/r, #t=db modified by query, #f=db NOT modified by query
(define (dbmod:with-db dbstruct run-id w/r proc params)
  (let* ((use-mutex (or (and w/r ;; use the mutex on queries that modify the db and for sync to disk
			     (> *api-process-request-count* 5)) ;; when writes are happening throttle more
			(> *api-process-request-count* 50)))
  (let* ((use-mutex w/r) ;; (or (and w/r ;; use the mutex on queries that modify the db and for sync to disk
			     ;; (> *api-process-request-count* 5)) ;; when writes are happening throttle more
			;; (> *api-process-request-count* 50)))
	 (dbdat     (dbmod:open-db dbstruct run-id (dbfile:db-init-proc)))
	 (dbh       (dbr:dbdat-dbh dbdat)) ;; this will be the cachedb handle
	 (dbfile    (dbr:dbdat-dbfile dbdat)))
    ;; if nfs mode do a sync if delta > 2
    (let* ((last-update (dbr:dbstruct-last-update dbstruct))
	   (sync-proc   (dbr:dbstruct-sync-proc dbstruct))
    #;(let* ((last-update (dbr:dbstruct-last-update dbstruct))
	   ;; (sync-proc   (dbr:dbstruct-sync-proc dbstruct))
	   (curr-secs   (current-seconds)))
      (if (> (- curr-secs last-update) 5)
	  (begin
	    (sync-proc last-update)

	    ;; MOVE THIS CALL TO INSIDE THE sync-proc CALL
	    (dbr:dbstruct-last-update-set! dbstruct curr-secs)

			      (debug:print-info 0 *default-log-port* "dbmod:with-db, database is busy, will try "count" more times.")
			      (thread-sleep! 1)
			      (loop (- count 1)))
			    (begin
			      (debug:print-info 0 *default-log-port* "dbmod:with-db, database is busy, giving up.")
			      (exit 1))))
		   (exn ()
			(dbfile:print-err exn "ERROR: Unknown error with db for run-id "
			(dbfile:print-err exn "ERROR: dbmod:with-db: Unknown error with database for run-id "run-id", message: "
					  run-id", message: "
					  ((condition-property-accessor 'exn 'message) exn)
					  ", details: "(condition->list exn))
			(exit 2))))))
      (if use-mutex (mutex-unlock! *db-with-db-mutex*))
      res)))

    (dbr:dbstruct-dbfile-set!    dbstruct dbfullname)
    (dbr:dbstruct-dbtmpname-set! dbstruct tmpdb)
    (dbr:dbstruct-dbfname-set!   dbstruct dbfname)
    (dbr:dbstruct-sync-proc-set! dbstruct
				 (lambda (last-update)
				   (if *sync-in-progress*
				       (debug:print 0 *default-log-port* "WARNING: overlapping calls to sync to disk")
				       (let* ((syncer-logfile    (conc areapath"/logs/"dbfname"-syncer.log"))
								       
					      (sync-cmd          (if (eq? syncdir 'todisk)
								     (conc "(NBFAKE_LOG="syncer-logfile" nbfake megatest -db2db -from "tmpdb" -to "dbfullname" -period 5 -timeout 10 > /dev/null 2&>1)&")
								     (conc "(NBFAKE_LOG="syncer-logfile" nbfake megatest -db2db -from "dbfullname" -to "tmpdb" -period 5 -timeout 10 > /dev/null 2&>1)&")))
					      (synclock-file     (conc dbfullname".lock"))
					      (syncer-running-file (conc dbfullname"-sync-running"))
					      (synclock-mod-time (if (file-exists? synclock-file)
								     (handle-exceptions
									 exn
								       #f
								       (file-modification-time synclock-file))
								     #f))
					      (thethread         (lambda ()
								   (thread-start!
								    (make-thread
								     (lambda ()
								       (set! *sync-in-progress* #t)
								       (debug:print 0 *default-log-port* "Running "sync-cmd)
								       (if (file-exists? syncer-running-file)
									   (debug:print-info 0 *default-log-port* "Syncer still running, skipping syncer start.")
									   (system sync-cmd))
								       (set! *sync-in-progress* #f)))))))
					 (if ((if (eq? syncdir 'todisk) < >) ;; use less than for todisk, greater than for from disk
					      (file-modification-time tmpdb)
					      (file-modification-time dbfullname))
					     (debug:print 4 *default-log-port* "Skipping sync, "tmpdb" older than "dbfullname)
					     (if synclock-mod-time
						 (if (> (- (current-seconds) synclock-mod-time) 20) ;; something wrong with sync, remove file
						     (begin
				       (begin
						       (handle-exceptions
							   exn
							 #f
                                                         (begin
                                                           (debug:print 0 *default-log-port* "Sync lock file " synclock-file "is older than 20 seconds ("  synclock-mod-time " seconds). Removing it")
                                         ;; turn off writes - send busy or block?
					 ;; call db2db internally
					 ;; turn writes back on
					 ;;
					 (set! *api-halt-writes* #t) ;; do we need a mutex?
					 ;; (dbmod:db-to-db-sync src-db dest-db last-update (dbfile:db-init-proc) keys)
					 (debug:print-info 2 *default-log-port* "Internal sync running from "tmpdb" to "dbfullname)
							   (delete-file synclock-file)
                                                         )
					 (dbmod:db-to-db-sync tmpdb dbfullname last-update (dbfile:db-init-proc) keys)
					 (set! *api-halt-writes* #f)
					 ))))
                                                       )
						       (thethread))
						     (debug:print 0 *default-log-port* "Skipping sync, lockfile "synclock-file" found."))
						 (thethread)))))))
    ;; (dbmod:sync-tables tables #f db cachedb)
    ;; 
    (thread-sleep! 1) ;; let things settle before syncing in needed data
    (dbmod:sync-gasket tables #f cachedb db dbfullname 'fromdest keys) ;; ) ;; load into cachedb
    (dbr:dbstruct-last-update-set! dbstruct (+ (current-seconds) -10)) ;; should this be offset back in time by one second?
    dbstruct))

		  (stmt3    (conc "UPDATE "table" SET ("no-id-fields-str") = ("questionmarks") WHERE id=?;"))
		  (start-ms (current-milliseconds)))
	     (debug:print 0 *default-log-port* "stmt3="stmt3)
	     (if (sqlite3:auto-committing? dbh1)
		 (begin
       	           (handle-exceptions
		     exn
		     (debug:print 0 *default-log-port* "Transaction update of "table" failed.")
		     (debug:print 0 *default-log-port* "Transaction update of "table" failed. "(condition->list exn))
		     (sqlite3:with-transaction
		      dbh1
		      (lambda ()
		        (sqlite3:execute dbh1 stmt1)    ;; get all new rows

		      #;(if (member "last_update" fields)
			  (sqlite3:execute dbh1 stmt8))    ;; get all updated rows

		   (sdb    (dbmod:safely-open-db src-db init-proc #t))
		   (ddb    (dbmod:safely-open-db dest-db init-proc d-wr))
		   (res    (dbmod:sync-gasket tables last-update sdb ddb dest-db 'todisk keys)))
	      (sqlite3:finalize! sdb)
	      (sqlite3:finalize! ddb)
	      res)))
      #f))
)

;; ======================================================================
;; dbstats
;;======================================================================

;; (define *dbstruct-dbs*         #f) ;; used to cache the dbstruct in db:setup. Goal is to remove this.
;; db stats
(define *db-stats*            (make-hash-table)) ;; hash of vectors < count duration-total >
(define *db-stats-mutex*      (make-mutex))

(define (rmt:print-db-stats)
  (let ((fmtstr "~40a~8-d~20-d~20,2-f")) ;; "~20,2-f"
    (debug:print 0 *default-log-port* "DB Stats\n========")
    (debug:print 0 *default-log-port* (format #f "~40a~8a~20a~10a" "Cmd" "Count" "TotTime" "Avg"))
    (for-each (lambda (cmd)
		(let* ((dat     (hash-table-ref *db-stats* cmd))
		       (count   (dbstat-cnt dat))
		       (tottime (dbstat-tottime dat)))
		  (debug:print 0 *default-log-port*
			       (format #f fmtstr cmd count tottime
				       (/ tottime count)))))
	      (sort (hash-table-keys *db-stats*)
		    (lambda (a b)
		      (> (dbstat-tottime (hash-table-ref *db-stats* a))
			 (dbstat-tottime (hash-table-ref *db-stats* b))))))))

(defstruct dbstat
  (cnt 0)
  (tottime 0))

(define (db:add-stats cmd run-id params delta)
  (let* ((modified-cmd (if (eq? cmd 'general-call)
			   (string->symbol (conc "general-call-" (car params)))
			   cmd))
	 (rec          (hash-table-ref/default *db-stats* modified-cmd #f)))
    (if (not rec)
	(let ((new-rec  (make-dbstat)))
	  (hash-table-set! *db-stats* modified-cmd new-rec)
	  (set! rec new-rec)))
    (dbstat-cnt-set! rec (+ (dbstat-cnt rec) 1))
    (dbstat-tottime-set! rec (+ (dbstat-tottime rec) delta))))
    


)


;; ATTIC

					 #;(let* ((syncer-logfile    (conc areapath"/logs/"dbfname"-syncer.log"))
					      (sync-cmd          (if (eq? syncdir 'todisk)
								     (conc "(NBFAKE_LOG="syncer-logfile" nbfake megatest -db2db -from "tmpdb" -to "dbfullname" -period 5 -timeout 10 > /dev/null 2&>1)&")
								     (conc "(NBFAKE_LOG="syncer-logfile" nbfake megatest -db2db -from "dbfullname" -to "tmpdb" -period 5 -timeout 10 > /dev/null 2&>1)&")))
					      (synclock-file     (conc dbfullname".lock"))
					      (syncer-running-file (conc dbfullname"-sync-running"))
					      (synclock-mod-time (if (file-exists? synclock-file)
								     (handle-exceptions
									 exn
								       #f
								       (file-modification-time synclock-file))
								     #f))
					      (thethread         (lambda ()
								   (thread-start!
								    (make-thread
								     (lambda ()
								       (set! *sync-in-progress* #t)
								       (debug:print-info "Running "sync-cmd)
								       (if (file-exists? syncer-running-file)
									   (debug:print-info 0 *default-log-port* "Syncer still running, skipping syncer start.")
									   (system sync-cmd))
								       (set! *sync-in-progress* #f)))))))
					 (if ((if (eq? syncdir 'todisk) < >) ;; use less than for todisk, greater than for from disk
					      (file-modification-time tmpdb)
					      (file-modification-time dbfullname))
					     (debug:print 4 *default-log-port* "Skipping sync, "tmpdb" older than "dbfullname)
					     (if synclock-mod-time
						 (if (> (- (current-seconds) synclock-mod-time) 20) ;; something wrong with sync, remove file
						     (begin
						       (handle-exceptions
							   exn
							 #f
                                                         (begin
                                                           (debug:print 0 *default-log-port* "Sync lock file " synclock-file "is older than 20 seconds ("  synclock-mod-time " seconds). Removing it")
							   (delete-file synclock-file)
                                                         )
                                                       )
						       (thethread))
						     (debug:print 0 *default-log-port* "Skipping sync, lockfile "synclock-file" found."))
						 (thethread))))

  please-update  
  tabdats
  update-mutex
  updaters 
  updating
  uidat ;; needs to move to tabdat at some time
  hide-not-hide-tabs
  target
  )

(define (dboard:commondat-make)
  (make-dboard:commondat
   curr-tab-num:         0
   tabdats:              (make-hash-table)
   please-update:        #t
   update-mutex:         (make-mutex)
   updaters:             (make-hash-table)
   updating:             #f
   hide-not-hide-tabs:   #f
   target:               ""
   ))

;; RADT => Matrix defstruct addition
(defstruct dboard:graph-dat
    ((id           #f) : string)
    ((color        #f) : vector)
    ((flag         #t) : boolean)
    ((cell         #f) : number)
    )

;; RA => returns the tabdat stored at hashkey passed in commondat-tabdats table (e.g. 0 gives summary)
;;
(define (dboard:common-get-tabdat commondat #!key (tab-num #f))
  (let* ((tnum (or tab-num
		   (dboard:commondat-curr-tab-num commondat)
		   0)) ;; tab-num value is curr-tab-num value in passed commondat
	 (ht   (dboard:commondat-tabdats commondat))
	 (res  (hash-table-ref/default ht tnum #f)))
    (or res
	(let ((new-tabdat  (dboard:tabdat-make-data)))
	  (hash-table-set! ht tnum new-tabdat)
	  new-tabdat))))

;; RA => sets the tabdat passed to the hashkey at commondat:tabdats hash table
;;
(define (dboard:common-set-tabdat! commondat tabnum tabdat)
  (hash-table-set!
   (dboard:commondat-tabdats commondat)
   tabnum
   tabdat))

;; data for runs, tests etc. was used in run summary?
;;
(defstruct dboard:runsdat
  ;; new system
  runs-index    ;; target/runname => colnum
  tests-index   ;; testname/itempath => rownum

	       #:value "megatest "
	       #:expand "HORIZONTAL"
	       #:readonly "YES"
	       #:font "Courier New, -12"
	       )))
      (dboard:tabdat-command-tb-set! data tb)
      tb)

    (iup:button "Execute" #:size "50x"
		#:action (lambda (obj)
			   ;; (let ((cmd (conc ;; "xterm -geometry 180x20 -e \""
                           (common:run-a-command (iup:attribute (dboard:tabdat-command-tb data) "VALUE")))))))
			 (let ((cmd (iup:attribute (dboard:tabdat-command-tb data) "VALUE")))
                           (if (substring-index "no-runname-specified" cmd)
                            (debug:print 0 *default-log-port* "ERROR: no runname specified")
                            (begin
                              (if (substring-index "no-target-selected" cmd)
                                (debug:print 0 *default-log-port* "ERROR: no target selected")
                                (begin
                                  (if (not (substring-index "-run" cmd))
                                    (debug:print 0 *default-log-port*  "ERROR: No target selected")
                                    (common:run-a-command (iup:attribute (dboard:tabdat-command-tb data) "VALUE"))
                                  )
                                )
                              )
                            )
                           )))))))
    ;; ";echo Press any key to continue;bash -c 'read -n 1 -s'\" &")))
    ;; (system cmd)))))))

(define (dcommon:command-action-selector commondat tabdat #!key (tab-num #f))
  (iup:frame
   #:title "Set the action to take"
   (iup:hbox

  (iup:frame
   #:title "Runname"
   (let* ((default-run-name (seconds->work-week/day (current-seconds)))
	  (tb (iup:textbox #:expand "HORIZONTAL"
			   #:action (lambda (obj val txt)
				      (debug:catch-and-dump
				       (lambda ()
					 ;; (print "obj: " obj " val: " val " unk: " unk)
					 (dboard:tabdat-run-name-set! tabdat txt) ;; (iup:attribute obj "VALUE"))
					 (dashboard:update-run-command tabdat))
				       "command-runname-selector tb action"))
			   #:value (or default-run-name (dboard:tabdat-run-name tabdat))))
	  (lb (iup:listbox #:expand "HORIZONTAL"
			   #:dropdown "YES"
			   #:action (lambda (obj val index lbstate)

					  (runs-for-targ (rmt:get-runs-by-patt (dboard:tabdat-keys tabdat) "%" #f #f #f #f 0))
					  (runs-header   (vector-ref runs-for-targ 0))
					  (runs-dat      (vector-ref runs-for-targ 1))
					  (run-names     (cons default-run-name 
							       (map (lambda (x)
								      (db:get-value-by-header x runs-header "runname"))
								    runs-dat))))
				     ;; (print "DEBUGINFO: run-names=" run-names)
				     ;; (iup:attribute-set! lb "REMOVEITEM" "ALL")
				     (iuplistbox-fill-list lb run-names selected-item: default-run-name))))))
     ;; (dboard:tabdat-updater-for-runs-set! tabdat refresh-runs-list)
     (dboard:commondat-add-updater commondat refresh-runs-list tab-num: tab-num)
     ;; (refresh-runs-list)
     (dboard:tabdat-run-name-set! tabdat default-run-name)
     (iup:hbox

megatest_manual.html : megatest_manual.txt  *.txt installation.txt *png *.dot
	asciidoc  -b html5 -a icons -a iconsdir=$(DISPATH)/images/icons -a toc2 megatest_manual.txt
#	dos2unix megatest_manual.html

megatest_manual.pdf : megatest_manual.txt *.txt *png *.dot
	a2x -a toc -f pdf megatest_manual.txt

%.pdf : %.dot
	dot -Tpdf $*.dot -o$*.pdf

server.ps : server.dot
	dot -Tps server.dot > server.ps

client.ps : client.dot
	dot -Tps client.dot > client.ps

complex-itemmap.png : complex-itemmap.dot

//     Megatest is distributed in the hope that it will be useful,
//     but WITHOUT ANY WARRANTY; without even the implied warranty of
//     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
//     GNU General Public License for more details.
// 
//     You should have received a copy of the GNU General Public License
//     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

digraph G {

    label = "Server Start Sequences";
    color=brown;
    rankdir="TB";

    subgraph cluster_1 {
	label="Find Prime Main Server";
	
        node [style=filled,shape=box];

	
	check_available_queue       -> remove_entries_over_10s_old;
	remove_entries_over_10s_old -> set_available [label="num_avail < 3"];
	remove_entries_over_10s_old -> exit [label="num_avail > 2"];

	set_available               -> delay_2s;
	delay_2s          -> check_place_in_queue;

	check_place_in_queue        -> "http:transport-launch" [label="at head"];
	check_place_in_queue        -> exit [label="not at head"];

	START;
        HaveServ [label="Look at .servinfo\nfiles for prime main"]; 
	AskPrime [label="Ask Prime for main"];
	PingPrime [label="Ping Prime"];
	AskPrime [label="Ask .servinfo prime for server"];
	StartServ [label="Launch Server Process for main.db"];
	
      	START -> HaveServ;
	HaveServ -> PingPrime;
	PingPrime -> AskPrime [label="Got response"];
	PingPrime -> StartServ [label="No reponse"];
	HaveServ -> StartServ [label="No files"];
	StartServ -> "Delay 2s" -> START;
        AskPrime -> DONE;
    }
	"client:login"              -> "server:shutdown" [label="login failed"];
	"server:shutdown"           -> exit;	

	subgraph cluster_2 {
    
    subgraph cluster_2 {
		"http:transport-launch"       -> "http:transport-run";
		"http:transport-launch"       -> "http:transport-keep-running";

	label="Starting non-prime server"
		"http:transport-keep-running" -> "tests running?";
		"tests running?"              -> "client:login" [label=yes];
        node [style=filled,shape=box];
	StartTCPServer [label="Start tcp server"];
		"tests running?"              -> "server:shutdown" [label=no];
		"client:login"                -> delay_5s [label="login ok"];
	FindPrimeMain [label="Find Prime Main Server"];
		delay_5s                      -> "http:transport-keep-running";
	}
	RegisterProcessViaPrime [label="Register process via prime server"];

	// start_server -> "server_running?";
	// "server_running?" -> set_available [label="no"];
	// "server_running?" -> delay_2s [label="yes"];
	// delay_2s -> "still_running?";
	// "still_running?" -> ping_server [label=yes];
	// "still_running?" -> set_available [label=no];
	// ping_server -> exit [label=alive];
	// ping_server -> remove_server_record [label=dead];
	StartTCPServer -> FindPrimeMain -> START;
	DONE -> RegisterProcessViaPrime -> READY;
    }

    subgraph cluster_3 {
	label="Start Prime Main"
        node [style=filled,shape=box];
	StartTCPServer_prime [label="Start tcp server"];
	GetServInfoFiles [label="Get servinfo files"];
	CreateServInfoFile [label="Create servinfo file"];
	RegisterProcess [label="Register process in no-sync (direct access)"];
	// remove_server_record -> set_available;
	// set_available -> avail_delay [label="delay 3s"];
	// avail_delay -> "first_in_queue?";
	ValidateServInfoFiles [label="Validate servinfo files with ping\nremove any files which do not respond to ping"];

	// 
	// "first_in_queue?" -> set_running [label=yes];
	CheckHost [label="Verify that current host matches\nexisting servinfo files host"]
	// set_running -> get_next_port -> handle_requests;
	// "first_in_queue?" -> "dead_entry_in_queue?" [label=no];
	// "dead_entry_in_queue?" -> "server_running?" [label=no];
	// "dead_entry_in_queue?" -> "remove_dead_entries" [label=yes];
	// remove_dead_entries -> "server_running?";
	StartTCPServer_prime -> GetServInfoFiles;
	GetServInfoFiles -> CreateServInfoFile [label="No servinfo\nfiles"];
	GetServInfoFiles -> ValidateServInfoFiles;
	// 
	// handle_requests -> start_shutdown [label="no traffic\nno running tests"];
	ValidateServInfoFiles -> CreateServInfoFile [label="No valid files"];
	CreateServInfoFile -> GetServInfoFiles [label="servinfo file created"];
	KeepRunning [label="READY"];
	// handle_requests -> shutdown_request;
	// start_shutdown -> shutdown_delay;
	// shutdown_request -> shutdown_delay;
	// shutdown_delay -> exit;
	
        label = "server:launch";
        color=brown;
		     
	ValidateServInfoFiles -> CheckHost;
	CheckHost -> RegisterProcess [label="Have valid\nservinfo files and same host"];
	RegisterProcess -> KeepRunning;
	CheckHost -> EXIT [label="Not same host"];	
    }

}
//     client_start_server -> start_server;
//     handle_requests -> read_write;
//     read_write -> handle_requests;
}

    ;; (tests:set-full-meta-info test-id run-id (calc-minutes) work-area)
    (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area 10)

    (let loop ((minutes   (calc-minutes))
	       (cpu-load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
	       (disk-free (get-df (current-directory)))
               (last-sync (current-seconds)))
      ;; (common:telemetry-log "zombie" (conc "launch:monitor-job - top of loop encountered at "(current-seconds)" with last-sync="last-sync))
      (let* ((over-time     (> (current-seconds) (+ last-sync update-period)))
             (new-cpu-load  (let* ((load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
                                   (delta (abs (- load cpu-load))))
                              (if (> delta 0.1) ;; don't bother updating with small changes
                                  load
                                  #f)))
             (new-disk-free (let* ((df    (if over-time ;; only get df every 30 seconds
                                              (get-df (current-directory))
                                              disk-free))
                                   (delta (abs (- df disk-free))))
                              (if (and (> df 0)
                                       (> (/ delta df) 0.1)) ;; (> delta 200) ;; ignore changes under 200 Meg
                                  df
                                  #f)))
             (do-sync       (or new-cpu-load new-disk-free over-time))

             (test-info   (rmt:get-test-state-status-by-id run-id test-id))
             (state       (car test-info));; (db:test-get-state test-info))
             (status      (cdr test-info));; (db:test-get-status test-info))
	     (killreq     (equal? state "KILLREQ"))
             (kill-reason  "no kill reason specified")
             (kill-job?    #f))
        ;; (common:telemetry-log "zombie" (conc "launch:monitor-job - decision time encountered at "(current-seconds)" with last-sync="last-sync" do-sync="do-sync" over-time="over-time" update-period="update-period))
        (cond
         ((test-get-kill-request run-id test-id)
         (killreq
          (set! kill-reason "KILLING TEST since received kill request (KILLREQ)")
          (set! kill-job? #t))
         ((and runtlim (> (- (current-seconds) start-seconds) runtlim))
          (set! kill-reason (conc "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" (- (current-seconds) start-seconds) " seconds, limit=" runtlim))
          (set! kill-job? #t))
         ((equal? status "DEAD")
          (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
          (rmt:set-state-status-and-roll-up-items run-id test-id 'foo "RUNNING" "n/a" "was marked dead; really still running.")
          ;;(set! kill-reason "KILLING TEST because it was marked as DEAD by launch:handle-zombie-tests (might indicate really overloaded server or else overzealous setup.deadtime)") ;; MARK RUNNING
          (set! kill-job? #f)))

        (debug:print 4 *default-log-port* "cpu: " new-cpu-load " disk: " new-disk-free " last-sync: " last-sync " do-sync: " do-sync)
        (if (common:low-noise-print 600 "run zombie") ;; every five minutes is plenty
	    (launch:handle-zombie-tests run-id))
        (when do-sync
          ;;(with-output-to-file (conc (getenv "MT_TEST_RUN_DIR") "/last-loadinfo.log" #:append)
          ;;  (lambda () (pp (list (current-seconds) new-cpu-load new-disk-free (calc-minutes)))))
          ;; (common:telemetry-log "zombie" (conc  "launch:monitor-job - dosync started at "(current-seconds)))
          (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
          (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f))
          ;; (common:telemetry-log "zombie" (conc "launch:monitor-job - dosync finished at "(current-seconds)))
	  )
        
	(if kill-job? 
	    (begin
              (debug:print-info 0 *default-log-port* "proceeding to kill test: "kill-reason)
	      (mutex-lock! m)
	      ;; NOTE: The pid can change as different steps are run. Do we need handshaking between this
	      ;;       section and the runit section? Or add a loop that tries three times with a 1/4 second

		      (debug:print-error 0 *default-log-port* "Nothing to kill, pid1=" pid1 ", pid2=" pid2)
		      (tests:test-set-status! run-id test-id "KILLED"  "FAILED TO KILL" (conc (args:get-arg "-m")" "kill-reason) #f) ;; BB ADDED kill-reason -- confirm OK with Matt
		      )))
	      (mutex-unlock! m)
	      ;; no point in sticking around. Exit now. But run end of run before exiting?
              (launch:end-of-run-check run-id)
	      (exit)))
	(if (hash-table-ref/default misc-flags 'keep-going #f)
	(if (hash-table-ref/default misc-flags 'keep-going #f)  ;; keep originals for cpu-load and disk-free unless they change more than the allowed delta
	    (begin
	      (thread-sleep! 3) ;; (+ 3 (random 6))) ;; add some jitter to the call home time to spread out the db accesses
	      (thread-sleep! 6) ;; was 3
	      (if (hash-table-ref/default misc-flags 'keep-going #f)  ;; keep originals for cpu-load and disk-free unless they change more than the allowed delta
		  (loop (calc-minutes)
                        (or new-cpu-load cpu-load)
                        (or new-disk-free disk-free)
                        (if do-sync (current-seconds) last-sync)))))))
    (tests:update-central-meta-info run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f))) ;; NOTE: Checking twice for keep-going is intentional
	      (loop (calc-minutes)
		    (or new-cpu-load cpu-load)
		    (or new-disk-free disk-free)
		    (if do-sync (current-seconds) last-sync))))))
    (tests:update-central-meta-info run-id test-id (commonmod:get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f))) ;; NOTE: Checking twice for keep-going is intentional


;; read testconfig and create .logpro and script files
;;    - use #f for tconfigreg to re-read the testconfigs from disk
;;
(define (launch:extract-scripts-logpro  test-dir test-name item-path tconfigreg-in)
  (let* ((tconfigreg      (or tconfigreg-in
			      (tests:get-all)))
	 (tconfig-fname   (conc test-dir "/.testconfig"))
	 (tconfig-tmpfile (conc tconfig-fname ".tmp"))
	 (tconfig         (tests:get-testconfig test-name item-path tconfigreg #t force-create: #t)) ;; 'return-procs)))
	 (scripts         (configf:get-section tconfig "scripts"))
	 (logpros         (configf:get-section tconfig "logpro")))
    ;; create .testconfig file
    (configf:write-alist tconfig tconfig-tmpfile)
    (file-move tconfig-tmpfile tconfig-fname #t)
    (delete-file* ".final-status")
    
    ;; extract scripts from testconfig and write them to files in test run dir
    (for-each
     (lambda (scriptdat)
       (match scriptdat
	      ((name content)
	       (debug:print-info 2 *default-log-port* "Creating script "(current-directory)"/"name)
	       (with-output-to-file name
		 (lambda ()
		   (print content)))
	       (change-file-mode name (bitwise-ior perm/irwxg perm/irwxu)))
	      (else
	       (debug:print-info 0 "Invalid script definiton found in [scripts] section of testconfig. \"" scriptdat "\""))))
     scripts)

    ;; extract logpro from testconfig and write them to files in test run dir
    (for-each
     (lambda (logprodat)
       (match logprodat
	      ((name content)
	       (debug:print-info 2 *default-log-port* "Creating logpro file "(current-directory)"/"name".logpro")
	       (with-output-to-file (conc name".logpro")
		 (lambda ()
		   (print content)
		   ;; (change-file-mode name (bitwise-ior perm/irwxg perm/irwxu))
		   )))
	      (else
	       (debug:print-info 0 "Invalid logpro definiton found in [logpro] section of testconfig. \"" logprodat "\""))))
     logpros)))

;; read testconfig and create .logpro and script files
;;    - use #f for tconfigreg to re-read the testconfigs from disk
;;
(define (launch:extract-scripts-logpro  test-dir test-name item-path tconfigreg-in)
  (let* ((tconfigreg      (or tconfigreg-in
			      (tests:get-all)))
	 (tconfig-fname   (conc test-dir "/.testconfig"))
	 (tconfig-tmpfile (conc tconfig-fname ".tmp"))
	 (tconfig         (tests:get-testconfig test-name item-path tconfigreg #t force-create: #t)) ;; 'return-procs)))
	 (scripts         (configf:get-section tconfig "scripts"))
	 (logpros         (configf:get-section tconfig "logpro")))
    ;; create .testconfig file
    (configf:write-alist tconfig tconfig-tmpfile)
    (file-move tconfig-tmpfile tconfig-fname #t)
    (delete-file* ".final-status")
    
    ;; extract scripts from testconfig and write them to files in test run dir
    (for-each
     (lambda (scriptdat)
       (match scriptdat
	      ((name content)
	       (debug:print-info 2 *default-log-port* "Creating script "(current-directory)"/"name)
	       (with-output-to-file name
		 (lambda ()
		   (print content)))
	       (change-file-mode name (bitwise-ior perm/irwxg perm/irwxu)))
	      (else
	       (debug:print-info 0 "Invalid script definiton found in [scripts] section of testconfig. \"" scriptdat "\""))))
     scripts)

    ;; extract logpro from testconfig and write them to files in test run dir
    (for-each
     (lambda (logprodat)
       (match logprodat
	      ((name content)
	       (debug:print-info 2 *default-log-port* "Creating logpro file "(current-directory)"/"name)
	       (with-output-to-file name
		 (lambda ()
		   (print content)
		   ;; (change-file-mode name (bitwise-ior perm/irwxg perm/irwxu))
		   )))
	      (else
	       (debug:print-info 0 "Invalid logpro definiton found in [logpro] section of testconfig. \"" logprodat "\""))))
     logpros)))

;; read testconfig and create .logpro and script files
;;    - use #f for tconfigreg to re-read the testconfigs from disk
;;
(define (launch:extract-scripts-logpro  test-dir test-name item-path tconfigreg-in)
  (let* ((tconfigreg      (or tconfigreg-in
			      (tests:get-all)))

;; load the ~/.megatestrc file, put (use trace)(trace-call-sites #t)(trace function-you-want-to-trace) in this file
;;
(let ((debugcontrolf (conc (get-environment-variable "HOME") "/.megatestrc")))
  (if (common:file-exists? debugcontrolf)
    (begin
      ;; for some reason, debug:print does not work here. Had to use print.
      (with-output-to-port (current-error-port)
	(lambda ()
      (print (conc "WARNING: loading " debugcontrolf))
	  (print (conc "WARNING: loading " debugcontrolf))))
      (load debugcontrolf)
    )
  )
)

;; usage logging, careful with this, it is not designed to deal with all real world challenges!
;;

			"-pathmod"
			"-env2file"
			"-envcap"
			"-envdelta"
			"-setvars"
			"-set-state-status"
			"-import-sexpr"
                        "-remove-dbs" ;; to be used only with -import-sexpr to remove megatest dbs first.
			"-period"  ;; sync period in seconds
			"-timeout" ;; exit sync if timeout in seconds exceeded since last change

                        ;; move runs stuff here
                        "-remove-keep"           
			"-set-run-status"
			"-age"

       (begin
        (debug:print 0 *default-log-port* "Attempt to start a server on a machine that is not the homehost. Aborting")
        (exit
        )))
      (case (rmt:transport-mode)
	((tcp)
	 (let* ((timeout    (server:expiration-timeout)))
	   (debug:print 0 *default-log-port* "INFO: Starting server for " dbfname " using tcp method with server timeout of "timeout)
	   (debug:print 0 *default-log-port* "INFO: megatest -server starting on " (get-host-name) " for " dbfname " using tcp method with timeout of "timeout)
	   (tt-server-timeout-param timeout)
	   (thread-start! (make-thread api:print-db-stats "print-db-stats"))
	   (if dbfname
	       (tt:start-server tl #f dbfname api:tcp-dispatch-request-make-handler keys)
	       (begin
		 (debug:print 0 *default-log-port* "ERROR: transport mode is tcp - -db is required.")
		 (exit 1)))))
	(else (debug:print 0 *default-log-port* "ERROR: rmt:transport-mode value not recognised "(rmt:transport-mode))))
      (set! *didsomething* #t)))

  )
)




(if (args:get-arg "-kill-servers")

  
  (let* ((tl (launch:setup)) ;; need this to initialize *toppath*
        (servdir (tt:get-servinfo-dir *toppath*))
        (servfiles (glob (conc servdir "/*:*.db")))
        (fmtstr  "~10a~22a~10a~25a~25a~8a\n")
        (dbfiles (append (glob (conc *toppath* "/.mtdb/main.db")) (glob (conc *toppath* "/.mtdb/?.db"))(glob (conc *toppath* "/.mtdb/??.db"))))
        (dbfiles (if (file-exists? (conc *toppath* "/.mtdb/main.db")) (append (glob (conc *toppath* "/.mtdb/main.db")) (glob (conc *toppath* "/.mtdb/?.db"))(glob (conc *toppath* "/.mtdb/??.db"))) '()))
        (ttdat (make-tt areapath: *toppath*))
     )
     (format #t fmtstr "DB" "host:port" "PID" "age" "last mod" "state")
     (for-each
        (lambda (dbfile)
          (let* (
            (dbfname (conc (pathname-file dbfile) ".db"))

	  (change-directory (getenv "MT_TEST_RUN_DIR"))
	  (let* ((testname (getenv "MT_TEST_NAME"))
		 (itempath (getenv "MT_ITEMPATH")))
	    (launch:extract-scripts-logpro (getenv "MT_TEST_RUN_DIR") testname itempath #f))
	  (set! *didsomething* #t))
	(debug:print 0 *default-log-port* "ERROR: Must run -regen-testfiles in a test environment (i.e. test xterm from dashboard)")))
		 	  
;;======================================================================
;; Utils for test areas
;;======================================================================

(if (args:get-arg "-regen-testfiles")
    (if (getenv "MT_TEST_RUN_DIR")
	(begin
	  (launch:setup)
	  (change-directory (getenv "MT_TEST_RUN_DIR"))
	  (let* ((testname (getenv "MT_TEST_NAME"))
		 (itempath (getenv "MT_ITEMPATH")))
	    (launch:extract-scripts-logpro (getenv "MT_TEST_RUN_DIR") testname itempath #f))
	  (set! *didsomething* #t))
	(debug:print 0 *default-log-port* "ERROR: Must run -regen-testfiles in a test environment (i.e. test xterm from dashboard)")))
		 	  
;;======================================================================
;; Archive tests
;;======================================================================
;; Archive tests matching target, runname, and testpatt
(if (equal? (args:get-arg "-archive") "replicate-db")
    (begin
          ;; check if source

    (begin
      (if (not (launch:setup))
	  (begin
	    (debug:print 0 *default-log-port* "Failed to setup, exiting") 
	    (exit 1)))
      ;; keep this one local
      ;; (open-run-close patch-db #f)
      (let ((dbstructs (db:setup #f)))
      (let ((dbstructs (db:setup)))
        (common:cleanup-db dbstructs full: #t))
      (set! *didsomething* #t)))

(if (args:get-arg "-cleanup-db")
    (begin
      (if (not (launch:setup))
	  (begin
	    (debug:print 0 *default-log-port* "Failed to setup, exiting") 
	    (exit 1)))

;;      (if (not (server:choose-server *toppath* 'home?))
;;	  (begin
;;	    (debug:print 0 *default-log-port* "Servers are not running on this host or no servers alive. Cannot run cleanup-db")
;;	    (exit 1)))

      (let ((dbstructs (db:setup #f)))
      (let ((dbstructs (db:setup)))
        (common:cleanup-db dbstructs))
      (set! *didsomething* #t)))

(if (args:get-arg "-mark-incompletes")
    (begin
      (if (not (launch:setup))
	  (begin

	(args:get-arg "-repl")
	(args:get-arg "-load"))
    (let* ((toppath (launch:setup))
	   (dbstructs (if (and toppath
			       ;; NOTE: server:choose-server is starting a server
			       ;;   either add equivalent for tcp mode or ????
                               #;(server:choose-server toppath 'home?))
                          (db:setup #t)
                          (db:setup)
                          #f))) ;; make-dbr:dbstruct path: toppath local: (args:get-arg "-local")) #f)))
      (if *toppath*
	  (cond
	   ((getenv "MT_RUNSCRIPT")
	    ;; How to run megatest scripts
	    ;;
	    ;; #!/bin/bash

;; ;; ;; redo me        (list "uname" "rundir" "final_logf" "comment"))
;; ;; ;; redo me       (set! *didsomething* #t)))

(if (args:get-arg "-import-megatest.db")
    (begin
      (launch:setup)
      (db:multi-db-sync 
       (db:setup #f)
       (db:setup)
       'killservers
       'dejunk
       'adj-testids
       'old2new
       )
      (set! *didsomething* #t)))

(if (args:get-arg "-import-sexpr")
    (begin
      (launch:setup)
      (rmt:import-sexpr (args:get-arg "-import-sexpr"))
      (set! *didsomething* #t)))
 (begin
   (launch:setup)
   (rmt:import-sexpr (args:get-arg "-import-sexpr"))
   (set! *didsomething* #t)))

(if (args:get-arg "-sync-to-megatest.db")
    (let* ((duh      (launch:setup))
	   (dbstruct (db:setup #t))
	   (dbstruct (db:setup))
	   (tmpdbpth (dbr:dbstruct-tmppath dbstruct))
	   (lockfile (conc tmpdbpth ".lock"))
	   (locked   (common:simple-file-lock lockfile)) 
	   (res      (if locked
			 (db:multi-db-sync 
			  dbstruct
			  'new2old)

;;     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
;;     GNU General Public License for more details.
;; 
;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.
;;

(require-extension (srfi 18) extras tcp s11n)

(declare (unit portlogger))
(declare (uses debugprint))
(declare (uses dbmod))

(module portlogger
*

(import scheme chicken data-structures)
(import srfi-1 posix srfi-69 hostinfo dot-locking z3
	(srfi 18) extras s11n)
(import scheme)

(cond-expand
 (chicken-4
  (import chicken data-structures)
  (import posix
	  ;; hostinfo
	  ;; dot-locking
	  extras
	  )
  
  (import (prefix sqlite3 sqlite3:))
  (import debugprint dbmod)
  )
 (chicken-5
  (import chicken.base
	  chicken.condition
	  chicken.file
	  chicken.pathname
	  chicken.process-context.posix
	  chicken.process
	  chicken.sort
	  chicken.string
	  chicken.time
	  chicken.random
	  
	  system-information
	  )
  (define file-write-access? file-writable?)
  (define random pseudo-random-integer)
  ))

(import srfi-1 srfi-69 z3
	(srfi 18) s11n)
(import (prefix sqlite3 sqlite3:))
(import debugprint dbmod)

;; lsof -i

(define (portlogger:open-db fname)
  (let* (;; (avail    (tasks:wait-on-journal fname 5 remove: #t)) ;; wait up to about 10 seconds for the journal to go away
         (avail #t)
	 (exists   (file-exists? fname))
	 (db       (if avail 

            port INTEGER PRIMARY KEY,
            state TEXT DEFAULT 'not-used',
            fail_count INTEGER DEFAULT 0,
            update_time TIMESTAMP DEFAULT (strftime('%s','now')) );")
    db))

(define (portlogger:open-run-close proc . params)
  (let* ((fname  (conc "/tmp/." (current-user-name) "-portlogger.db"))
	 (avail  (tasks:wait-on-journal fname 10))) ;; wait up to about 10 seconds for the journal to go away
  (let* ((fname  (conc "/tmp/." (current-user-name) "-portlogger.db")))
	 ;; (avail  (tasks:wait-on-journal fname 10))) ;; wait up to about 10 seconds for the journal to go away
    (handle-exceptions
     exn
     (begin
       ;; (release-dot-lock fname)
       (debug:print-error 0 *default-log-port* "portlogger:open-run-close failed. " proc " " params)
       (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
       ;; (debug:print 5 *default-log-port* "exn=" (condition->list exn))

     ((http)(make-remote))
     ((tcp) (tt:make-remote areapath))
     (else #f)))

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

(define *send-receive-mutex* (make-mutex)) ;; should have separate mutex per run-id
(define *ttdat* #f)
;; how to make area-dat
(define (rmt:set-ttdat areapath ttdat)
  (if ttdat
    ttdat
    (if *ttdat*
       *ttdat*
       (begin
         (debug:print-info 2 *default-log-port* "rmt:set-ttdat: Initialize new ttdat")
         (let* ((newremote  (make-and-init-remote areapath)))
           (set! *ttdat* newremote)
	   newremote

;; RA => e.g. usage (rmt:send-receive 'get-var #f (list varname))
;;
(define (rmt:send-receive cmd rid params #!key (attemptnum 1)(area-dat #f)) ;; start attemptnum at 1 so the modulo below works as expected
         )
       )
     )
   )
)

;; NB// area-dat replaced by ttdat
;; 
(define (rmt:send-receive cmd run-id params #!key (attemptnum 1)(ttdat #f))
  (assert (or (not run-id) (number? run-id)) "FATAL: run-id is required to be a number or #f")
  (assert *toppath* "FATAL: rmt:send-receive called with *toppath* not set.")
	
  ;; 1. check if server is started IFF cmd is a write OR if we are not on the homehost, store in runremote
  ;; 2. check the age of the connections. refresh the connection if it is older than timeout-20 seconds.
  ;; 3. do the query, if on homehost use local access
  ;;
  (let* ((start-time    (current-seconds)) ;; snapshot time so all use cases get same value
         (areapath      *toppath*);; TODO - resolve from dbstruct to be compatible with multiple areas
  (let* ((areapath      *toppath*) ;; TODO - resolve from dbstruct to be compatible with multiple areas
	 (runremote     (or area-dat
			    *runremote*))
         (attemptnum    (+ 1 attemptnum))
	 (readonly-mode (rmtmod:calc-ro-mode runremote *toppath*))
	 (testsuite     (common:get-testsuite-name))
	 (readonly-mode (rmtmod:calc-ro-mode ttdat *toppath*))
	 (testsuite     (common:get-testsuite-name)))
	 (mtexe         (common:find-local-megatest)))

    (case (rmt:transport-mode)
      ((tcp)
       (let* ((start-time    (current-seconds)) ;; snapshot time so all use cases get same value
      ((http)(http-transport-handler runremote cmd rid params attemptnum area-dat areapath readonly-mode))
      ((tcp) (tcp-transport-handler  runremote cmd rid params attemptnum area-dat areapath readonly-mode testsuite mtexe))
      ((nfs) (nfs-transport-handler  runremote cmd rid params attemptnum area-dat areapath readonly-mode testsuite mtexe))
      )))
	      (attemptnum    (+ 1 attemptnum))
	      (mtexe         (common:find-local-megatest))
	      (dbfname       (conc (dbfile:run-id->dbnum run-id)".db"))
	      (ttdat         (rmt:set-ttdat areapath ttdat))
	      (conn          (tt:get-conn ttdat dbfname))
	      (is-main       (equal? dbfname "main.db")) ;; why not (not run-id) ?
	      (server-start-proc (if is-main
				     #f
				     (lambda ()
				       ;; (debug:print-info 0 *default-log-port* "starting server for dbfname: "dbfname)
				       (rmt:start-server ;; tt:server-process-run
					areapath
					testsuite ;; (dbfile:testsuite-name)
					mtexe
					run-id)))))
	 ;; here we look at ttdat, if dbfname is NOT main.db we check that a conn exists for it
	 ;; and if there is no conn we first send a request to the main.db server to start a
	 ;; server for the dbfname.
	 #;(if (and (not is-main)(not conn)) ;; no existing connection to non-main server, call in a start up request
	 (begin
	 (server-start-proc)
	 (thread-sleep! 1)))
	 (tt:handler ttdat cmd run-id params attemptnum readonly-mode dbfname testsuite mtexe server-start-proc)))
      ((nfs)
       (nfs-transport-handler cmd run-id params attemptnum areapath readonly-mode testsuite))
      (else
       (debug:print-info 0 *default-log-port* "rmt:transport-mode is "(rmt:transport-mode))
       (assert #f "FATAL: rmt:transport-mode set to invalid value.")))))

(define (nfs-transport-handler  runremote cmd run-id params attemptnum area-dat areapath readonly-mode testsuite mtexe)
(define (nfs-transport-handler cmd run-id params attemptnum areapath readonly-mode testsuite)
  (let* ((keys     (common:get-fields *configdat*))
	 (dbstruct (dbmod:nfs-get-dbstruct run-id keys (dbfile:db-init-proc) areapath)))
    (api:dispatch-request dbstruct cmd run-id params)))
	 
	
(define (tcp-transport-handler runremote cmd run-id params attemptnum area-dat areapath readonly-mode testsuite mtexe)
  (if (not runremote)
      (let* ((newremote  (make-and-init-remote areapath)))
	(set! *runremote* newremote)
	(set! runremote newremote)))
  (let* ((dbfname (conc (dbfile:run-id->dbnum run-id)".db"))) ;;(dbfile:run-id->path areapath run-id)))
    (tt:handler runremote cmd run-id params attemptnum area-dat areapath readonly-mode dbfname testsuite mtexe)))
	
(define (rmt:print-db-stats)
  (let ((fmtstr "~40a~7-d~9-d~20,2-f")) ;; "~20,2-f"
    (debug:print 18 *default-log-port* "DB Stats\n========")
    (debug:print 18 *default-log-port* (format #f "~40a~8a~10a~10a" "Cmd" "Count" "TotTime" "Avg"))
    (for-each (lambda (cmd)
		(let ((cmd-dat (hash-table-ref *db-stats* cmd)))
		  (debug:print 18 *default-log-port* (format #f fmtstr cmd (vector-ref cmd-dat 0) (vector-ref cmd-dat 1) (/ (vector-ref cmd-dat 1)(vector-ref cmd-dat 0))))))
	      (sort (hash-table-keys *db-stats*)
		    (lambda (a b)
		      (> (vector-ref (hash-table-ref *db-stats* a) 0)
			 (vector-ref (hash-table-ref *db-stats* b) 0)))))))

(define (rmt:get-max-query-average run-id)
  (mutex-lock! *db-stats-mutex*)
  (let* ((runkey (conc "run-id=" run-id " "))
	 (cmds   (filter (lambda (x)
			   (substring-index runkey x))
			 (hash-table-keys *db-stats*)))
	 (res    (if (null? cmds)

			     (loop (car tal)(cdr tal) newmax-cmd currmax)))))))
    (mutex-unlock! *db-stats-mutex*)
    res))

(define (rmt:open-qry-close-locally cmd run-id params #!key (remretries 5))
  (let* ((qry-is-write    (not (member cmd api:read-only-queries)))
	 (db-file-path    (common:make-tmpdir-name *toppath* "")) ;;  0))
	 (dbstructs-local (db:setup #t))
	 (dbstructs-local (db:setup))
	 (read-only       (not (file-write-access? db-file-path)))
	 (start           (current-milliseconds))
	 (resdat          (if (not (and read-only qry-is-write))
			      (let ((v (api:execute-requests dbstructs-local (vector (symbol->string cmd) params))))
			;;	(handle-exceptions ;; there has been a long history of receiving strange errors from values returned by the client when things go wrong..
			;;	 exn               ;;  This is an attempt to detect that situation and recover gracefully
			;;	 (begin

	      #f))
	(begin
	  ;; (rmt:update-db-stats run-id cmd params duration)
	  ;; mark this run as dirty if this was a write, the watchdog is responsible for syncing it
	  (if qry-is-write
	      (let ((start-time (current-seconds)))
		(mutex-lock! *db-multi-sync-mutex*)
/		(set! *db-last-access* start-time)  ;; THIS IS PROBABLY USELESS? (we are on a client)
		(set! *db-last-access* start-time)  ;; THIS IS PROBABLY USELESS? (we are on a client)
                (mutex-unlock! *db-multi-sync-mutex*)))))
    res))

;;======================================================================
;;
;; A C T U A L   A P I   C A L L S  
;;
;;======================================================================

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

(define (rmt:kill-server run-id)
  (rmt:send-receive 'kill-server run-id (list run-id)))

(define (rmt:start-server run-id)
  (rmt:send-receive 'start-server 0 (list run-id)))
(define (rmt:start-server areapath testsuite mtexe run-id) ;; run on main.db server
  (rmt:send-receive 'start-server #f (list areapath testsuite mtexe run-id)))

;;======================================================================
;;  M I S C
;;======================================================================

(define (rmt:login run-id)
  (rmt:send-receive 'login run-id (list *toppath* megatest-version (client:get-signature))))

;; This login does no retries under the hood - it acts a bit like a ping.
;; Deprecated for nmsg-transport.
;;
;; (define (rmt:login-no-auto-client-setup runremote)
;;   (rmt:send-receive-no-auto-client-setup runremote 'login 0 (list *toppath* megatest-version (client:get-signature))))
;;   (rmt:send-receive-no-auto-client-setup runremote 'login #f (list *toppath* megatest-version (client:get-signature))))


;; given a hostname, return a pair of cpu load and update time representing latest intelligence from tests running on that host
(define (rmt:get-latest-host-load hostname)
  (rmt:send-receive 'get-latest-host-load 0 (list hostname)))
  (rmt:send-receive 'get-latest-host-load #f (list hostname)))

(define (rmt:sdb-qry qry val run-id)
  ;; add caching if qry is 'getid or 'getstr
  (rmt:send-receive 'sdb-qry run-id (list qry val)))

;; NOT COMPLETED
(define (rmt:runtests user run-id testpatt params)

(define (rmt:delete-run run-id)
  (rmt:send-receive 'delete-run #f (list run-id)))

(define (rmt:update-run-stats run-id stats)
  (rmt:send-receive 'update-run-stats #f (list run-id stats)))

(define (rmt:delete-old-deleted-test-records)
  (rmt:send-receive 'delete-old-deleted-test-records #f '()))
(define (rmt:delete-old-deleted-test-records run-id)
  (rmt:send-receive 'delete-old-deleted-test-records run-id (list run-id)))

(define (rmt:get-runs runpatt count offset keypatts)
  (rmt:send-receive 'get-runs #f (list runpatt count offset keypatts)))

(define (rmt:simple-get-runs runpatt count offset target last-update)
  (rmt:send-receive 'simple-get-runs #f (list runpatt count offset target last-update)))

(define (rmt:find-and-mark-incomplete run-id ovr-deadtime)
  (let* ((cfg-deadtime             (configf:lookup-number *configdat* "setup" "deadtime"))
	 (test-stats-update-period (configf:lookup-number *configdat* "setup" "test-stats-update-period")))
   (rmt:find-and-mark-incomplete-engine run-id ovr-deadtime cfg-deadtime test-stats-update-period)
   ;;call end of eud of run detection for posthook
   (launch:end-of-run-check run-id)))

;; orphaned from cherrypick merge
;;         (debug:print 0 *default-log-port* "Inserting " (length tests-data) " tests in run " runname)

     data)))

(define (rmt:import-run target run-dat)
  (let* ((runname    (car run-dat))
	 (all-dat    (cdr run-dat))
	 (tests-data (alist-ref "data" all-dat equal?))
	 (run-meta   (alist-ref "meta" all-dat equal?))
         (run-id     (string->number (alist-ref "id"   run-meta equal?))))

	 (run-id     (rmt:insert-run target runname run-meta)))
    ;; (rmt:insert-run run-id target runname run-meta)
    (if (list? tests-data)
      (begin
        (debug:print 0 *default-log-port* "import-run: inserting " (length tests-data) " tests")
      (for-each
        (lambda (test-dat)
          (let* ((test-id  (car test-dat))
        (for-each
          (lambda (test-dat)
            (let* ((test-id  (car test-dat))
	      (test-rec (cdr test-dat)))
	  (rmt:insert-test run-id test-rec)))
      tests-data)
	      (rmt:insert-test run-id test-rec)))
         tests-data)
      )
      (debug:print 0 *default-log-port* "import-run: run has no tests")
    )
  )
)

;; insert run if not there, return id either way
(define (rmt:insert-run target runname run-meta)
(define (rmt:insert-run run-id target runname run-meta)
  ;; look for id, return if found
  (debug:print 0 *default-log-port* "Insert run: "target"/"runname)
  (let* ((runs (rmtmod:send-receive 'simple-get-runs #f
				    ;;    runpatt count offset target last-update)
				    (list runname #f    #f     target #f))))
    (if (null? runs)
       (begin
        (debug:print 0 *default-log-port* "inserting run for runname " runname " target " target)
	(rmtmod:send-receive 'insert-run #f (list target runname run-meta))
	(simple-run-id (car runs)))))
	(rmtmod:send-receive 'insert-run #f (list run-id target runname run-meta))
       )
       (begin
        (debug:print 0 *default-log-port* "Found run-id " (simple-run-id (car runs)) " for runname " runname " target " target)
	(simple-run-id (car runs))
       ))))

(define (rmt:insert-test run-id test-rec)
  (let* ((testname  (alist-ref "testname" test-rec equal?))
	 (item-path (alist-ref "item_path" test-rec equal?))
         (test-id (rmt:get-test-id run-id testname item-path))
         )
        (if test-id
          (debug:print 0 *default-log-port* "test "testname"/"item-path " already exists in run-id " run-id)
          (begin
            (debug:print 0 *default-log-port* "   Insert test in run "run-id": "testname"/"item-path)
            (debug:print 0 *default-log-port* " Insert test in run "run-id": "testname"/"item-path)
            (rmtmod:send-receive 'insert-test run-id test-rec)
          )
        )
  )
)

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

    ;; first verify we are able to write the output file
    (if (not (file-read-access? infile))
        (begin 
	  (debug:print 2 *default-log-port* "ERROR: cannot read " infile)
          (debug:print 2 *default-log-port* "ERROR: run-dir is " run-dir)
          #f
          )
        (with-input-from-file infile read-lines)
	)))
	(let ((res (with-input-from-file infile read-lines)))
	  (if (null? res)
	      #f
	      (string-split (car res)))))))
  
;;  select end_time-now from
;;      (select testname,item_path,event_time+run_duration as
;;                          end_time,strftime('%s','now') as now from tests where state in
;;      ('RUNNING','REMOTEHOSTSTART','LAUNCHED'));
;;
;; NOT EASY TO MIGRATE TO db{file,mod}

  ;; Take advantage of a good place to exit if running the one-pass methodology
  (if (and (> (runs:dat-can-run-more-tests-count runsdat) 20)
	   (args:get-arg "-one-pass"))
      (exit 0))

  (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))

  (let* ((num-running             (rmt:get-count-tests-running run-id))
  (let* ((num-running             (rmt:get-count-tests-running-for-run-id run-id))
	 (num-running-in-jobgroup (rmt:get-count-tests-running-in-jobgroup run-id jobgroup))
	 (job-group-limit         (let ((jobg-count (configf:lookup *configdat* "jobgroups" jobgroup)))
				    (if (string? jobg-count)
					(string->number jobg-count)
					jobg-count))))
    (if (> (+ num-running num-running-in-jobgroup) 0)
	(runs:inc-can-run-more-tests-count runsdat)) ;; (set! *runs:can-run-more-tests-count* (+ *runs:can-run-more-tests-count* 1)))

    (if (not (null? required-tests))
	(debug:print-info 1 *default-log-port* "Adding \"" (string-intersperse required-tests " ") "\" to the run queue"))
    ;; NOTE: these are all parent tests, items are not expanded yet.
    (debug:print-info 4 *default-log-port* "test-records=" (hash-table->alist test-records))
    (let ((reglen (configf:lookup *configdat* "setup" "runqueue")))
      (if (> (length (hash-table-keys test-records)) 0)
	  (let* ((keep-going        #t)
	  (let* ()
		 (run-queue-retries 5)
		 (run-ids (rmt:get-all-run-ids)))
	    #;(for-each (lambda (run-id)
			(if keep-going
			    (handle-exceptions
				exn
			      (debug:print 0 *default-log-port* "error in calling find-and-mark-incomplete for run-id " run-id ", exn=" exn)
			      (rmt:find-and-mark-incomplete run-id #f)))) ;; ovr-deadtime))) ;; could be root of https://hsdes.intel.com/appstore/article/#/220546828/main -- Title: Megatest jobs show DEAD even though they are still running (1.64/27)
		      run-ids)
	    (runs:run-tests-queue run-id runname test-records keyvals flags test-patts required-tests
				  (any->number reglen) all-tests-registry)
	    (set! keep-going #f)
	    (if (> run-count 0) ;; handle reruns
		(begin
		  (if (not (hash-table-ref/default flags "-preclean" #f))
		      (hash-table-set! flags "-preclean" #t))
		  (if (not (hash-table-ref/default flags "-rerun" #f))
		      (hash-table-set! flags "-rerun" "ABORT,STUCK/DEAD,n/a,ZERO_ITEMS"))
      (runs:run-tests target runname test-patts user flags run-count: (- run-count 1)))
                (launch:end-of-run-check run-id)))
                  (runs:run-tests target runname test-patts user flags run-count: (- run-count 1)))
                 (launch:end-of-run-check run-id)))
	  (debug:print-info 0 *default-log-port* "No tests to run")))
    (debug:print-info 4 *default-log-port* "All done by here")
    ;; TODO: try putting post hook call here
      
    ;  (debug:print-info 2 *default-log-port* " run-count " run-count)
    ;  (runs:run-post-hook run-id))
    ;  (debug:print-info 2 *default-log-port* "Not calling post hook runcount = " run-count ))   

		      (if (eq? nth-try 'removed) ;; removed is removed - drop it NOW
			  (if (null? tal)
			      #f ;; yes, really
			      (list (car tal)(cdr tal) reg reruns))
			  (begin
			    (if (runs:lownoise (conc "FAILED prerequisites or other issue" hed) 60)
				(debug:print 0 *default-log-port* "WARNING: test " hed " has FAILED prerequisites (" (runs:pretty-string fails)") or other issue. Internal state >" nth-try "< will be overridden and we'll retry."))
			    (let* ((test-id      (rmt:get-test-id run-id testname item-path))
				   (test-info    (rmt:get-testinfo-state-status run-id test-id)) ;; we need *current* info
				   (status       (db:test-status test-info)))
			    ;; was: (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "KEEP_TRYING" #f)
                            (mt:test-set-state-status-by-testname-unless-completed run-id test-name item-path "COMPLETED" "PREQ_FAIL" #f)
			      (if (equal? status "KEEP_TRYING")
				  (mt:test-set-state-status-by-testname-unless-completed run-id test-name item-path "COMPLETED" "PREQ_FAIL" #f)
				  (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "KEEP_TRYING" #f)))
			    (hash-table-set! test-registry hed 'removed) ;; was 0
                            (if (not (and (null? reg) (null? tal)))
                                (runs:loop-values tal reg reglen regfull reruns)
                                #f))))
		     (else
		      (if (runs:lownoise (conc "FAILED prerequitests and we tried" hed) 60)
			  (debug:print 0 *default-log-port* "WARNING: test " hed " has FAILED prerequitests: " (runs:pretty-string fails) "  and we've tried at least 10 times to run it. Giving up now."))

    ;; Here is where the test_meta table is best updated
    ;; Yes, another use of a global for caching. Need a better way?
    ;;
    ;; There is now a single call to runs:update-all-test_meta and this 
    ;; per-test call is not needed. Given the delicacy of the move to 
    ;; v1.55 this code is being left in place for the time being.
    ;;
    (if (not (hash-table-ref/default *test-meta-updated* test-name #f))
    (if (not (hash-table-exists? *test-meta-updated* test-name))
        (begin
          (hash-table-set! *test-meta-updated* test-name #t)
          (runs:update-test_meta test-name test-conf)))
    
    ;; itemdat => ((ripeness "overripe") (temperature "cool") (season "summer"))
    (let* ((new-test-path (string-intersperse (cons test-path (map cadr itemdat)) "/"))
	   (test-id       (rmt:get-test-id run-id test-name item-path))

	       (begin
		 (case action
                   ((kill-runs)
                    (tasks:kill-runner target run-name "%")
                    (debug:print 1 *default-log-port* "Killing tests for run: " runkey " " (db:get-value-by-header run header "runname"))
                    )
		   ((remove-runs)
		    ;; use this location to cleanup old DELETED records? No. See below for same call
		    ;; (rmt:delete-old-deleted-test-records run-id)
		    ;; (if (tasks:need-server run-id)(tasks:start-and-wait-for-server tdbdat run-id 10))
		    ;; seek and kill in flight -runtests with % as testpatt here
		    ;; (if (equal? testpatt "%")
		    (tasks:kill-runner target run-name testpatt)
		    ;; (debug:print 0 *default-log-port* "not attempting to kill any run launcher processes as testpatt is " testpatt))
		    (debug:print 1 *default-log-port* "Removing tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((set-state-status)

                    (debug:print 2 *default-log-port* "run-paths-hash: " (hash-table-keys run-paths-hash))

                    (debug:print 1 *default-log-port* "Removing target " target "run: " run-name)
                    (if (not keep-records)
                      (begin
                        (debug:print 1 *default-log-port* "Removing DB records for the run.")
                        (rmt:delete-run run-id)
                        (rmt:delete-old-deleted-test-records))
                        (rmt:delete-old-deleted-test-records run-id))
                    )
                    (if (not (equal?  linkspath "/does/not/exist/I"))
	               (begin 
                         (debug:print 1 *default-log-port* "Recursively removing links dir " linkspath)
                         (runs:recursive-delete-with-error-msg linkspath)))

                   (for-each (lambda(runpath)

		      (with-output-to-file start-flag (lambda () (print server-key)))
		      (thread-sleep! 0.25)
		      (set! new-server-key (with-input-from-file start-flag (lambda () (read-line))))
		      (equal? server-key new-server-key)))
	       #t
               ;; If either of the above conditions is not true, print a "Gating server start" message, wait <idle-time> + 1, then call this function recursively. 
	       (begin
		 (debug:print-info 0 *default-log-port* "Gating server start, last start: "
		 (debug:print-info 2 *default-log-port* "Gating server start, last start: "
				   (seconds->time-string fmodtime) ", time since last start: " delta ", required idletime: " idletime ", gating reason:" (if old-enough "another job started a server" "too soon to start another server"))
		 
		 (thread-sleep! ( + 1 idletime))
		 (server:wait-for-server-start-last-flag areapath)))))))

;; oldest server alive determines host then choose random of youngest
;; five servers on that host

;;     (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
;; This is currently broken. Just use the number of hours with no unit.
;; Default is 60 seconds.
;; Default is 600 seconds.
;;
(define (server:expiration-timeout)
  (let* ((tmo (configf:lookup *configdat* "server" "timeout")))
    (if (string? tmo)
	(let* ((num (string->number tmo)))
	  (if num
	      (* 3600 num)

(declare (uses portlogger))

(use address-info tcp)

(module tcp-transportmod
	*
	
  (import scheme
	  (prefix sqlite3 sqlite3:)
(import scheme)

(cond-expand
 (chicken-4
  (import (prefix sqlite3 sqlite3:)
	  chicken
	  extras
	  hostinfo

	  ports
	  posix
	  files
	  data-structures
	  tcp
	  ))
 (chicken-5
  (import chicken.base
	  chicken.condition
	  chicken.file
	  chicken.pathname
	  chicken.process-context.posix
	  chicken.process
	  chicken.sort
	  chicken.string
	  chicken.time
	  chicken.tcp
	  chicken.random
	  chicken.file.posix
	  chicken.pretty-print
	  chicken.io
	  chicken.port
	  chicken.process-context

	  system-information)
  (define unsetenv unset-environment-variable!)
  ))
 
	  address-info
 (import  address-info
	  directory-utils
	  extras
	  files
	  hostinfo
	  matchable
	  md5
	  message-digest
	  ports
	  posix
	  regex
	  regex-case
	  s11n
	  srfi-1
	  srfi-18
	  srfi-4
	  srfi-69
	  stack
	  typed-records
	  tcp-server
	  tcp
	  
	  debugprint
	  commonmod
	  dbfile
	  dbmod
	  portlogger
	)

;; parameters
;;
(define tt-server-timeout-param (make-parameter 600))

;; make ttdat visible
(define *server-info* #f)
(define *server-run*  #t)

(define (tt:make-remote areapath)
  (make-tt areapath: areapath))

;; 1 ... or #f
;; and check that dbfname matches. FIXME: the propagation of dbfname and run-id
;; might not make the best sense
;;
(define (tt:valid-run-id run-id dbfname)
  (and (or (number? run-id)
	   (not run-id))
       (equal? (dbfile:run-id->dbfname run-id) dbfname)))

(tcp-buffer-size 2048)
;; (max-connections 4096) 
;; (max-connections 4096)

(define (tt:get-conn ttdat dbfname)
  (hash-table-ref/default (tt-conns ttdat) dbfname #f))

;; do all the busy work of finding and setting up conn for
;; connecting to a server
;; This function, `tt:client-connect-to-server`, is designed to manage connections between a client and a server within a testing framework.
;; The function takes four arguments:
;; 1. `ttdat`: a data structure that holds information about the testing environment or connections.
;; 2. `dbfname`: The name of the database file that the client wants to connect to.

;;    - If the server is neither running nor starting, it checks if it's been more than 10 seconds since the last server start attempt. If so, it attempts to start the server using `server-start-proc` and then sleeps for 1 second before retrying the connection.
;; 9. If no server information is found (`else` case), it checks if it's been more than 3 seconds since the last server start attempt. If so, it starts a new server using `server-start-proc`, updates the last server start time, and sleeps for 4 seconds.
;; 10. It then sleeps for 1 second and prints debug information before recursively calling itself to retry the connection.
;;
;; The function uses recursion to keep trying to connect to the server, with various sleep intervals to prevent overwhelming the system with connection attempts or server starts.
;; It also uses a hash table to cache connections and avoid reconnecting to a server if a connection already exists.
;; The function is designed to handle different server states and ensure that a server is running and available before returning a valid connection to the caller.
;; 
(define (tt:client-connect-to-server ttdat dbfname run-id testsuite)
;;
;; Here's a step-by-step explanation of what the function does:
;;
;; 1. It first asserts that the `run-id` is valid for the given `dbfname` using the `tt:valid-run-id` function. If the `run-id` is not valid, it raises a fatal error.
;; 2. It prints debug information indicating that the function `tt:client-connect-to-server` has been called with the given `dbfname`.
;; 3. It attempts to retrieve an existing connection to the server from a hash table (`tt-conns`) using the `dbfname` as the key. If a connection already exists, it prints debug information and returns the existing connection.
;; 4. If no existing connection is found, it retrieves the current server information from the servinfo file, using the `tt:get-current-server-info` function.
;; 5. It uses pattern matching to destructure the server information into variables (`host`, `port`, `start-time`, `server-id`, `pid`, `dbfname2`, `servinffile`). It then asserts that the `dbfname` from the server info matches the one provided to the function.
;; 6. It constructs a connection object (`conn`) with the server information.
;; 7. It attempts to ping the server using `tt:timed-ping` to verify that the server is running and can be communicated with.
;; 8. Depending on the result of the ping:
;;    - If the server is running (`running`), it prints debug information, saves the connection in the hash table, and returns the connection.
;;    - If the server is starting (`starting`), it sleeps for 2 seconds and then recursively calls itself to retry the connection.
;;    - If the server is neither running nor starting, it checks if it's been more than 10 seconds since the last server start attempt. If so, it attempts to start the server using `server-start-proc` and then sleeps for 1 second before retrying the connection.
;; 9. If no server information is found (`else` case), it checks if it's been more than 3 seconds since the last server start attempt. If so, it starts a new server using `server-start-proc`, updates the last server start time, and sleeps for 4 seconds.
;; 10. It then sleeps for 1 second and prints debug information before recursively calling itself to retry the connection.
;;
;; The function uses recursion to keep trying to connect to the server, with various sleep intervals to prevent overwhelming the system with connection attempts or server starts.
;; It also uses a hash table to cache connections and avoid reconnecting to a server if a connection already exists.
;; The function is designed to handle different server states and ensure that a server is running and available before returning a valid connection to the caller.
;; 
(define (tt:client-connect-to-server ttdat dbfname run-id testsuite server-start-proc)
  (assert (tt:valid-run-id run-id dbfname) "FATAL: invalid run-id "run-id)
  (debug:print-info 2 *default-log-port* "tt:client-connect-to-server " dbfname)
  (let* ((conn (hash-table-ref/default (tt-conns ttdat) dbfname #f))
	 (server-start-proc (lambda ()
			      (tt:server-process-run
			       (tt-areapath ttdat)
			       testsuite ;; (dbfile:testsuite-name)
			       (common:find-local-megatest)
			       run-id))))
  (let* ((conn              (tt:get-conn ttdat dbfname))
	 (server-start-proc (or server-start-proc
				(lambda ()
				  (assert (equal? dbfname "main.db") ;; only main.db is started here
					  "FATAL: called server-start-proc for db other than main.db")
				  (tt:server-process-run
				   (tt-areapath ttdat)
				   testsuite ;; (dbfile:testsuite-name)
				   (common:find-local-megatest)
				   run-id)))))
    (if conn
	(begin 
          (debug:print-info 2 *default-log-port* "already connected to a server for " dbfname)
           conn) ;; we are already connected to the server
	(let* ((sdat (tt:get-current-server-info ttdat dbfname)))
	  (match sdat

	;; no conn
        (let* ((sdats (tt:get-server-info-sorted ttdat dbfname))
	       (sdat  (if (null? sdats)
			  #f
			  (car sdats))))
	   (debug:print-info 2 *default-log-port* "found sdat " sdat)
           (match sdat
	    ((host port start-time server-id pid dbfname2 servinffile)
	     (assert (equal? dbfname dbfname2) "FATAL: read server info from wrong file.")
             (debug:print-info 2 *default-log-port* "no conn - in match servinffile:" servinffile)
	     (let* ((host-port (conc host":"port))
		    (conn (make-tt-conn
			   host: host
			   port: port

		   ((running)
                    (debug:print-info 2 *default-log-port* "Setting conn = " conn " in hash table")
		    (hash-table-set! (tt-conns ttdat) dbfname conn) ;;; is this ok to save before validating that the connection is good?
		    conn)
		   ((starting)
		    (thread-sleep! 2)
                    (debug:print-info 0 *default-log-port* "server for " dbfname " is in starting state, retrying connect")
		    (tt:client-connect-to-server ttdat dbfname run-id testsuite))
		    (tt:client-connect-to-server ttdat dbfname run-id testsuite server-start-proc))
		   (else
		    (let* ((curr-secs (current-seconds)))
		      ;; rm the (last server) would go here
		      (if (> (- curr-secs (tt-last-serv-start ttdat)) 10)
			  (begin
			    (debug:print-info 0 *default-log-port* "Unreachable server at "
					      host":"port" with servinfo file "servinffile", removing it")
			    (if (file-exists? servinffile)
				(handle-exceptions
				 exn
				 #f
				 (delete-file servinffile)))
			    (tt-last-serv-start-set! ttdat curr-secs)
                            (debug:print-info 0 *default-log-port* "Starting a new server on " (get-host-name))
			    (server-start-proc))) ;; start server if 10 sec since last attempt
		      (thread-sleep! 1)
                      (debug:print-info 2 *default-log-port* "server ping result was neither running nor starting. Retrying connect")
		      (tt:client-connect-to-server ttdat dbfname run-id testsuite)))))))
		      (tt:client-connect-to-server ttdat dbfname run-id testsuite server-start-proc)))))))

	    (else ;; no good server found, if haven't started server in > 5 secs, start another
	     (if (> (- (current-seconds) (tt-last-serv-start ttdat)) 3) ;; BUG - grow this number really do not want to swamp the machine with servers
		 (begin
		   (debug:print-info 0 *default-log-port* "Starting server for "dbfname)
		   (debug:print-info 0 *default-log-port* "Starting server for "dbfname " on " (get-host-name))
		   (server-start-proc)
		   (tt-last-serv-start-set! ttdat (current-seconds))
                   (thread-sleep! 4)
                   ))
	     (thread-sleep! 1)
             (debug:print-info 0 *default-log-port* "Connect to server for " dbfname)
	     (tt:client-connect-to-server ttdat dbfname run-id testsuite)))))))
             (debug:print-info 0 *default-log-port* "Connect to server from " (get-host-name) " for " dbfname)
	     (tt:client-connect-to-server ttdat dbfname run-id testsuite server-start-proc)))))))

;; returns ( result . ping_time )
(define (tt:timed-ping host port server-id)
  (let* ((start-time (current-milliseconds))
	 (result     (tt:ping host port server-id)))
    (cons result (- (current-milliseconds) start-time))))
    

(define (tt:ping host port server-id #!optional (tries-left 5))

       ;; (debug:print 0 *default-log-port* "res not in form (status errmsg result meta), got: "res)
       (try-again)))))

;; client side handler
;;
;;(tt:handler #<tt> get-keys #f () 2 #f "/home/matt/data/megatest/ext-tests" #f "main.db" "ext-tests" "/home/matt/data/megatest/bin/.22.04/../megatest")
;;
(define (tt:handler ttdat cmd run-id params attemptnum area-dat areapath readonly-mode dbfname testsuite mtexe)
(define (tt:handler ttdat cmd run-id params attemptnum readonly-mode dbfname testsuite mtexe server-start-proc)
  (debug:print 2 *default-log-port* "tt:handler cmd: " cmd " run-id: " run-id " attemptnum: " attemptnum)
  ;; connect-to-server will start a server if needed.
  ;; NOTE: areapath is passed in and in tt struct. We'll use passed in value for now.
  (let* ((areapath (tt-areapath ttdat))
  ;; connect-to-server will start a server if needed.
  (let* ((conn (tt:client-connect-to-server ttdat dbfname run-id testsuite))) ;; (hash-table-ref/default (tt-conns ttdat) dbfname #f)))
	 (conn     (tt:client-connect-to-server ttdat dbfname run-id testsuite server-start-proc))) ;; looks up conn keyed by dbfname
    (if conn
	;; have connection, call the server
	(let* ((res (tt:send-receive ttdat conn cmd run-id params)))
	  ;; res is (status errmsg result meta)
         ; (debug:print 0 *default-log-port* "conn:" conn " res: " res)
	  (match res
	    ((status errmsg result meta)
	     (if (list? meta)
		 (let* ((delay-wait (alist-ref 'delay-wait meta)))
		   (if (and (number? delay-wait)
			    (> delay-wait 0))
		       (begin
			 (debug:print 0 *default-log-port* "Server is loaded, delaying "delay-wait" seconds")
			 (thread-sleep! delay-wait)))))
	     (case status
	       ((busy) ;; result will be how long the server wants you to delay
		(let* ((dly  (if (number? result) result 0.1)))
		  (debug:print 0 *default-log-port* "WARNING: server for "dbfname" is busy, will try again in "dly" seconds.")
		(let* ((raw-dly  (if (number? result) result 0.1))
		       (dly      (+ raw-dly (/ attemptnum 10)))) ;; (* raw-dly (/ attemptnum 2))))
		  (debug:print 0 *default-log-port* "WARNING: server for "dbfname" is busy, cmd is "cmd", will try again in "dly" seconds. This is attempt "(- attemptnum 1))
		  (thread-sleep! dly)
		  (tt:handler  ttdat cmd run-id params (+ attemptnum 1) area-dat areapath readonly-mode dbfname testsuite mtexe)))
		  (tt:handler  ttdat cmd run-id params (+ attemptnum 1) readonly-mode dbfname testsuite mtexe server-start-proc)))
	       ((loaded)
		(debug:print 0 *default-log-port* "WARNING: server for "dbfname" is loaded, slowing queries.")
		(tt:backoff-incr (tt-conn-host conn)(tt-conn-port conn))
		result) ;; (tt:handler  ttdat cmd run-id params (+ attemptnum 1) area-dat areapath readonly-mode dbfname testsuite mtexe))
		result) ;; (tt:handler  ttdat cmd run-id params (+ attemptnum 1) readonly-mode dbfname testsuite mtexe))
	       (else
		result)))
	    (else ;; did not receive properly formated result
	     (if (not res) ;; tt:send-receive telling us that communication failed
		 (let* ((host    (tt-conn-host conn))
			(port    (tt-conn-port conn))
			;; (dbfname (tt-conn-port conn)) ;; 192.168.0.127:4242-726924:4.db
			(pid     (tt-conn-pid  conn))
                        ;;(servinf (tt-conn-servinf-file conn))) 
			(servinf (tt-servinf-file ttdat))) ;; (conc areapath"/.servinfo/"host":"port"-"pid":"dbfname))) ;; TODO, use (server:get-servinfo-dir areapath)
		   (hash-table-set! (tt-conns ttdat) dbfname #f)
		   (hash-table-set! (tt-conns ttdat) dbfname #f) ;; clear out the conn for this dbfname to force finding new server
		   (if (and servinf (file-exists? servinf))
		       (begin
			 (if (< attemptnum 10)
			     (begin
			       (thread-sleep! 0.5)
			       (tt:handler ttdat cmd run-id params (+ attemptnum 1) area-dat areapath readonly-mode dbfname testsuite mtexe))
			       (tt:handler ttdat cmd run-id params (+ attemptnum 1) readonly-mode dbfname testsuite mtexe server-start-proc))
			     (begin
			       (debug:print 0 *default-log-port* "INFO: no response from server "host":"port" for "dbfname)
			       (if (and (file-exists? servinf)
					(> (- (current-seconds)(file-modification-time servinf)) 60))
				   (begin
				     (debug:print 0 *default-log-port* "INFO: "servinf" file seems old and no ping response, removing it.")
				     (handle-exceptions
					 exn
				       #f
				       (delete-file* servinf))
				     (tt:handler ttdat cmd run-id params (+ attemptnum 1) area-dat areapath readonly-mode dbfname testsuite mtexe))
				     (tt:handler ttdat cmd run-id params (+ attemptnum 1) readonly-mode dbfname testsuite mtexe server-start-proc))
				   (begin
				     ;; start server - addressed in client-connect-to-server
				     ;; delay        - addressed in client-connect-to-server
				     ;; try again
				     (thread-sleep! 0.25) ;; dunno, I think this needs to be here
				     (tt:handler ttdat cmd run-id params (+ attemptnum 1) area-dat areapath readonly-mode dbfname testsuite mtexe))
				     (tt:handler ttdat cmd run-id params (+ attemptnum 1) readonly-mode dbfname testsuite mtexe server-start-proc))
				   ))))
		       (begin ;; no server file, delay and try again
			 (debug:print 2 *default-log-port* "INFO: connection to server "host":"port" broken for "dbfname", no servinf file. Server exited? ")
			 (thread-sleep! 0.5)
			 (tt:handler ttdat cmd run-id params (+ attemptnum 1) area-dat areapath readonly-mode dbfname testsuite mtexe))))
			 (tt:handler ttdat cmd run-id params (+ attemptnum 1) readonly-mode dbfname testsuite mtexe server-start-proc))))
		 (begin ;; this case is where res is malformed. Probably should abort
		   (assert #f "FATAL: tt:handler received bad data "res)
		   ;; (debug:print 0 *default-log-port* "INFO: got corrupt data from server "host":"port", "res", for "dbfname", will try again.")
		   ;; (tt:handler ttdat cmd run-id params (+ attemptnum 1) area-dat areapath readonly-mode dbfname testsuite mtexe)
		   ;; (tt:handler ttdat cmd run-id params (+ attemptnum 1) readonly-mode dbfname testsuite mtexe)
		   )))))
	(begin
	  (thread-sleep! 1) ;; no conn yet set up, give it a rest and try again
	  (tt:handler ttdat cmd run-id params attemptnum area-dat areapath readonly-mode dbfname testsuite mtexe)))))
	  (tt:handler ttdat cmd run-id params attemptnum readonly-mode dbfname testsuite mtexe server-start-proc)))))

(define (tt:bid-for-servership run-id)
  #f)

;; gets server info and appends path to server file
;; sorts by age, oldest first
;;
;; returns list of (host port startseconds server-id servinfofile)
;;
(define (tt:get-server-info-sorted ttdat dbfname)

       (if (or (> (length sorted) 1)
	       (common:low-noise-print 120 "server info sorted"))
	   (debug:print 2 *default-log-port* "SERVER #"count": "(string-intersperse (map conc sorted) ", ")))
       (set! count (+ count 1)))
     sorted)
    sorted))
    
(define (tt:get-current-server-info ttdat dbfname)
  (assert (tt-areapath ttdat) "FATAL: areapath not set in ttdat.")
  ;;
  ;; TODO - replace most of below with tt;get-server-info-sorted
  ;;
  (let* ((areapath (tt-areapath ttdat))
	 (sfiles   (tt:find-server areapath dbfname))
	 (sdats    (filter car (map tt:server-get-info sfiles))) ;; first element is #f if the file disappeared while being read
	 (sorted   (sort sdats (lambda (a b)
				 (< (list-ref a 2)(list-ref b 2))))))
    (if (null? sorted)
	#f  ;; we'll want to wait until extra servers have exited
	(car sorted))))

(define (tt:send-receive ttdat conn cmd run-id params)
  (let* ((host-port (tt-conn-host-port conn)) ;; (conc (tt-conn-host conn)":"(tt-conn-port conn)))
	 (host      (tt-conn-host conn))
	 (port      (tt-conn-port conn))
	 (dat       (list cmd run-id params #f))) ;; no meta data yet
    (tt:send-receive-direct host port dat)))

			       (if (> adj wait-delay)
				   0
				   (- wait-delay adj))
			       0)))
	  (if (> new-wait 0)
	      (begin
		(if (common:low-noise-print 10 "delay wait message")
		    (debug:print-info 0 *default-log-port* "Server loaded, DelayWait: "new-wait))
		    (debug:print-info 0 *default-log-port* "Server on host " host " loaded, DelayWait: "new-wait))
		(tt:backoff-wait-delay-set! bkoff new-wait)
		(tt:backoff-last-adj-t-set! bkoff (current-seconds))
		(thread-sleep! new-wait))
	      (hash-table-delete! *tt:backoff-smoothing* host-port))))))

(define (tt:send-receive-direct host port dat #!key (ping-mode #f)(tries-remaining 25))
  (assert (number? port) "FATAL: tt:send-receive-direct called with port not a number "port)
  (assert (number? port) "FATAL: tt:send-receive-direct called with  a port that is not a number "port)
  (tt:backoff-decr-and-wait host port)
  (let* ((retry          (lambda ()
			   (tt:send-receive-direct host port dat tries-remaining: (- tries-remaining 1))))
	 (full-err-print (lambda (exn msg)
			   (if (condition? exn)
			       (begin
				 (pp (condition->list exn) *default-log-port*)

  #f)

;; start the listener and start responding to requests
;;
;; NOTE: organise by dbfname, not run-id so we don't need
;;       to pull in more modules
;;
;; This is the routine called in megatest.scm to start a server
;; This is the routine called in megatest.scm to start a server. NOTE: sequence is different for main.db vs. X.db
;;
;; Server viability is checked in keep-running. Blindly start and run here.
;;
(define (tt:start-server areapath run-id dbfname-in handler keys)
  (assert areapath "FATAL: areapath not provided for tt:start-server")
  ;; is there already a server for this dbfile? Then exit.
  (debug:print 2 *default-log-port* "tt:start-server: " dbfname-in)
  (let* ((ttdat   (make-tt areapath: areapath))
	 (dbfname (or dbfname-in (dbmod:run-id->dbfname run-id)))
	 (servers (tt:find-server areapath dbfname))) ;; should use tt:get-current-server-info instead
         (debug:print 0 *default-log-port* "Found " (length servers) " already running for " dbfname)
    (if (> (length servers) 0)
	 (dbfname (or dbfname-in (dbmod:run-id->dbfname run-id))))
    (set! *server-info* ttdat)
    ;; (debug:print 0 *default-log-port* "Found " (length servers) " already running for " dbfname)
    ;; (if (> (length servers) 0)
	(begin
	  (debug:print 0 *default-log-port* "INFO: found server(s) already running for db "dbfname", "(string-intersperse servers ",")" Exiting.")
	  (exit))
	(let* ((dbstruct   (dbmod:open-dbmoddb areapath run-id dbfname (dbfile:db-init-proc) keys)))
	  (tt-handler-set! ttdat (handler dbstruct))
	  (let* ((tcp-thread (make-thread
			      (lambda ()
				(tt:start-tcp-server ttdat)) ;; start the tcp-server which applies handler to incoming data
			      "tcp-server-thread"))
		 (run-thread (make-thread
			      (lambda ()
				(tt:keep-running ttdat dbfname dbstruct)))))
	    (thread-start! tcp-thread)
	    (thread-start! run-thread)
    (let* ((dbstruct   (dbmod:open-dbmoddb areapath run-id dbfname (dbfile:db-init-proc) keys)))
      (tt-handler-set! ttdat (handler dbstruct))
      (let* ((servinf-created #f)
	     (tcp-thread      (make-thread
			       (lambda ()
				 ;; NOTE: tt-port and tt-host are set in connect-listener which is called under tt:start-tcp-server
				 (tt:start-tcp-server ttdat)) ;; start the tcp-server which applies handler to incoming data
			       "tcp-server-thread"))
	     (run-thread      (make-thread
			       (lambda ()
				 (tt:keep-running ttdat dbfname dbstruct)))))
	(thread-start! tcp-thread)

	(let* ((areapath     (tt-areapath ttdat))
	       (nosyncdbpath (conc areapath"/.mtdb"))
	       (servers      ;; (tt:find-server areapath dbfname)))
		(tt:get-server-info-sorted ttdat dbfname)) ;; (host port startseconds server-id servinfofile)
	       (good-srvrs  
		;; contact servers via ping, if no response remove the .servinfo file
		(let loop ((servrs     servers)
			   (prime-host #f)
			   (result    '()))
		  (if (null? servrs)
		      (reverse result)
		      (let* ((servdat (car servrs)))
			(match servdat
			     ((host port startseconds server-id pid dbfilename servinfofile)
                              (debug:print-info 0 *default-log-port* "Good servinfo file: " servdat)
			      (let* ((ping-res  (tt:timed-ping host port server-id))
				     (good-ping (match ping-res
						   ((result . ping-time)
						    (not result)) ;; we couldn't reach the server or it was not a megatest server
						   (else #f))) ;; the ping failed completely?
				     (same-host (or (not prime-host) ;; i.e. this is the first host
						    (equal? prime-host host)))
				     (keep-srv  (and good-ping same-host)))
				(if keep-srv	
				    (loop (cdr servrs)
					  host
					  (cons servdat result))
				    (begin
                                      ;; (debug:print-info 0 *default-log-port* "good-ping: " good-ping " same-host: " same-host "keep-srv: " keep-srv)
				      (handle-exceptions
				       exn
				       (debug:print-info 0 *default-log-port* "Error removing server info file: "servinfofile", "
							 (condition->list exn))
				       (delete-file* servinfofile))
				      (loop (cdr servrs) prime-host result)))))
			     (else
			      ;; can't delete it as we don't have a filename. NOTE: Should really never get here.
			      (debug:print-info 0 *default-log-port* "ERROR: bad servinfo record \""servdat"\"")
			      (loop (cdr servrs) prime-host result)) ;; drop 
			     )))))
	       (home-host (if (null? good-srvrs)
			      #f
			      (caar good-srvrs))))
	  ;; by here we have a trustworthy list of servers and we have removed the .servinfo file for any unresponsive servers
	  ;; and the list is in good-srvrs
	  (cond
	   ((not home-host) ;; no servers yet, go ahead and start
	    (debug:print-info 0 *default-log-port* "No servers yet, starting on "(get-host-name)))
	   ((> (length good-srvrs) 2) ;; don't need more, just exit
	    (debug:print-info 0 *default-log-port* "Have "(length good-srvrs)", no need for more, exiting.")
	    (exit))
	   ((not (equal? home-host (get-host-name))) ;; there is a home-host and we are not on it
	    (debug:print-info 0 *default-log-port* "Prime main server is on host "home-host", but we are on host "(get-host-name)", exiting.")
	    (exit))
	   (else
	    (debug:print-info 0 *default-log-port* "Starting on host "(get-host-name)", along with "(length good-srvrs)" other servers.")))

	    (let* ((areapath     (tt-areapath ttdat))
		   (nosyncdbpath (conc areapath"/.mtdb")))
	      ;; this didn't seem to work, is port not available yet?
	      (let loop ((count 0))
		(if (tt-port ttdat)
		    (begin
		      (procinf-port-set! *procinf* (tt-port ttdat))
		      (procinf-dbname-set! *procinf* dbfname)
		      (dbfile:with-no-sync-db
		       nosyncdbpath
		       (lambda (nsdb)
			 (dbfile:insert-or-update-process nsdb *procinf*))))
		    (if (< count 5)
			(begin
			  (thread-sleep! 0.5)
			  (loop (+ count 1)))
			(debug:print 0 *default-log-port* "ERROR: (tt-port ttdat) no port set!"))))
	    
	      (thread-join! run-thread) ;; run thread will exit on timeout or other conditions
	      ;; replace with call to (dbfile:set-process-done nsdb host pid reason)
	      (procinf-status-set! *procinf* "done")
	      (procinf-end-set! *procinf* (current-seconds))
	      (dbfile:with-no-sync-db
	       nosyncdbpath
	       (lambda (nsdb)
		 (dbfile:insert-or-update-process nsdb *procinf*)))
              (debug:print 0 *default-log-port* "Exiting now.")
	      (exit)))))))
	  ;; this didn't seem to work, is port not available yet?
	  (let loop ((count 0))
	    (if (tt-port ttdat)
		(begin
		  (procinf-port-set! *procinf* (tt-port ttdat))
		  (procinf-dbname-set! *procinf* dbfname)
		  (dbfile:with-no-sync-db
		   nosyncdbpath
		   (lambda (nsdb)
		     (dbfile:insert-or-update-process nsdb *procinf*))))
		(if (< count 10)
		    (begin
		      (thread-sleep! 0.25)
		      (loop (+ count 1)))
		    (begin
		      (debug:print 0 *default-log-port* "ERROR: (tt-port ttdat) no port set! Exiting.")
		      (exit)))))
	  
	  ;; create a servinfo file start keep-running
          (debug:print 0 *default-log-port* "Creating servinfo file for " dbfname)
	  (tt:create-server-registration-file ttdat dbfname)
	  (procinf-status-set! *procinf* "running")
	  (tt-state-set! ttdat 'running)
	  (dbfile:with-no-sync-db
	   nosyncdbpath
	   (lambda (nsdb)
	     (dbfile:insert-or-update-process nsdb *procinf*)))
	  (thread-start! run-thread)

	  (thread-join! run-thread) ;; run thread will exit on timeout or other conditions
	  
	  ;; (tcp-close (tt-socket ttdat)) ;; close up ports here

	  ;; replace with call to (dbfile:set-process-done nsdb host pid reason)
	  (procinf-status-set! *procinf* "done")
	  (procinf-end-set! *procinf* (current-seconds))
	  ;; either convert this to use set-process-done or get rid of set-process-done
	  (dbfile:with-no-sync-db
	   nosyncdbpath
	   (lambda (nsdb)
	     (dbfile:insert-or-update-process nsdb *procinf*)))
	  (debug:print 0 *default-log-port* "Exiting now.")
	  (exit))))))

(define (tt:keep-running ttdat dbfname dbstruct)
  ;; verfiy conn for ready
  ;; listener socket has been started by this stage
  ;; wait for a port before creating the registration file
  ;;
  (let* ((db-locked-in #f)
	 (areapath     (tt-areapath ttdat))
	 (nosyncdbpath (conc areapath"/.mtdb"))
	 (cleanup (lambda ()
		    (if (tt-cleanup-proc ttdat)
			((tt-cleanup-proc ttdat)))
		    (dbfile:with-no-sync-db nosyncdbpath
					    (lambda (db)
					      (let* ((dbtmpname (dbr:dbstruct-dbtmpname dbstruct)))
						(debug:print-info 0 *default-log-port* "keep-running: removing lock for file "dbtmpname)
						(db:no-sync-del! db dbfname)
  ;; (debug:print-info 0 *default-log-port* "keep-running: removing lock for file "dbtmpname)
  
                                                ))))))
    (set! *server-info* ttdat)
    (let loop ((count 0))
      (if (> count 240)
	  (begin
	    (debug:print 0 *default-log-port* "FATAL: Could not start a tcp server, giving up.")
	    (exit 1))
  ;; at this point the server is running and responding to calls, we just monitor
  ;; for db calls and exit if there are none.
	  (if (not (tt-port ttdat)) ;; no connection yet
	      (begin
		(thread-sleep! 0.25)
		(loop (+ count 1))))))
    
    (tt:create-server-registration-file ttdat dbfname)

  ;; if I am not in the first 3 servers, exit
    ;; now start watching the last-access, if it hasn't been touched
    ;; in over ten seconds we exit
  (let* ((start-time (current-seconds)))
    (thread-sleep! 0.05) ;; any real need for delay here?
    (let loop ()
      (let* ((servers (tt:get-server-info-sorted ttdat dbfname))
      (let* ((servers   (tt:get-server-info-sorted ttdat dbfname))
	     (ok      (cond
		       ((null? servers) #f) ;; not ok
	     (home-host (if (null? servers)
		       ((equal? (list-ref (car servers) 6) ;; compare the servinfofile
				(tt-servinf-file ttdat))
			(let* ((res (if db-locked-in
					#t
			    #f
					(let* ((lock-result  ;; this is the primary lock - need to double verify that got it
						(dbfile:with-no-sync-db
						 nosyncdbpath
						 (lambda (db)
			    (caar servers)))
	     (my-index  (list-index (lambda (x)
						   (db:no-sync-lock-and-check db dbfname
									      (tt-servinf-file ttdat)
				      (equal? (list-ref x 6)
					      (tt-servinf-file ttdat)))
									      ;; (dbr:dbstruct-dbtmpname dbstruct)
									      ))))
					       (success (car lock-result)))
				    servers))
	     (ok         (cond
					  (if success
					      (begin
						(tt-state-set! ttdat 'running)
						(debug:print 0 *default-log-port* "Got server lock for " dbfname)
			  ((not *server-run*)
			   (debug:print 0 *default-log-port* "WARNING: received a stop server from client by remote request.")
						(set! db-locked-in #t)
						#t)
					      (begin
						(debug:print 0 *default-log-port* "Failed to get server lock for "dbfname)
						#f))))))
			   #f)
			  ((null? servers)
			   (debug:print 0 *default-log-port* "WARNING: no servinfo files found, this cannot be.")
			   #f) ;; not ok
			  ((> my-index 2)
			   (debug:print 0 *default-log-port* "WARNING: there are more than two servers ahead of me, I'm not needed, exiting.")
			   #f) ;; not ok to not be in first three
			  ((eq? (tt-state ttdat) 'running) #t) ;; we are good to keep going
			  ((> (- (current-seconds) start-time) 30)
			   (debug:print 0 *default-log-port* "WARNING: over 30 seconds and not yet in runnning mode. Exiting.")
			   #f)
			  (if (and res (common:low-noise-print 120 "top server message"))
			      (debug:print-info 0 *default-log-port* "Keep running, I'm the top server for "
						dbfname" on "(tt-host ttdat)":"(tt-port ttdat)))
			  res))
		       (else
			  (else #t))))
                        ;; wrong servinfo file
			(debug:print-info 0 *default-log-port* "I'm not the lead server: "servers)
			(let* ((leadsrv (car servers)))
			  (match leadsrv
			    ((host port startseconds server-id pid dbfname servinfofile)
			     (let* ((result  (tt:timed-ping host port server-id))
				    (res     (car result))
				    (ping    (cdr result)))
			       (debug:print-info 0 *default-log-port* "Ping to "host":"port", with server-id "server-id
						 ", and file "servinfofile" returned "res)
			       (if res
				   #f ;; not the server, but all good, want to exit
				   (if (and (file-exists? servinfofile)
					  (> (- (current-seconds)(file-modification-time servinfofile)) 30))
				     (begin
				       ;; can't ping and file has been on disk 15 seconds, go ahead and try to remove it
				       (debug:print-info 0 *default-log-port* "Removing apparently dead server info file: "servinfofile)
                                       (handle-exceptions
                                        exn
                                        (debug:print-info 0 *default-log-port* "Error removing server info file: "servinfofile)
				        (delete-file* servinfofile)
                                       )
				       #t) ;; not the server but the server is not reachable
				     (begin
				       (debug:print 0 *default-log-port* "I'm not the server but could not ping "host":"port", will try again.")
				       (thread-sleep! 1) ;; just because
				       #t)))))
			    (else ;; should never get here
			     (debug:print 0 *default-log-port* "BAD SERVER RECORD: "leadsrv)
			     (assert #f "Bad server record "leadsrv))))))))
	(if ok
	    (tt-last-access-set! ttdat *db-last-access*) ;; bit silly, just use db-last-access
	    (begin
	      (debug:print 0 *default-log-port* "Exiting immediately")
	      (cleanup)
	      (tt:shutdown-server ttdat)
	      (exit)))

	(let* ((last-update (dbr:dbstruct-last-update dbstruct))
	       (curr-secs   (current-seconds)))
	  (if (and (eq? (tt-state ttdat) 'running)
		   (> (- curr-secs last-update) 3)) ;; every 3-4 seconds update the db?
	      (begin
		(set! (file-modification-time (tt-servinf-file ttdat)) (current-seconds))
		   (> (- curr-secs last-update) 5)) ;; every 5 seconds update the db?
	      (let* ((sinfo-file (tt-servinf-file ttdat)))
		;; (debug:print 0 *default-log-port* "INFO: touching "sinfo-file)
		(set! (file-modification-time sinfo-file) (current-seconds))
		((dbr:dbstruct-sync-proc dbstruct) last-update)
		(dbr:dbstruct-last-update-set! dbstruct curr-secs))))
	  
	
	(if (< (- (current-seconds) (tt-last-access ttdat)) (tt-server-timeout-param))
	    (begin
	      (thread-sleep! 5)
	      (loop)))))
    (cleanup)
    (tt:shutdown-server ttdat)
    (debug:print 0 *default-log-port* "INFO: Server timed out, exiting from tt:keep-running.")))

  

;; ;; given an already set up uconn start the cmd-loop
;; ;;
;; (define (tt:cmd-loop ttdat)
;;   (let* ((serv-listener (-socket uconn))
;; 	 (listener      (lambda ()
;; 			  (let loop ((state 'start))
;; 			    (let-values (((inp oup)(tcp-accept serv-listener)))
;; 			      ;; (mutex-lock! *send-mutex*) ;; DOESN'T SEEM TO HELP
;; 			      (let* ((rdat  (deserialize inp)) ;; '(my-host-port qrykey cmd params)
;; 				     (resp  (ulex-handler uconn rdat)))
;; 				(serialize resp oup)
;; 				(close-input-port inp)
;; 				(close-output-port oup)
;; 				;; (mutex-unlock! *send-mutex*) ;; DOESN'T SEEM TO HELP
;; 				)
;; 			      (loop state))))))
;;     ;; start N of them
;;     (let loop ((thnum   0)
;; 	       (threads '()))
;;       (if (< thnum 100)
;; 	  (let* ((th (make-thread listener (conc "listener" thnum))))
;; 	    (thread-start! th)
;; 	    (loop (+ thnum 1)
;; 		  (cons th threads)))
;; 	  (map thread-join! threads)))))
;; 
;; 
;; 
;; (define (wait-and-close uconn)
;;   (thread-join! (udat-cmd-thread uconn))
;;   (tcp-close (udat-socket uconn)))
;; 
;; 

(define (tt:shutdown-server ttdat)
  (let* ((cleanproc (tt-cleanup-proc ttdat))
	 (port      (tt-port         ttdat)))
  (let* ((host (tt-host ttdat))
	 (port (tt-port ttdat))
	 (sinf (tt-servinf-file ttdat)))
    (tt-state-set! ttdat 'shutdown)
    (portlogger:open-run-close portlogger:set-port port "released")
    (if cleanproc (cleanproc))
    (if (file-exists? sinf)
    (tcp-close (tt-socket ttdat)) ;; close up ports here
	(delete-file* sinf))
    ))

;; (define (wait-and-close uconn)
;;   (thread-join! (tt-cmd-thread uconn))
;;   (tcp-close (tt-socket uconn)))

;; return servid
;; side-effects:
;;   ttdat-cleanup-proc is populated with function to remove the serverinfo file
(define (tt:create-server-registration-file ttdat dbfname)
  (let* ((areapath (tt-areapath ttdat))
	 (servdir  (tt:get-servinfo-dir areapath))
	 (host     (tt-host ttdat))
	 (port     (tt-port ttdat))
	 (servinf (conc servdir"/"host":"port"-"(current-process-id)":"dbfname))
	 (serv-id (tt:mk-signature areapath))
	 (serv-id (tt:mk-signature areapath)))
	 (clean-proc (lambda ()
		       (delete-file* servinf)
		       )))
    (assert (and host port) "FATAL: tt:create-server-registration-file called with no conn, dbfname="dbfname)
    (tt-cleanup-proc-set! ttdat clean-proc)
    (tt-servinf-file-set! ttdat servinf)
    (with-output-to-file servinf
      (lambda ()
	(print "SERVER STARTED: "host":"port" AT "(current-seconds)" server-id: "serv-id" pid: "(current-process-id)" dbfname: "dbfname)))
      serv-id))

;; find valid server

           )
           sfiles
         )
    (debug:print-info 2 *default-log-port* "tt:find-server: good-files: " good-files " sfiles: " sfiles)
    good-files))

;; given a path to a server info file return: host port startseconds server-id pid dbfname logf
;; example of what it's looking for in the log file:
;; example of what it's looking for in the file:
;;     SERVER STARTED: 10.38.175.67:50216 AT 1616502350.0 server-id: 4907e90fc55c7a09694e3f658c639cf4 
;;
(define (tt:server-get-info logf)
  (let ((server-rx    (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+) server-id: (\\S+) pid: (\\d+) dbfname: (\\S+)")) ;; SERVER STARTED: host:port AT timesecs server id
	(bad-dat      (list #f #f #f #f #f #f logf)))
     (let ((fdat     (handle-exceptions
			 exn
		       (begin
			 ;; WARNING: this is potentially dangerous to blanket ignore the errors
			 (debug:print-info 0 *default-log-port* "Unable to get server info from "logf", exn="(condition->list exn))
			 ;; BUG, TODO: add err checking, for now blanket ignore the errors?
			 (debug:print-info 0 *default-log-port* "Unable to get server info from "logf
					   ", exn="(condition->list exn))
			 '()) ;; no idea what went wrong, call it a bad server, return empty list
		       (with-input-from-file logf read-lines))))
       (if (null? fdat) ;; bad data, return bad-dat
	   bad-dat
	   (let loop ((inl  (car fdat))
		      (tail (cdr fdat))
		      (lnum 0))

			    (string->number pid)
			    dbfname
			    logf))
		     (else
		      (debug:print 0 *default-log-port* "ERROR: did not recognise SERVER line info "mlst)
		      bad-dat)))))))))

(define *last-server-start* (make-hash-table))

(define (tt:too-recent-server-start dbfname)
  (let* ((last-run-time (hash-table-ref/default *last-server-start* dbfname #f)))
    (and last-run-time
	 (< (- (current-seconds) last-run-time) 5))))
    
;; Given an area path,  start a server process    ### NOTE ### > file 2>&1 
;; if the target-host is set 
;; try running on that host
;;   incidental: rotate logs in logs/ dir.
;;
(define  (tt:server-process-run areapath testsuite mtexe run-id #!key (profile-mode "")) ;; areapath is *toppath* for a given testsuite area
  (assert areapath  "FATAL: tt:server-process-run called without areapath defined.")
  (assert testsuite "FATAL: tt:server-process-run called without testsuite defined.")
  (assert mtexe     "FATAL: tt:server-process-run called without mtexe defined.")
  ;; mtest -server - -m testsuite:ext-tests -db 6.db
  (let* ((dbfname  (dbmod:run-id->dbfname run-id))
	 (load     (get-normalized-cpu-load))
	 (trying   (length (tt:find-server areapath dbfname)))
	 (nrun     (number-of-processes-running (conc "mtest.*server.*"testsuite".*"dbfname))))
    (cond
     ((> load 2.0)
      (debug:print 0 *default-log-port* "Normalized load "load" on " (get-host-name) " is over the limit of 2.0. Not starting a server.")
      (thread-sleep! 1))
     ((> nrun 100)
      (debug:print 0 *default-log-port* nrun" servers running on " (get-host-name) ", not starting another.")
      (thread-sleep! 1))
     ((> trying 4)
      (debug:print 0 *default-log-port* trying" servers registered in .servinfo dir. not starting another.")
      (thread-sleep! 1))
     (else
      (if (not (file-exists? (conc areapath"/logs")))
	      (create-directory (conc areapath"/logs") #t))
	  (let* ((logfile   (conc areapath "/logs/server-"dbfname"-"(current-process-id)".log")) ;; -" curr-pid "-" target-host ".log"))
		 (cmdln     (conc
			     mtexe
			     " -startdir "areapath
			     " -server - ";; (or target-host "-")
			     " -m testsuite:"testsuite
			     " -db "dbfname ;; (dbmod:run-id->dbfname run-id)
			     " " profile-mode
			     (conc " >> " logfile " 2>&1 &"))))
	    ;; we want the remote server to start in *toppath* so push there
	    ;; (push-directory areapath) ;; use cd in the command line instead
  (let* ((dbfname  (dbmod:run-id->dbfname run-id)))
    (if (tt:too-recent-server-start dbfname)
	#f
	(let* ((load     (get-normalized-cpu-load))
	       (srvrs    (tt:find-server areapath dbfname))
	       (trying   (length srvrs))
	       (nrun     (number-of-processes-running (conc "mtest.*server.*"testsuite".*"dbfname))))
	  (cond
	   ((> load 2.0)
	    (debug:print 0 *default-log-port* "Normalized load "load" on " (get-host-name) " is over the limit of 2.0. Not starting a server. Please reduce the load on "(get-host-name)" by killing some processes")
	    (thread-sleep! 1)
	    #f)
	   ((> nrun 100)
	    (debug:print 0 *default-log-port* nrun" servers running on " (get-host-name) ", not starting another.")
	    (thread-sleep! 1)
	    #f)
	   ((> trying 2)
	    (debug:print 0 *default-log-port* trying" servers registered in .servinfo dir. not starting another.")
	    (thread-sleep! 1)
	    #f)
	   (else
	    (if (not (file-exists? (conc areapath"/logs")))
		(create-directory (conc areapath"/logs") #t))
	    (let* ((logfile   (conc areapath "/logs/server-"dbfname"-"(current-process-id)".log")) ;; -" curr-pid "-" target-host ".log"))
		   (cmdln (conc
			       mtexe
			       " -startdir "areapath
			       " -server - ";; (or target-host "-")
			       " -m testsuite:"testsuite
			       " -db "dbfname ;; (dbmod:run-id->dbfname run-id)
			       " " profile-mode
			       #;(conc " >> " logfile " 2>&1 &"))))
	      ;; we want the remote server to start in *toppath* so push there
	      ;; (push-directory areapath) ;; use cd in the command line instead
	    (debug:print 2 *default-log-port* "INFO: Trying to start server in tcp mode (" cmdln ") at "(common:human-time)" for "areapath)
	    ;; (debug:print 0 *default-log-port* "INFO: starting server at " (common:human-time))
	      ;; (debug:print 0 *default-log-port* "INFO: starting server at " (common:human-time))

	      (setenv "NBFAKE_QUIET" "yes") ;; BUG: change to with-environment-variable ...
	      (setenv "NBFAKE_LOG" logfile)
	      (system (conc "cd "areapath" ; nbfake " cmdln))
	      (unsetenv "NBFAKE_QUIET")
	      (unsetenv "NBFAKE_LOG")
	    (system cmdln)
	    ;; ;; use below to go back to nbfake - nbfake does cause trouble ...
	    ;; (setenv "NBFAKE_QUIET" "yes") ;; BUG: change to with-environment-variable ...
	    ;; (setenv "NBFAKE_LOG" logfile)
	    ;; (system (conc "cd "areapath" ; nbfake " cmdln))
	    ;; (unsetenv "NBFAKE_QUIET")
	    ;; (unsetenv "NBFAKE_LOG")
	    
	    ;;(pop-directory)
	    )))))
	      ;; (system cmdln)
	      (hash-table-set! *last-server-start* dbfname (current-seconds))
	      ;; ;; use below to go back to nbfake - nbfake does cause trouble ...
	      ;; (setenv "NBFAKE_QUIET" "yes") ;; BUG: change to with-environment-variable ...
	      ;; (setenv "NBFAKE_LOG" logfile)
	      ;; (system (conc "cd "areapath" ; nbfake " cmdln))
	      ;; (unsetenv "NBFAKE_QUIET")
	      ;; (unsetenv "NBFAKE_LOG")
	      
	      ;;(pop-directory)
	      #t)))))))

;;======================================================================
;; tcp connection stuff
;;======================================================================

;; find a port and start tcp-server. This only starts the tcp portion of
;; the server, look at (tt:start-server ...) above for the entry point

;; 	 (setup-listener uconn (+ port 1)))
;;        #f)
;;    (connect-listener uconn port)))

(define (setup-listener-portlogger uconn)
  (let ((port (portlogger:open-run-close portlogger:find-port)))
    (assert (tt? uconn) "FATAL: setup-listener called with wrong struct "uconn)
    (debug:print 2 *default-log-port* "setup-listener-portlogger got port " port)
    (handle-exceptions
	exn
      (if (< port 65535)
	  (begin
            (debug:print 0 *default-log-port* "setup-listener-portlogger: exception finding port. Retrying")
	    (portlogger:open-run-close portlogger:set-failed port)
	    (thread-sleep! 0.25)
	    (setup-listener-portlogger uconn))
          (begin
            (assert #t "setup-listener-portlogger: could not get a port")
	    #f
          )
      )
      (debug:print 2 *default-log-port* "setup-listener-portlogger: got port " port)
      (connect-listener uconn port))))

(define (connect-listener uconn port)
  ;; (tcp-listener-socket LISTENER)(socket-name so)
  ;; sockaddr-address, sockaddr-port, sockaddr->string
  (let* ((tlsn (tcp-listen port 10000 #f)) ;; (tcp-listen TCPPORT [BACKLOG [HOST]])
	 (addr  (tt:get-best-guess-address (get-host-name)))) ;; (get-my-best-address))) ;; (hostinfo-addresses (host-information (current-hostname)))

;;======================================================================
;; test steps
;;======================================================================

;; teststep-set-status! used to be here

;; NOT NEEDED
(define (test-get-kill-request run-id test-id) ;; run-id test-name itemdat)
#;(define (test-get-kill-request run-id test-id) ;; run-id test-name itemdat)
  (let* ((testdat   (rmt:get-test-state-status-by-id run-id test-id)))
    (and testdat
	 (equal? (car testdat) "KILLREQ"))))

(define (test:tdb-get-rundat-count tdb)
  (if tdb
      (let ((res 0))

  (if (and uname hostname)
      (rmt:general-call 'update-uname-host run-id uname hostname test-id)))
  
;; This one is for running with no db access (i.e. via rmt: internally)
(define (tests:set-full-meta-info db test-id run-id minutes work-area remtries)
;; (define (tests:set-full-meta-info test-id run-id minutes work-area)
;;  (let ((remtries 10))
  (let* ((cpuload  (get-cpu-load))
  (let* ((cpuload  (commonmod:get-cpu-load))
	 (diskfree (get-df (current-directory)))
	 (uname    (get-uname "-srvpio"))
	 (hostname (get-host-name)))
    (tests:update-central-meta-info run-id test-id cpuload diskfree minutes uname hostname)))
    
;; (define (tests:set-partial-meta-info test-id run-id minutes work-area)
#;(define (tests:set-partial-meta-info test-id run-id minutes work-area remtries)

;;======================================================================
;; set up transport, db cache and sync methods
;;
;; sync-method:        'original, 'attach or 'none
;; cache-method:       'tmp 'none
;; rmt:transport-mode: 'http, 'tcp, 'nfs
;;
;; NOTE: NOT ALL COMBINATIONS WORK
;;
;;======================================================================

;; uncomment this block to test without tcp
;; (dbfile:sync-method 'none)
;; (dbfile:cache-method 'none)
;; (rmt:transport-mode 'nfs)

;; uncomment this block to test with tcp
(dbfile:sync-method 'attach) ;; attach) ;; original
(dbfile:cache-method 'tmp)
(rmt:transport-mode 'tcp)

#!/bin/bash

if [ -z "megatest.config" ]; then
    echo "The file 'megatest.config' does not exist. This must be run in a megatest area."
    exit 1
fi
if [ -d ".mtdb" ]; then
    echo "The .mtdb directory already exists. Will not do the conversion"
    exit 1
fi
if [ -d ".megatest" ]; then
    echo "Found a .megatest directory. Will convert from megatest 1.70 to 1.71/1.80 format"
    /p/foundry/env/pkgs/megatest/1.70/16/bin/megatest -list-runs % -dumpmode sexpr > data.sexpr
else 
    if [ -f "megatest.db" ]; then
        echo "Found megatest.db. Will convert from megatest 1.65 to 1.71/1.80 format"
        /p/foundry/env/pkgs/megatest/1.65/92/bin/megatest -list-runs % -dumpmode sexpr > data.sexpr
    else
        echo "Did not find .megatest or megatest.db. Cannot do the conversion"
        exit 1
    fi
fi
which megatest
megatest -import-sexpr data.sexpr