Megatest

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

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

$(PREFIX)/bin/mtest-reaper: helpers/mtest-reaper.scm helpers/ducttape-lib.scm helpers/inteldate.scm helpers/mimetypes.scm
	make -C helpers $@ PREFIX=$(PREFIX) INSTALL=$(INSTALL) ARCHSTR=$(ARCHSTR)

mtest-reaper: $(PREFIX)/bin/mtest-reaper

# install dashboard as dboard so wrapper script can be called dashboard
$(PREFIX)/bin/.$(ARCHSTR)/dboard : dboard $(FILES) utils/mk_wrapper
	utils/mk_wrapper $(PREFIX) dboard $(PREFIX)/bin/dashboard
	chmod a+x $(PREFIX)/bin/dashboard
	$(INSTALL) dboard $(PREFIX)/bin/.$(ARCHSTR)/dboard

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/loadrunner $(PREFIX)/bin/viewscreen $(PREFIX)/bin/mt_xterm \
	  $(PREFIX)/share/docs/megatest_manual.html $(PREFIX)/bin/remrun

$(PREFIX)/bin/.$(ARCHSTR) : 
	mkdir -p $(PREFIX)/bin/.$(ARCHSTR)
	mkdir -p $(PREFIX)/bin/.$(ARCHSTR)/lib

test: tests/tests.scm
	cd tests;csi -I .. -b -n tests.scm

ext-tests/.fslckout : $(MTQA_FOSSIL)
	mkdir -p ext-tests
	cd ext-tests;fossil open --nested $(MTQA_FOSSIL)

	fi
	if csi -ne '(use postgresql)';then \
	   echo "(use postgresql)(hash-table-set! *available-db* 'postgresql #t)" >> altdb.scm;\
	fi

portlogger-example : portlogger-example.scm api.o archive.o client.o common.o configf.o daemon.o dashboard-tests.o db.o dcommon.o ezsteps.o filedb.o genexample.o gutils.o http-transport.o items.o keys.o launch.o lock-queue.o margs.o megatest-version.o mt.o ods.o portlogger.o process.o rmt.o rpc-transport.o runconfig.o runs.o sdb.o server.o synchash.o tasks.o tdb.o tests.o tree.o
	csc $(CSCOPTS) portlogger-example.scm api.o archive.o client.o common.o configf.o daemon.o dashboard-tests.o db.o dcommon.o ezsteps.o filedb.o genexample.o gutils.o http-transport.o items.o keys.o launch.o lock-queue.o margs.o megatest-version.o mt.o ods.o portlogger.o process.o rmt.o rpc-transport.o runconfig.o runs.o sdb.o server.o synchash.o tasks.o tdb.o tests.o tree.o

    get-count-tests-running-for-run-id
    get-run-info
    get-run-status
    get-run-stats
    get-targets
    get-target
    ;; register-run
    get-tests-tags
    get-tests-for-run
    get-test-id
    get-tests-for-runs-mindata
    get-run-name-from-id
    get-runs
    get-num-runs
    get-all-run-ids

    testmeta-get-record
    have-incompletes?
    synchash-get
    ))

(define api:write-queries
  '(
    get-keys-write ;; dummy "write" query to force server start

    ;; SERVERS
    start-server
    kill-server

    ;; TESTS
    test-set-state-status-by-id
    delete-test-records

;;
;;    - returns #( flag result )
;;
(define (api:execute-requests dbstruct dat)
  (handle-exceptions
   exn
   (let ((call-chain (get-call-chain)))
     (debug:print 0 *default-log-port* "WARNING: api:execute-requests received an exception from peer")
     (print-call-chain (current-error-port))
     (debug:print 0 *default-log-port* " message: "  ((condition-property-accessor 'exn 'message) exn))       
     (vector #f (vector exn call-chain dat))) ;; return some stuff for debug if an exception happens
   (if (not (vector? dat))                    ;; it is an error to not receive a vector
       (vector #f #f "remote must be called with a vector")       
       (vector                                   ;; return a vector + the returned data structure
	#t 

	    ((sync-inmem->db)               (let ((run-id (car params)))
					      (db:sync-touched dbstruct run-id force-sync: #t)))
	    ((mark-incomplete)              (apply db:find-and-mark-incomplete dbstruct params))

	    ;; TESTMETA
	    ((testmeta-add-record)       (apply db:testmeta-add-record dbstruct params))
	    ((testmeta-update-field)     (apply db:testmeta-update-field dbstruct params))
            ((get-tests-tags)            (db:get-tests-tags dbstruct))

	    ;; TASKS
	    ((tasks-add)                 (apply tasks:add dbstruct params))   
	    ((tasks-set-state-given-param-key) (apply tasks:set-state-given-param-key dbstruct params))
	    ((tasks-get-last)            (apply tasks:get-last dbstruct params))

	    ;; ARCHIVES
	    ;; ((archive-get-allocations)   
	    ((archive-register-disk)     (apply db:archive-register-disk dbstruct params))
	    ((archive-register-block-name)(apply db:archive-register-block-name dbstruct params))
	    ((archive-allocate-testsuite/area-to-block)(apply db:archive-allocate-testsuite/area-to-block dbstruct block-id testsuite-name areakey))

	    ;;======================================================================
	    ;; READ ONLY QUERIES
	    ;;======================================================================

	    ;; KEYS
	    ((get-key-val-pairs)               (apply db:get-key-val-pairs dbstruct params))
	    ((get-keys)                        (db:get-keys dbstruct))
            ((get-keys-write)                        (db:get-keys dbstruct)) ;; force a dummy "write" query to force server
	    ((get-key-vals)                    (apply db:get-key-vals dbstruct params))
	    ((get-target)                      (apply db:get-target dbstruct params))
	    ((get-targets)                     (db:get-targets dbstruct))

	    ;; ARCHIVES
	    ((test-get-archive-block-info)     (apply db:test-get-archive-block-info dbstruct params))
	    

	    ((get-steps-data)               (apply db:get-steps-data dbstruct params))
	    ((get-steps-for-test)           (apply db:get-steps-for-test dbstruct params))

	    ;; TEST DATA
	    ((read-test-data)               (apply db:read-test-data dbstruct params))

	    ;; MISC
            ((get-latest-host-load)         (apply db:get-latest-host-load dbstruct params))
	    ((have-incompletes?)            (apply db:have-incompletes? dbstruct params))
	    ((login)                        (apply db:login dbstruct params))
	    ((general-call)                 (let ((stmtname   (car params))
						  (run-id     (cadr params))
						  (realparams (cddr params)))
					      (db:with-db dbstruct run-id #t ;; these are all for modifying the db
							  (lambda (db)

    ok))

(define (client:connect iface port)
  (case (server:get-transport)
    ((rpc)  (rpc:client-connect  iface port))
    ((http) (http:client-connect iface port))
    ((zmq)  (zmq:client-connect  iface port))
    (else
     (debug:print 0 *default-log-port* "ERROR: transport " (remote-transport *runremote*) " not supported (5)")
     (exit))))

(define (client:setup  run-id #!key (remaining-tries 10) (failed-connects 0))
  (case (server:get-transport)
    ((rpc) (let ((res (client:setup-rpc run-id remaining-tries: remaining-tries)))
             (remote-conndat-set! *runremote* res)
             res))
    ((http)(client:setup-http run-id remaining-tries: remaining-tries failed-connects: failed-connects))
    (else
     (debug:print 0 *default-log-port* "ERROR: transport " (remote-transport *runremote*) " not supported (6)")
     (exit)))) ;; (client:setup-rpc run-id))))

;; (define (client:login-no-auto-setup server-info run-id)
;;   (case (server:get-transport)
;;     ((rpc)  (rpc:login-no-auto-client-setup server-info run-id))
;;     ((http) (rmt:login-no-auto-client-setup server-info run-id))
;;     (else   (rpc:login-no-auto-client-setup server-info run-id))))
;; 

;;   2. We are a run tests, list runs or other interactive process and we must figure out
;;      *transport-type* and *runremote* from the monitor.db
;;
;; client:setup
;;
;; lookup_server, need to remove *runremote* stuff
;;

(define (client:setup-rpc run-id #!key (remaining-tries 10) (failed-connects 0))
  (debug:print-info 2 *default-log-port* "client:setup-rpc remaining-tries=" remaining-tries)
  (let* ((server-dat (tasks:get-server (db:delay-if-busy (tasks:open-db)) run-id))
         (num-available (tasks:num-in-available-state (db:delay-if-busy (tasks:open-db)) run-id)))
    (cond
     ((<= remaining-tries 0)
      (debug:print-error 0 *default-log-port* "failed to start or connect to server for run-id " run-id)
      (exit 1))
     (server-dat
      (debug:print-info 4 *default-log-port* "client:setup-rpc server-dat=" server-dat ", remaining-tries=" remaining-tries)

      (rpc-transport:client-setup run-id server-dat remaining-tries: remaining-tries))
     (else
      (if (< num-available 2)
          (server:try-running run-id))
      (thread-sleep! (+ 2 (random (- 20 remaining-tries)))) ;; give server a little time to start up, randomize a little to avoid start storms.
      (client:setup run-id remaining-tries: (- remaining-tries 1))))))


(define (client:setup-http run-id #!key (remaining-tries 10) (failed-connects 0))
  (debug:print-info 2 *default-log-port* "client:setup remaining-tries=" remaining-tries)
  (let* ((tdbdat (tasks:open-db)))
    (if (<= remaining-tries 0)
	(begin
	  (debug:print-error 0 *default-log-port* "failed to start or connect to server for run-id " run-id)
	  (exit 1))
	(let* ((server-dat (tasks:get-server (db:delay-if-busy tdbdat) run-id)))
	  (debug:print-info 4 *default-log-port* "client:setup server-dat=" server-dat ", remaining-tries=" remaining-tries)
	  (if server-dat
	      (let* ((iface     (tasks:hostinfo-get-interface server-dat))
		     (hostname  (tasks:hostinfo-get-hostname  server-dat))
		     (port      (tasks:hostinfo-get-port      server-dat))

		     (start-res (http-transport:client-connect iface port))



		     (ping-res  (rmt:login-no-auto-client-setup start-res)))






		(if (and start-res
			 ping-res)
		    (begin
		      (remote-conndat-set! *runremote* start-res) ;; (hash-table-set! *runremote* run-id start-res)
		      (debug:print-info 2 *default-log-port* "connected to " (http-transport:server-dat-make-url start-res))
		      start-res)
		    (begin    ;; login failed but have a server record, clean out the record and try again

		(let ((num-available (tasks:num-in-available-state (db:dbdat-get-db tdbdat) run-id)))
		  (debug:print-info 0 *default-log-port* "client:setup, no server registered, remaining-tries=" remaining-tries " num-available=" num-available)
		  (if (< num-available 2)
		      (server:try-running run-id))
		  (thread-sleep! (+ 5 (random (- 20 remaining-tries))))  ;; give server a little time to start up, randomize a little to avoid start storms.
		  (client:setup run-id remaining-tries: (- remaining-tries 1)))))))))

;; keep this as a function to ease future ;; this is unused, not porting for rpc -BB
(define (client:start run-id server-info)
  (http-transport:client-connect (tasks:hostinfo-get-interface server-info)
				 (tasks:hostinfo-get-port server-info)))

;; ;; client:signal-handler
;; (define (client:signal-handler signum)
;;   (signal-mask! signum)

(import (prefix sqlite3 sqlite3:))
(import (prefix base64 base64:))

(declare (unit common))

(include "common_records.scm")
(include "thunk-utils.scm")


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

;; (define old-exit exit)
;; 
;; (define (exit . code)

  (mutex-lock! *context-mutex*)
  (let ((cxt (hash-table-ref/default *contexts* toppath #f)))
    (if (not cxt)
        (set! cxt (let ((x (make-cxt)))(hash-table-set! *contexts* toppath x) x)))
    (let ((cxt-mutex (cxt-mutex cxt)))
      (mutex-unlock! *context-mutex*)
      (mutex-lock! cxt-mutex)
      ;; here we guard proc with exception handler so
      ;; no matter how proc succeeds or fails,
      ;; the cxt-mutex will be unlocked afterward.
      (let* ((EXCEPTION-SYMBOL (gensym)) ;; use a generated symbol
             (guarded-proc               ;; to avoid collision
              (lambda args
                (let* ((res (condition-case
                             (apply proc args)
                             [x () (cons EXCEPTION-SYMBOL x)])))
                 (mutex-unlock! cxt-mutex)
                 (if (and (pair? res) (eq? (car res) EXCEPTION))
                     (abort (cdr res))
                     res)))))
        (guarded-proc cxt)))))
        
(define *db-keys* #f)

(define *configinfo*   #f)   ;; raw results from setup, includes toppath and table from megatest.config
(define *runconfigdat* #f)   ;; run configs data
(define *configdat*    #f)   ;; megatest.config data
(define *configstatus* #f)   ;; status of data; 'fulldata : all processing done, #f : no data yet, 'partialdata : partial read done

;; 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*
;; task db
(define *task-db*             #f) ;; (vector db path-to-db)
(define *db-access-allowed*   #t) ;; flag to allow access
(define *db-access-mutex*     (make-mutex))
(define *db-cache-path*       #f)

;; SERVER
(define *my-client-signature* #f)
(define *transport-type*  #f)             ;; override with [server] transport http|rpc|nmsg

(define *DEFAULT-TRANSPORT* "http")
(define (common:set-transport-type)
  (set! *transport-type*
        (string->symbol
         (or
          (args:get-arg "-transport")
          (configf:lookup *configdat* "server" "transport")
          *DEFAULT-TRANSPORT*)))
  *transport-type*)
  
(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)
(define *time-to-exit*      #f)
(define *server-run*        #t)

(define *test-ids*          (make-hash-table)) ;; cache run-id, testname, and item-path => test-id
(define *test-info*         (make-hash-table)) ;; cache the test info records, update the state, status, run_duration etc. from testdat.db

(define *run-info-cache*     (make-hash-table)) ;; run info is stable, no need to reget
(define *launch-setup-mutex* (make-mutex))     ;; need to be able to call launch:setup often so mutex it and re-call the real deal only if *toppath* not set
(define *homehost-mutex*     (make-mutex))

(defstruct remote
  (hh-dat            (common:get-homehost)) ;; homehost record ( addr . hhflag )
  (server-url        (if *toppath* (server:read-dotserver *toppath*))) ;; (server:check-if-running *toppath*) #f))
  (last-server-check 0)  ;; last time we checked to see if the server was alive
  (conndat           #f)
  (transport         *transport-type*)
  (server-timeout    (or (server:get-timeout) 100))) ;; default to 100 seconds

;; launching and hosts
(defstruct host
  (reachable    #f)
  (last-update  0)
  (last-used    0)
  (last-cpuload 1))

  (message-digest-string (md5-primitive) *toppath*))

;;======================================================================
;; E X I T   H A N D L I N G
;;======================================================================

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

;;   (let ((ohh (common:on-homehost?))
;; 	(srv (args:get-arg "-server")))
;;     (and ohh srv)))
    ;; (debug:print-info 0 *default-log-port* "common:run-sync? ohh=" ohh ", srv=" srv)

;;;; run-ids
;;    if #f use *db-local-sync* : or 'local-sync-flags
;;    if #t use timestamps      : or 'timestamps
(define (common:sync-to-megatest.db dbstruct) 
  (let ((start-time         (current-seconds))
	(res                (db:multi-db-sync dbstruct 'new2old)))
    (let ((sync-time (- (current-seconds) start-time)))
      (debug:print-info 3 *default-log-port* "Sync of newdb to olddb completed in " sync-time " seconds pid="(current-process-id))
      (if (common:low-noise-print 30 "sync new to old")
	  (debug:print-info 0 *default-log-port* "Sync of newdb to olddb completed in " sync-time " seconds pid="(current-process-id))))
    res))

;; currently the primary job of the watchdog is to run the sync back to megatest.db from the db in /tmp
;; if we are on the homehost and we are a server (by definition we are on the homehost if we are a server)
;;
(define (common:watchdog)
  (thread-sleep! 0.05) ;; delay for startup
  (let ((legacy-sync (common:run-sync?))
	(debug-mode  (debug:debug-mode 1))
	(last-time   (current-seconds)))
    (debug:print-info 0 *default-log-port* "watchdog starting. legacy-sync is " legacy-sync" pid="(current-process-id))
    (if legacy-sync
	(let ((dbstruct (db:setup)))
	  (debug:print-info 0 *default-log-port* "Server running, periodic sync started.")
	  (let loop ()
            ;;(BB> "watchdog loop.  pid="(current-process-id))
	    ;; sync for filesystem local db writes
	    ;;
	    (mutex-lock! *db-multi-sync-mutex*)
	    (let* ((need-sync        (>= *db-last-access* *db-last-sync*)) ;; no sync since last write
		   (sync-in-progress *db-sync-in-progress*)
		   (should-sync      (> (- (current-seconds) *db-last-sync*) 5)) ;; sync every five seconds minimum
		   (will-sync        (and (or need-sync should-sync)
					  (not sync-in-progress)))
		   (start-time       (current-seconds)))
	      ;; (debug:print-info 0 *default-log-port* "need-sync: " need-sync " sync-in-progress: " sync-in-progress " should-sync: " should-sync " will-sync: " will-sync)
	      (if will-sync (set! *db-sync-in-progress* #t))

		  (if (and (not *time-to-exit*)
			   (< count 4)) ;; was 11, changing to 4. 
		      (begin
			(thread-sleep! 1)
			(delay-loop (+ count 1))))
		  (loop)))
	    (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)
  (on-exit (lambda () 0))
  (let ((no-hurry  (if *time-to-exit* ;; hurry up
		       #f
		       (begin
			 (set! *time-to-exit* #t)
			 #t))))
    (debug:print-info 4 *default-log-port* "starting exit process, finalizing databases.")
    (if (and no-hurry (debug:debug-mode 18))
	(rmt:print-db-stats))
    (let ((th1 (make-thread (lambda () ;; thread for cleaning up, give it five seconds
                              (if *dbstruct-db* (db:close-all *dbstruct-db*)) ;; one second allocated
			      (if *task-db*    
				  (let ((db (cdr *task-db*)))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! *task-db* 0 #f)
					  (set! *task-db* #f)))))
                              (if (and *runremote*
                                       (remote-conndat *runremote*))
                                  (begin
                                    (http-client#close-all-connections!))) ;; for http-client
                              (if (not (eq? *default-log-port* (current-error-port)))
                                  (close-output-port *default-log-port*))
			      (set! *default-log-port* (current-error-port))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 *default-log-port* "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
                                  (begin
                                    (thread-sleep! 5)) ;; give the clean up few seconds to do it's stuff
                                  (begin
      				  (thread-sleep! 2)))
      			      (debug:print 4 *default-log-port* " ... done")
      			      )
			    "clean exit")))

      ;; let's try to clean up open sockets
      (if *runremote*
          (case (remote-transport *runremote*)
            ((http) #t)
            ((rpc)  (rpc:close-all-connections!))
            (else
             (debug:print-info 0 *default-log-port* "Transport "(remote-transport *runremote*)" not supported"))))

      (thread-start! th1)
      (thread-start! th2)
      (thread-join! th1)
      )
    )

  0)

(define (std-signal-handler signum)
  ;; (signal-mask! signum)
  (set! *time-to-exit* #t)
  (debug:print-error 0 *default-log-port* "Received signal " signum " exiting promptly")
  ;; (std-exit-procedure) ;; shouldn't need this since we are exiting and it will be called anyway
  (exit))

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

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

(define (common:args-get-testpatt rconf)
  (let* ((tagexpr (args:get-arg "-tagexpr"))
         (tags-testpatt (if tagexpr (string-join (runs:get-tests-matching-tags tagexpr) ",") #f))
         (testpatt-key  (if (args:get-arg "-mode") (args:get-arg "-mode") "TESTPATT"))
         (args-testpatt (or (args:get-arg "-testpatt") (args:get-arg "-runtests") "%"))


         (rtestpatt     (if rconf (runconfigs-get rconf testpatt-key) #f)))
    (cond
     (tags-testpatt
      (debug:print-info 0 *default-log-port* "-tagexpr "tagexpr" selects testpatt "tags-testpatt)
      tags-testpatt)
     ((and (equal? args-testpatt "%") rtestpatt)
      (debug:print-info 0 *default-log-port* "testpatt defined in "testpatt-key" from runconfigs: " rtestpatt)
      rtestpatt)
     (else args-testpatt))))
     
(define (common:get-linktree)
  (or (getenv "MT_LINKTREE")
      (if *configdat*
	  (configf:lookup *configdat* "setup" "linktree"))))

(define (common:args-get-runname)
  (let ((res (or (args:get-arg "-runname")

	   (with-input-from-pipe 
	    (conc "ssh " remote-host " cat /proc/loadavg")
	    (lambda ()(list (read)(read)(read)))))
      (with-input-from-file "/proc/loadavg" 
	(lambda ()(list (read)(read)(read))))))

;; get normalized cpu load by reading from /proc/loadavg and /proc/cpuinfo return all three values and the number of real cpus and the number of threads
;; returns alist '((adj-cpu-load . normalized-proc-load) ... etc.
;;  keys: adj-proc-load, adj-core-load, 1m-load, 5m-load, 15m-load
;;
(define (common:get-normalized-cpu-load remote-host)
  (let ((data (if remote-host
                  (with-input-from-pipe 
                   (conc "ssh " remote-host " cat /proc/loadavg;cat /proc/cpuinfo;echo end")
                   read-lines)
                  (append 

(define (common:unix-ping hostname)
  (let ((res (system (conc "ping -c 1 " hostname " > /dev/null"))))
    (eq? res 0)))

;; ideally put all this info into the db, no need to preserve it across moving homehost
;;
;; return list of
;;  ( reachable? cpuload update-time )
(define (common:get-host-info hostname)
  (let* ((loadinfo (rmt:get-latest-host-load hostname))
         (load (car loadinfo))
         (load-sample-time (cdr loadinfo))
         (load-sample-age (- (current-seconds) load-sample-time))
         (loadinfo-timeout-seconds 20)
         (host-last-update-timeout-seconds 10)
         (host-rec (hash-table-ref/default *host-loads* hostname #f))
         )
    (cond
     ((< load-sample-age loadinfo-timeout-seconds)
      (list #t
            load-sample-time
            load))
     ((and host-rec
           (< (current-seconds) (+ (host-last-update host-rec) host-last-update-timeout-seconds)))
      (list #t
            (host-last-update host-rec)
            (host-last-cpuload host-rec )))
     ((common:unix-ping hostname)
      (list #t
            (current-seconds)
            (alist-ref 'adj-core-load (common:get-normalized-cpu-load hostname))))
     (else
      (list #f 0 -1)))))
    
(define (common:update-host-loads-table hosts-raw)
  (let* ((hosts (filter (lambda (x)
                          (string-match (regexp "^\\S+$") x))
                        hosts-raw)))










    (for-each
     (lambda (hostname)
       (let* ((rec       (let ((h (hash-table-ref/default *host-loads* hostname #f)))
                          (if h
                              h
                              (let ((h (make-host)))
                                (hash-table-set! *host-loads* hostname h)
                                h))))

              (host-info         (common:get-host-info hostname))
              (is-reachable      (car host-info))
              (last-reached-time (cadr host-info))

              (load              (caddr host-info)))
         (host-reachable-set!    rec is-reachable)
         (host-last-update-set!  rec last-reached-time)
         (host-last-cpuload-set! rec load)))
     hosts)))

(define (common:get-least-loaded-host hosts-raw)
  (let* ((hosts (filter (lambda (x)
                          (string-match (regexp "^\\S+$") x))
                        hosts-raw))
         (best-host #f)
         (best-load 99999)
         (curr-time (current-seconds)))
    (common:update-host-loads-table hosts)
    (for-each
     (lambda (hostname)
       (let* ((rec
               (let ((h (hash-table-ref/default *host-loads* hostname #f)))
                 (if h
                     h
                     (let ((h (make-host)))
                       (hash-table-set! *host-loads* hostname h)
                       h))))
              (reachable (host-reachable rec))
              (load      (host-last-cpuload   rec)))
         (cond
          ((not reachable) #f)
          ((< (+ load (/ (random 250) 1000))         ;; add a random factor to keep from getting in a rut
              (+ best-load (/ (random 250) 1000))  )
           (set! best-load load)




           (set! best-host hostname)))))
     hosts)
    best-host))




(define (common:wait-for-cpuload maxload numcpus waitdelay #!key (count 1000) (msg #f)(remote-host #f))
  (let* ((loadavg (common:get-cpu-load remote-host))
	 (first   (car loadavg))
	 (next    (cadr loadavg))
	 (adjload (* maxload numcpus))
	 (loadjmp (- first next)))

       (let* ((this-loc (vector-ref frame 0))
              (this-func (cadr (string-split this-loc " "))))
         (if (equal? this-func "BB>")
             (set! location this-loc))))
     stack)
    (let ((dp-args (append (list 0 *default-log-port* location"   "  ) in-args)))
      (apply debug:print dp-args))))

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



;; register hash tables with BBpp.
(hash-table-set! *BBpp_custom_expanders_list* HASH_TABLE:
                 (cons hash-table? hash-table->alist))

;; test name converter
(define (BBpp_custom_converter arg)
  (let ((res #f))
    (for-each
     (lambda (custom-type-name)
       (let* ((custom-type-info      (hash-table-ref *BBpp_custom_expanders_list* custom-type-name))
              (custom-type-test      (car custom-type-info))
              (custom-type-converter (cdr custom-type-info)))
         (when (and (not res) (custom-type-test arg))
           (set! res (custom-type-converter arg)))))
     (hash-table-keys *BBpp_custom_expanders_list*))
    (if res (BBpp_ res) arg)))

(define (BBpp_ arg)
  (cond
   ;;((SOMESTRUCT? arg) (cons SOMESTRUCT: (SOMESTRUCT->alist arg)))
   ;;((dboard:tabdat? arg) (cons dboard:tabdat: (dboard:tabdat->alist arg)))
   ((hash-table? arg)
    (let ((al (hash-table->alist arg)))
      (BBpp_ (cons HASH_TABLE: al))))
   ((null? arg) '())
   ;;((list? arg) (cons (BBpp_ (car arg)) (BBpp_ (cdr arg))))
   ((pair? arg) (cons (BBpp_ (car arg)) (BBpp_ (cdr arg))))
   (else (BBpp_custom_converter arg))))

;; Brandon's pretty printer.  It expands hashes and custom types in addition to regular pp
(define (BBpp arg)
  (pp (BBpp_ arg)))

;(use define-macro)
(define-syntax inspect
  (syntax-rules ()
    [(_ x)
    ;; (with-output-to-port (current-error-port)
       (printf "~a is: ~a\n" 'x (with-output-to-string (lambda () (BBpp x))))
     ;;  )
     ]
    [(_ x y ...) (begin (inspect x) (inspect y ...))]))

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

  (runs-summary-source-runname-label #f)
  (runs-summary-dest-runname-label #f)
  ;; runs summary view
  
  tests-tree       ;; used in newdashboard
  )

;; register tabdat with BBpp
;; this is used by BBpp (Brandon's pretty printer) to convert dboard:tabdat into a composition of lists that pp will handle
(hash-table-set! *BBpp_custom_expanders_list* TABDAT:
                 (cons dboard:tabdat?
                       (lambda (tabdat-item)
                         (filter
                          (lambda (alist-entry)
                            (member (car alist-entry)
                                    '(allruns-by-id allruns))) ;; FIELDS OF INTEREST
                          (dboard:tabdat->alist tabdat-item)))))

(define (dboard:tabdat-target-string vec)
  (let ((targ (dboard:tabdat-target vec)))
    (if (list? targ)(string-intersperse targ "/") "no-target-specified")))

(define (dboard:tabdat-test-patts-use vec)    
  (let ((val (dboard:tabdat-test-patts vec)))(if val val ""))) ;;RADT => What is the if for?

  ((tests-by-name (make-hash-table)) : hash-table) ;; hash of testfullname => testdat
  key-vals
  ((last-update   0)                 : fixnum) ;; last query to db got records from before last-update
  ((data-changed  #f)                : boolean)
  ((run-data-offset  0)              : number)      ;; get only 100 items per call, set back to zero when received less that 100 items
  (db-path #f)
  )

;; register dboard:rundat with BBpp
;; this is used by BBpp (Brandon's pretty printer) to convert dboard:rundat into a composition of lists that pp will handle
(hash-table-set! *BBpp_custom_expanders_list* RUNDAT:
                 (cons dboard:rundat?
                       (lambda (tabdat-item)
                         (filter
                          (lambda (alist-entry)
                            (member (car alist-entry)
                                    '(run run-data-offset ))) ;; FIELDS OF INTEREST
                          (dboard:rundat->alist tabdat-item)))))




(define (dboard:rundat-make-init #!key (run #f)(key-vals #f)(tests #f));; -100 is before time began
  (make-dboard:rundat 
   run: run
   tests: (or tests (make-hash-table))
   key-vals: key-vals 
   )) 

	 (start-time  (current-seconds))
	 (runs-hash   (let ((ht (make-hash-table)))
			 (for-each (lambda (run)
				     (hash-table-set! ht (db:get-value-by-header run header "id") run))
				   runs-tree) ;; (vector-ref runs-dat 1))
			 ht))
	 (tb          (dboard:tabdat-runs-tree tabdat)))
    ;;(BB> "In update-rundat")
    ;;(inspect allruns runs-hash)
    (dboard:tabdat-last-runs-update-set! tabdat (- (current-seconds) 2))
    (dboard:tabdat-header-set! tabdat header)
    ;; 
    ;; trim runs to only those that are changing often here
    ;; 
    (if (null? runs)
	(begin

	    (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)
		  (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

     (let* ((tabdat (dboard:common-get-tabdat commondat tab-num: tab-num))
	    (dbkeys (dboard:tabdat-dbkeys tabdat)))
       ;;(print "RA => calling runs-tab-updater with commondat " commondat " tab-num " tab-num)
       ;;(tabdat-values tabdat) ;;RA added 
       ;; (pp (dboard:tabdat->alist tabdat))
       ;; (if (dashboard:database-changed? commondat tabdat context-key: 'runs-rundat)      
       (dashboard:do-update-rundat tabdat)
       ;;(BB> "dashboard:runs-tab-updater")
       ;;(inspect tabdat)

       (let ((uidat (dboard:commondat-uidat commondat)))
	 ;;(print "RA => Calling update-buttons with tabdat : " tabdat " uidat " uidat)
	 (update-buttons tabdat uidat (dboard:tabdat-numruns tabdat) (dboard:tabdat-num-tests tabdat)))
       ))
   "dashboard:runs-tab-updater"))

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

    (debug:print-info 4 *default-log-port* "Syncing for run-id: " run-id)
    (mutex-lock! *db-multi-sync-mutex*)
    (let ((update_info (cons (if force-sync 0 *db-last-sync*) "last_update")))
      (mutex-unlock! *db-multi-sync-mutex*)
      (db:sync-tables (db:sync-all-tables-list dbstruct) update_info tmpdb refndb mtdb))
    (mutex-lock! *db-multi-sync-mutex*)
    (set! *db-last-sync* start-t)
    (set! *db-last-access* start-t)
    (mutex-unlock! *db-multi-sync-mutex*)))

;; close all opened run-id dbs
(define (db:close-all dbstruct)
  (if (dbr:dbstruct? dbstruct)
      (begin
        ;; (db:sync-touched dbstruct 0 force-sync: #t) ;; NO. Do not do this here. Instead we rely on a server to be started when there are writes, even if the server itself is not going to be used as a server.

	   ;;      			    (> (- (current-seconds)(file-modification-time tdatpath)) 600)))) ;; no change in 10 minutes to testdat.db - she's dead Jim
	   ;;      		    incompleted))
	   (min-incompleted-ids (map car incompleted)) ;; do 'em all
	   (all-ids             (append min-incompleted-ids (map car oldlaunched))))
      (if (> (length all-ids) 0)
	  (begin
	    (debug:print 0 *default-log-port* "WARNING: Marking test(s); " (string-intersperse (map conc all-ids) ", ") " as INCOMPLETE")
            (for-each
             (lambda (test-id)
               (db:test-set-status-state dbstruct run-id test-id "COMPLETE" "DEAD" "Test failed to complete"))
             all-ids))))))

;; ALL REPLACED BY THE BLOCK ABOVE
;;
;; 	    (sqlite3:execute 
;; 	     db
;; 	     (conc "UPDATE tests SET state='INCOMPLETE' WHERE run_id=? AND id IN (" 
;; 		   (string-intersperse (map conc all-ids) ",")
;; 		   ");")
;;              run-id))))
;; 
;;     ;; Now do rollups for the toplevel tests
;;     ;;
;;     ;; (db:delay-if-busy dbdat)
;;     (for-each
;;      (lambda (toptest)
;;        (let ((test-name (list-ref toptest 3)))
;; ;;	     (run-id    (list-ref toptest 5)))
;; 	 (db:top-test-set-per-pf-counts dbstruct run-id test-name)))
;;      toplevels)))

;; BUG: Probably broken - does not explicitly use run-id in the query
;;
(define (db:top-test-set-per-pf-counts dbstruct run-id test-name)
  (db:general-call (db:get-db dbstruct run-id) 'top-test-set-per-pf-counts (list test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name))) 
 
		     

	     (lambda (state status count)
	       (let ((netstate (if (equal? state "COMPLETED") status state)))
		 (if (string? netstate)
		     (begin
		       (hash-table-set! totals netstate (+ (hash-table-ref/default totals netstate 0) count))
		       (hash-table-set! curr   netstate (+ (hash-table-ref/default curr   netstate 0) count))))))
	     db
	     "SELECT state,status,count(id) FROM tests AS t WHERE run_id=? GROUP BY state,status ORDER BY state,status DESC;"
             run-id)
	    ;; add the per run counts to res
	    (for-each (lambda (state)
			(set! res (cons (list run-name state (hash-table-ref curr state)) res)))
		      (sort (hash-table-keys curr) string>=))
	    (set! curr (make-hash-table))))))
     runs-info)
    (for-each (lambda (state)

		      (string-intersperse testnames "','")
		      "') AND NOT (uname = 'n/a' AND item_path='');")) ;; should this include the (uname = 'n/a' ...) ???
	       ))
	    0)))))
             ;; DEBUG FIXME - need to merge this v.155 query correctly   
             ;; AND testname in (SELECT testname FROM test_meta WHERE jobgroup=?)
             ;; AND NOT (uname = 'n/a' AND item_path = '');"

;; tags: '("tag%" "tag2" "%ag6")
;;

;; done with run when:
;;   0 tests in LAUNCHED, NOT_STARTED, REMOTEHOSTSTART, RUNNING
(define (db:estimated-tests-remaining dbstruct run-id)
  (db:with-db
   dbstruct
   run-id

    (sqlite3:for-each-row
     (lambda (state status count)
       (set! res (cons (vector state status count) res)))
     db
     "SELECT state,status,count(state) FROM tests WHERE run_id=? AND testname=? AND item_path='' GROUP BY state,status;"
     run-id testname)
    res))


(define (db:get-latest-host-load dbstruct raw-hostname)
  (let* ((hostname (string-substitute "\\..*$" "" raw-hostname))
        (res  (cons -1 0))
        (mydb (db:dbdat-get-db (db:get-db dbstruct 0)))
        )
    (db:with-db
     dbstruct
     0
     #f
     (lambda (db)
       (sqlite3:for-each-row
        (lambda (cpuload update-time)  (set! res (cons cpuload update-time)))
        db
        "SELECT tr.cpuload, tr.update_time FROM test_rundat tr, tests t WHERE t.host=? AND tr.cpuload != -1  AND tr.test_id=t.id ORDER BY tr.update_time DESC LIMIT 1;"
        hostname))) res ))


(define (db:set-top-level-from-items dbstruct run-id testname)
  (let* ((dbdat (db:get-db dbstruct run-id))
	 (db    (db:dbdat-get-db dbdat))
	 (summ  (db:get-state-status-summary db run-id testname))
	 (find  (lambda (state status)
		  (if (null? summ) 

	"SELECT id,item_path,state,status,run_duration,final_logf,comment FROM tests WHERE testname=? AND item_path != '';"
	test-name)
       res))))

;;======================================================================
;; Tests meta data
;;======================================================================

;; returns a hash table of tags to tests
;;
(define (db:get-tests-tags dbstruct)
  (let* ((dbdat   (db:get-db dbstruct #f))
	 (db      (db:dbdat-get-db dbdat))
         (res     (make-hash-table)))
    (sqlite3:for-each-row
     (lambda (testname tags-in)
       (let ((tags (string-split tags-in ",")))
         (for-each
          (lambda (tag)
            (hash-table-set! res tag
                             (delete-duplicates
                              (cons testname (hash-table-ref/default res tag '())))))
          tags)))
     db
     "SELECT testname,tags FROM test_meta")
    res))

;; read the record given a testname
(define (db:testmeta-get-record dbstruct testname)
  (let ((res   #f))
    (db:with-db
     dbstruct
     #f

				      (res #f))
				 (cond
				  ((equal? (uri-path (request-uri (current-request)))
					   '(/ "api"))
				   (send-response body:    (api:process-request *dbstruct-db* $) ;; the $ is the request vars proc
						  headers: '((content-type text/plain)))
				   (mutex-lock! *heartbeat-mutex*)
				   (set! *db-lastaccess* (current-seconds))
				   (mutex-unlock! *heartbeat-mutex*))
				  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ ""))
				   (send-response body: (http-transport:main-page)))
				  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ "json_api"))
				   (send-response body: (http-transport:main-page)))

;;
(define (http-transport:client-api-send-receive run-id serverdat cmd params #!key (numretries 3))
  (let* ((fullurl    (if (vector? serverdat)
			 (http-transport:server-dat-get-api-req serverdat)
			 (begin
			   (debug:print 0 *default-log-port* "FATAL ERROR: http-transport:client-api-send-receive called with no server info")
			   (exit 1))))
	 (res        (vector #f "uninitialized"))
	 (success    #t)
	 (sparams    (db:obj->string params transport: 'http)))
       (debug:print-info 11 *default-log-port* "fullurl=" fullurl ", cmd=" cmd ", params=" params ", run-id=" run-id "\n")
       ;; set up the http-client here
       (max-retry-attempts 1)
       ;; consider all requests indempotent
       (retry-request? (lambda (request)

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

;; run http-transport:keep-running in a parallel thread to monitor that the db is being 
;; used and to shutdown after sometime if it is not.
;;
(define (http-transport:keep-running server-id run-id)
  ;; if none running or if > 20 seconds since 

		      (begin
			(tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "dbprep")
			(thread-sleep! 0.5) ;; give some margin for queries to complete before switching from file based access to server based access
			(set! *dbstruct-db*  (db:setup)) ;;  run-id))
			(set! server-going #t)
			(tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "running")
			(server:write-dotserver *toppath* (conc iface ":" port))
                        (server:dotserver-starting-remove))
                      (begin ;; gotta exit nicely
			(tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "collision")
			(http-transport:server-shutdown server-id port))))))

      ;; when things go wrong we don't want to be doing the various queries too often
      ;; so we strive to run this stuff only every four seconds or so.
      (let* ((sync-time (- (current-milliseconds) start-time))
	    (rem-time  (quotient (- 4000 sync-time) 1000)))

;; (if (and (<= rem-time 4)
;; 	     (> rem-time 0))
;; 	(thread-sleep! rem-time)
;; 	(thread-sleep! 4))) ;; fallback for if the math is changed ...

(define (http-transport:server-shutdown server-id port)
  (let ((tdbdat (tasks:open-db)))
    (debug:print-info 0 *default-log-port* "Starting to shutdown the server. pid="(current-process-id))
    ;;
    ;; start_shutdown
    ;;
    (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "shutting-down")
    (set! *time-to-exit* #t) ;; tell on-exit to be fast as we've already cleaned up
    (portlogger:open-run-close portlogger:set-port port "released")
    (thread-sleep! 5)

    (exit)))

;; all routes though here end in exit ...
;;
;; start_server? 
;;
(define (http-transport:launch run-id)

  (server:dotserver-starting)


  (let* ((tdbdat (tasks:open-db)))
    (set! *run-id*   run-id)
    (if (args:get-arg "-daemonize")
	(begin
	  (daemon:ize)
	  (if *alt-log-file* ;; we should re-connect to this port, I think daemon:ize disrupts it
	      (begin
		(current-error-port *alt-log-file*)
		(current-output-port *alt-log-file*)))))
    (if (and (server:read-dotserver *toppath*)
             (server:check-if-running run-id))
	(begin
	  (debug:print 0 *default-log-port* "INFO: Server for run-id " run-id " already running")
	  (exit 0))
	(begin ;; ok, no server detected, clean out any lingering records
	   (tasks:server-force-clean-running-records-for-run-id  (db:delay-if-busy tdbdat) run-id "notresponding")))
    (let loop ((server-id (tasks:server-lock-slot (db:delay-if-busy tdbdat) run-id 'http))
	       (remtries  4))
      (if (not server-id)
	  (if (> remtries 0)
	      (begin
		(thread-sleep! 2)
		(loop (tasks:server-lock-slot (db:delay-if-busy tdbdat) run-id 'http)
		      (- remtries 1)))
	      (begin
		;; since we didn't get the server lock we are going to clean up and bail out
		(debug:print-info 2 *default-log-port* "INFO: server pid=" (current-process-id) ", hostname=" (get-host-name) " not starting due to other candidates ahead in start queue")
		(tasks:server-delete-records-for-this-pid (db:delay-if-busy tdbdat) " http-transport:launch")
                
                (server:dotserver-starting-remove)
		))
	  (let* ((th2 (make-thread (lambda ()
				     (debug:print-info 0 *default-log-port* "Server run thread started")
				     (http-transport:run 
				      (if (args:get-arg "-server")
					  (args:get-arg "-server")
					  "-")

	       " ms</td></tr>"
	       "<tr><td>Number non-cached queries</td> <td>"  *number-non-write-queries* "</td></tr>"
	       "<tr><td>Average non-cached time</td>   <td>" (if (eq? *number-non-write-queries* 0)
								 "n/a (no queries)"
								 (/ *total-non-write-delay* 
								    *number-non-write-queries*))
	       " ms</td></tr>"
	       "<tr><td>Last access</td><td>"              (seconds->time-string *last-db-access*) "</td></tr>"
	       "</table>")))
    (mutex-unlock! *heartbeat-mutex*)
    res))

(define (http-transport:runs linkpath)
  (conc "<h3>Runs</h3>"
	(string-intersperse

    (rmt:teststep-set-status! run-id test-id stepname "start" "-" #f #f)
    ;; now launch the actual process
    (call-with-environment-variables 
     (list (cons "PATH" (conc (get-environment-variable "PATH") ":.")))
     (lambda () ;; (process-run "/bin/bash" "-c" "exec ls -l /tmp/foobar > /tmp/delme-more.log 2>&1")
       (let* ((cmd (conc stepcmd " > " stepname ".log 2>&1")) ;; >outfile 2>&1 
	      (pid (process-run "/bin/bash" (list "-c" cmd))))

         (with-output-to-file "Makefile.ezsteps"
           (lambda ()
             (print stepname ".log :")
             (print "\t" cmd)
             (if (file-exists? (conc stepname ".logpro"))
                 (print "\tlogpro " stepname ".logpro " stepname ".html < " stepname ".log"))
             (print)
             (print stepname " : " stepname ".log")
             (print))
           #:append)

	 (rmt:test-set-top-process-pid run-id test-id pid)
	 (let processloop ((i 0))
	   (let-values (((pid-val exit-status exit-code)(process-wait pid #t)))
		       (mutex-lock! m)
		       (launch:einf-pid-set!         exit-info pid)         ;; (vector-set! exit-info 0 pid)
		       (launch:einf-exit-status-set! exit-info exit-status) ;; (vector-set! exit-info 1 exit-status)
		       (launch:einf-exit-code-set!   exit-info exit-code)   ;; (vector-set! exit-info 2 exit-code)

			     (current-seconds) 
			     start-seconds)))))
	 (kill-tries 0))
    ;; (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area)
    ;; (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))))
      (let ((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    (get-df (current-directory)))
				  (delta (abs (- df disk-free))))
			     (if (> delta 200) ;; ignore changes under 200 Meg
				 df
				 #f))))

	      (set! *runconfigdat* rdat)
	      (set! *toppath*      toppath)
	      (set! *configstatus* 'partial))
	    (begin
	      (debug:print-error 0 *default-log-port* "No " mtconfig " file found. Giving up.")
	      (exit 2))))))
    ;; additional house keeping
    (common:set-transport-type)
    (let* ((linktree (or (getenv "MT_LINKTREE")
			 (if *configdat* (configf:lookup *configdat* "setup" "linktree") #f))))
      (if linktree
	  (begin
	    (if (not (file-exists? linktree))
		(begin
		  (handle-exceptions

	    )))
    (if (and *toppath*
	     (directory-exists? *toppath*))
	(begin
	  (setenv "MT_RUN_AREA_HOME" *toppath*)
	  (setenv "MT_TESTSUITE_NAME" (common:get-testsuite-name)))
	(begin
	  (debug:print-error 0 *default-log-port* "failed to find the top path to your Megatest area.")
          ;;(exit 1)
          #f
          ))
    *toppath*))

(define (get-best-disk confdat testconfig)
  (let* ((disks   (or (and testconfig (hash-table-ref/default testconfig "disks" #f))
		      (hash-table-ref/default confdat "disks" #f)))
	 (minspace (let ((m (configf:lookup confdat "setup" "minspace")))
		     (string->number (or m "10000")))))
    (if disks 
	(let ((res (common:get-disk-with-most-free-space disks minspace))) ;; min size of 1000, seems tad dumb
	  (if res
	      (cdr res)
	      (begin
		(if (common:low-noise-print 20 "No valid disks or no disk with enough space")
		    (debug:print-error 0 *default-log-port* "No valid disks found in megatest.config. Please add some to your [disks] section and ensure the directory exists and has enough space!\n    You can change minspace in the [setup] section of megatest.config. Current setting is: " minspace))
		(exit 1))))))) ;; TODO - move the exit to the calling location and return #f

;; Desired directory structure:
;;
;;  <linkdir> - <target> - <testname> -.
;;                                     |
;;                                     v
;;  <rundir>  -  <target>  -    <testname> -|- <itempath(s)>

;; 2. create run dir on disk, path name is meaningful
;; 3. create link from run dir to megatest runs area 
;; 4. remotely run the test on allocated host
;;    - could be ssh to host from hosts table (update regularly with load)
;;    - could be netbatch
;;      (launch-test db (cadr status) test-conf))
(define (launch-test test-id run-id run-info keyvals runname test-conf test-name test-path itemdat params)
  (mutex-lock! *launch-setup-mutex*) ;; setting variables and processing the testconfig is NOT thread-safe, reuse the launch-setup mutex
  (let* ((item-path       (item-list->path itemdat)))
    (let loop ((delta        (- (current-seconds) *last-launch*))
	       (launch-delay (string->number (or (configf:lookup *configdat* "setup" "launch-delay") "5"))))
      (if (> launch-delay delta)
	  (begin
	    (debug:print-info 0 *default-log-port* "Delaying launch of " test-name " for " (- launch-delay delta) " seconds")
	    (thread-sleep! (- launch-delay delta))
	    (loop (- (current-seconds) *last-launch*) launch-delay))))

    (change-directory *toppath*)
    (alist->env-vars ;; consolidate this code with the code in megatest.scm for "-execute", *maybe* - the longer they are set the longer each launch takes (must be non-overlapping with the vars)
     (append
      (list
       (list "MT_RUN_AREA_HOME" *toppath*)
       (list "MT_TEST_NAME" test-name)
       (list "MT_RUNNAME"   runname)
       (list "MT_ITEMPATH"  item-path)
       )
      itemdat))
    (let* ((tregistry       (tests:get-all)) ;; third param (below) is system-allowed
           ;; for tconfig, why do we allow fallback to test-conf?
	   (tconfig         (or (tests:get-testconfig test-name tregistry #t force-create: #t)
				(begin
                                  (debug:print 0 *default-log-port* "WARNING: falling back to pre-calculated testconfig. This is likely not desired.")
                                  test-conf))) ;; force re-read now that all vars are set
	   (useshell        (let ((ush (config-lookup *configdat* "jobtools"     "useshell")))
			      (if ush 
				  (if (equal? ush "no") ;; must use "no" to NOT use shell
				      #f
				      ush)
				  #t)))     ;; default is yes
	   (runscript       (config-lookup tconfig   "setup"        "runscript"))

	   (cmdparms   #f)
	   (fullcmd    #f) ;; (define a (with-output-to-string (lambda ()(write x))))
	   (mt-bindir-path #f)
	   (testinfo   (rmt:get-test-info-by-id run-id test-id))
	   (mt_target  (string-intersperse (map cadr keyvals) "/"))
	   (debug-param (append (if (args:get-arg "-debug")  (list "-debug" (args:get-arg "-debug")) '())
				(if (args:get-arg "-logging")(list "-logging") '()))))

      ;; (if hosts (set! hosts (string-split hosts)))
      ;; set the megatest to be called on the remote host
      (if (not remote-megatest)(set! remote-megatest local-megatest)) ;; "megatest"))
      (set! mt-bindir-path (pathname-directory remote-megatest))
      (if launcher (set! launcher (string-split launcher)))
      ;; set up the run work area for this test
      (if (and (args:get-arg "-preclean") ;; user has requested to preclean for this run
	       (not (member (db:test-get-rundir testinfo)(list "n/a" "/tmp/badname")))) ;; n/a is a placeholder and thus not a read dir
	  (begin
	    (debug:print-info 0 *default-log-port* "attempting to preclean directory " (db:test-get-rundir testinfo) " for test " test-name "/" item-path)
	    (runs:remove-test-directory testinfo 'remove-data-only))) ;; remove data only, do not perturb the record
      
      ;; prevent overlapping actions - set to LAUNCHED as early as possible
      ;;
      ;; the following call handles waiver propogation. cannot yet condense into roll-up-pass-fail
      (tests:test-set-status! run-id test-id "LAUNCHED" "n/a" #f #f) ;; (if launch-results launch-results "FAILED"))
      (rmt:roll-up-pass-fail-counts run-id test-name item-path #f "LAUNCHED" #f)
      ;; (pp (hash-table->alist tconfig))
      (set! diskpath (get-best-disk *configdat* tconfig))
      (if diskpath
	  (let ((dat  (create-work-area run-id run-info keyvals test-id test-path diskpath test-name itemdat)))
	    (set! work-area (car dat))
	    (set! toptest-work-area (cadr dat))
	    (debug:print-info 2 *default-log-port* "Using work area " work-area))
	  (begin

	(if (not useshell)(debug:print 0 *default-log-port* "WARNING: internal launching will not work well without \"useshell yes\" in your [jobtools] section"))
	(set! fullcmd (append (list remote-megatest "-m" test-sig "-execute" cmdparms) debug-param (list (if useshell "&" ""))))))
      ;; (set! fullcmd (list remote-megatest test-sig "-execute" cmdparms (if useshell "&" "")))))
      (if (args:get-arg "-xterm")(set! fullcmd (append fullcmd (list "-xterm"))))
      (debug:print 1 *default-log-port* "Launching " work-area)
      ;; set pre-launch-env-vars before launching, keep the vars in prevvals and put the envionment back when done
      (debug:print 4 *default-log-port* "fullcmd: " fullcmd)
      (set! *last-launch* (current-seconds)) ;; all that junk above takes time, set this as late as possible.
      (let* ((commonprevvals (alist->env-vars
			      (hash-table-ref/default *configdat* "env-override" '())))
	     (miscprevvals   (alist->env-vars ;; consolidate this code with the code in megatest.scm for "-execute"
			      (append (list (list "MT_TEST_RUN_DIR" work-area)
					    (list "MT_TEST_NAME" test-name)
					    (list "MT_ITEM_INFO" (conc itemdat)) 
					    (list "MT_RUNNAME"   runname)

					  (if launchwait
					      cmdstr
					      (conc cmdstr " >> mt_launch.log 2>&1")))
					(car fullcmd))
				    (if useshell
					'()
					(cdr fullcmd)))))
        (mutex-unlock! *launch-setup-mutex*) ;; yes, really should mutex all the way to here. Need to put this entire process into a fork.
	(if (not launchwait) ;; give the OS a little time to allow the process to start
	    (thread-sleep! 0.01))
	(with-output-to-file "mt_launch.log"
	  (lambda ()
	    (print "LAUNCHCMD: " (string-intersperse fullcmd " "))
	    (if (list? launch-results)
		(apply print launch-results)

Selectors (e.g. use for -runtests, -remove-runs, -set-state-status, -list-runs etc.)
  -target key1/key2/...   : run for key1, key2, etc.
  -reqtarg key1/key2/...  : run for key1, key2, etc. but key1/key2 must be in runconfig
  -testpatt patt1/patt2,patt3/...  : % is wildcard
  -runname                : required, name for this particular test run
  -state                  : Applies to runs, tests or steps depending on context
  -status                 : Applies to runs, tests or steps depending on context
  -mode key               : load testpatt from <key> in runconfigs instead of default TESTPATT
  -tagexpr tag1,tag2%,..  : select tests with tags matching expression

Test helpers (for use inside tests)
  -step stepname
  -test-status            : set the state and status of a test (use :state and :status)
  -setlog logfname        : set the path/filename to the final log relative to the test
                            directory. may be used with -test-status
  -set-toplog logfname    : set the overall log for a suite of sub-tests

			":runname"
			"-runname"
			":state"  
			"-state"
			":status"
			"-status"
			"-list-runs"
			"-testpatt"
                        "-mode"
                        "-tagexpr"
			"-itempatt"
			"-setlog"
			"-set-toplog"
			"-runstep"
			"-logpro"
			"-m"
			"-rerun"

			"-q" ;; quiet 0, errors/warnings only
		       )
		 args:arg-hash
		 0))

;; Add args that use remargs here
;;

(if (and (not (null? remargs))
	 (not (or
	       (args:get-arg "-runstep")
	       (args:get-arg "-envcap")
	       (args:get-arg "-envdelta")
	       )
	      ))

(if (args:get-arg "-server")

    ;; Server? Start up here.
    ;;
    (let ((tl        (launch:setup))
	;; (run-id    (and (args:get-arg "-run-id")
	;; 		  (string->number (args:get-arg "-run-id"))))
          (transport-type *transport-type*  ))


      (server:launch 0 transport-type)
      (set! *didsomething* #t)))
;;     ;; (debug:print-error 0 *default-log-port* "server requires run-id be specified with -run-id")))
;; 
;;     ;; Not a server? This section will decide how to communicate
;;     ;;
;;     ;;  Setup client for all expect listed here

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


      (runs:update-all-test_meta #f)
      (set! *didsomething* #t)))

;;======================================================================
;; Start a repl
;;======================================================================

;; fakeout readline

          (debug:print 0 *default-log-port* "Failed to create HTML output in " toppath "/lt/runs-index.html"))
      (set! *didsomething* #t)))

;;======================================================================
;; Exit and clean up
;;======================================================================



(if (not *didsomething*)
    (debug:print 0 *default-log-port* help))


(thread-join! *watchdog*)
(set! *time-to-exit* #t)

(if (not (eq? *globalexitstatus* 0))
    (if (or (args:get-arg "-run")(args:get-arg "-runtests")(args:get-arg "-runall"))
        (begin
           (debug:print 0 *default-log-port* "NOTE: Subprocesses with non-zero exit code detected: " *globalexitstatus*)
           (exit 0))
        (case *globalexitstatus*
         ((0)(exit 0))
         ((1)(exit 1))
         ((2)(exit 2))
         (else (exit 3)))))

;;======================================================================
;; Copyright 2006-2013, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.
;;======================================================================
;;
(use format typed-records) ;; RADT => purpose of json format??

(declare (unit rmt))
(declare (uses api))
(declare (uses tdb))
(declare (uses http-transport))
(declare (uses rpc-transport))
;;(declare (uses nmsg-transport))
(include "common_records.scm")

;;
;; THESE ARE ALL CALLED ON THE CLIENT SIDE!!!
;;

;; generate entries for ~/.megatestrc with the following
;;
;;  grep define ../rmt.scm | grep rmt: |perl -pi -e 's/\(define\s+\((\S+)\W.*$/\1/'|sort -u









;;======================================================================
;;  S U P P O R T   F U N C T I O N S
;;======================================================================

;; if a server is either running or in the process of starting call client:setup
;; else return #f to let the calling proc know that there is no server available
;;

      (rmt:send-receive cmd rid params attemptnum: attemptnum))
     ;; on homehost and this is a read
     ((and (cdr (remote-hh-dat *runremote*))   ;; on homehost
           (member cmd api:read-only-queries)) ;; this is a read
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  3")
      (rmt:open-qry-close-locally cmd 0 params))

     ;; on homehost and this is a write, we already have a server, but server has died
     ((and (cdr (remote-hh-dat *runremote*))         ;; on homehost
           (not (member cmd api:read-only-queries))  ;; this is a write
           (remote-server-url *runremote*)           ;; have a server
           (not (server:read-dotserver *toppath*)))  ;; server has died.
      (set! *runremote* #f)
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  4.1")
      (rmt:send-receive cmd rid params attemptnum: attemptnum))

     ;; on homehost and this is a write, we already have a server
     ((and (cdr (remote-hh-dat *runremote*))         ;; on homehost
           (not (member cmd api:read-only-queries))  ;; this is a write
           (remote-server-url *runremote*))          ;; have a server
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  4")
      (rmt:open-qry-close-locally cmd 0 params))

     ;; commented by bb; this was blocking server passive start on write on homehost (case 5)
     ;; ;; on homehost and this is a write, we have a server (we know because case 4 checked)
     ;; ((and (cdr (remote-hh-dat *runremote*))         ;; on homehost
     ;;       (not (member cmd api:read-only-queries)))
     ;;  (mutex-unlock! *rmt-mutex*)
     ;;  (debug:print-info 12 *default-log-port* "rmt:send-receive, case  4.1")
     ;;  (rmt:open-qry-close-locally cmd 0 params))

     
     ;; no server contact made and this is a write, passively start a server 
     ((and (not (remote-server-url *runremote*))
	   (not (member cmd api:read-only-queries)))
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  5")
      (let ((serverconn (server:read-dotserver *toppath*))) ;; (server:check-if-running *toppath*))) ;; Do NOT want to run server:check-if-running - very expensive to do for every write call
	(if serverconn
	    (remote-server-url-set! *runremote* serverconn) ;; the string can be consumed by the client setup if needed

     ;; if not on homehost ensure we have a connection to a live server
     ;; NOTE: we *have* a homehost record by now
     ((and (not (cdr (remote-hh-dat *runremote*)))        ;; are we on a homehost?
           (not (remote-conndat *runremote*)))            ;; and no connection
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  6  hh-dat: " (remote-hh-dat *runremote*) " conndat: " (remote-conndat *runremote*))
      (mutex-unlock! *rmt-mutex*)
      (tasks:start-and-wait-for-server (tasks:open-db) 0 15)
      (let* ((cinfo (rmt:get-connection-info 0))
             (transport (if cinfo
                            (vector-ref cinfo 6)
                            (server:get-transport)))) ;; TODO: replace with tasks:server-dat-accessor-?? for transport
        (remote-conndat-set! *runremote* cinfo) ;; calls client:setup which calls client:setup-http
        (remote-transport-set! *runremote* transport))
      (rmt:send-receive cmd rid params attemptnum: attemptnum))
     ;; all set up if get this far, dispatch the query
     ((cdr (remote-hh-dat *runremote*)) ;; we are on homehost
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  7")
      (rmt:open-qry-close-locally cmd (if rid rid 0) params))
     ;; not on homehost, do server query
     (else
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  9")
      (let* ((conninfo (remote-conndat *runremote*))
	     (dat      (case (remote-transport *runremote*)
			 ((http) (condition-case ;; handling here has caused a lot of problems. However it is needed to deal with attemtped communication to servers that have gone away
                                  (http-transport:client-api-send-receive 0 conninfo cmd params)
                                  ((commfail)(vector #f "communications fail"))
                                  ((exn)(vector #f "other fail" (print-call-chain)))))
                         ((rpc) (condition-case ;; handling here has caused a lot of problems. However it is needed to deal with attemtped communication to servers that have gone away
                                  (rpc-transport:client-api-send-receive 0 conninfo cmd params)
                                  ((commfail)(vector #f "communications fail"))
                                  ((exn)(vector #f "other fail" (print-call-chain)))))
			 (else
			  (debug:print 0 *default-log-port* "ERROR: transport " (remote-transport *runremote*) " not supported (1)")
			  (exit))))
	     (success  (if (vector? dat) (vector-ref dat 0) #f))
	     (res      (if (vector? dat) (vector-ref dat 1) #f)))
	(if (vector? conninfo)(http-transport:server-dat-update-last-access conninfo)) ;; refresh access time
        (debug:print-info 12 *default-log-port* "rmt:send-receive, case  9. conninfo=" conninfo " dat=" dat)
	(if success
	    (case (remote-transport *runremote*)
	      ((http rpc) res)
	      (else
	       (debug:print 0 *default-log-port* "ERROR: transport " (remote-transport *runremote*) " is unknown")
	       (exit 1)))
	    (begin
	      (debug:print 0 *default-log-port* "WARNING: communication failed. Trying again, try num: " attemptnum)
	      (remote-conndat-set!    *runremote* #f)
	      (remote-server-url-set! *runremote* #f)

	      #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)
                (mutex-unlock! *db-multi-sync-mutex*)))))
    res))

(define (rmt:send-receive-no-auto-client-setup connection-info cmd run-id params)
  (let* ((run-id   (if run-id run-id 0))
         (transport (or (remote-transport *runremote*) (server:get-transport)))
	 (res  	   (handle-exceptions
		    exn
		    #f
                    (case transport
                      ((http) (http-transport:client-api-send-receive run-id connection-info cmd params))
                      ((rpc) (rpc-transport:client-api-send-receive run-id connection-info cmd params))
                      (else
			  (debug:print 0 *default-log-port* "ERROR: transport " (remote-transport *runremote*) " not supported (2)")
			  (exit))

                    ))))
    (if (and res (vector-ref res 0))
	(vector-ref res 1) ;;; YES!! THIS IS CORRECT!! CHANGE IT HERE, THEN CHANGE rmt:send-receive ALSO!!!
	#f)))

;; ;; Wrap json library for strings (why the ports crap in the first place?)
;; (define (rmt:dat->json-str dat)
;;   (with-output-to-string 

  (rmt:send-receive 'login run-id (list *toppath* megatest-version *my-client-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 connection-info)
  (case *transport-type* ;; run-id of 0 is just a placeholder
    ((http rpc)(rmt:send-receive-no-auto-client-setup connection-info 'login 0 (list *toppath* megatest-version *my-client-signature*)))
    (else
     (debug:print 0 *default-log-port* "ERROR: transport " (remote-transport *runremote*) " not supported (3)")
     (exit))

    
    ;;((nmsg)(nmsg-transport:client-api-send-receive run-id connection-info 'login (list *toppath* megatest-version run-id *my-client-signature*)))
    ))

;; hand off a call to one of the db:queries statements
;; added run-id to make looking up the correct db possible 
;;
(define (rmt:general-call stmtname run-id . params)
  (rmt:send-receive 'general-call run-id (append (list stmtname run-id) params)))


;; 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)))

;; (define (rmt:sync-inmem->db run-id)
;;   (rmt:send-receive 'sync-inmem->db run-id '()))

(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)
  (rmt:send-receive 'runtests run-id testpatt))

;;======================================================================
;;  T E S T   M E T A 
;;======================================================================

(define (rmt:get-tests-tags)
  (rmt:send-receive 'get-tests-tags #f '()))

;;======================================================================
;;  K E Y S 
;;======================================================================

;; These require run-id because the values come from the run!
;;
(define (rmt:get-key-val-pairs run-id)
  (rmt:send-receive 'get-key-val-pairs run-id (list run-id)))

(define (rmt:get-keys)
  (if *db-keys* *db-keys* 
     (let ((res (rmt:send-receive 'get-keys #f '())))
       (set! *db-keys* res)
       res)))

(define (rmt:get-keys-write) ;; dummy query to force server start
  (let ((res (rmt:send-receive 'get-keys-write #f '())))
    (set! *db-keys* res)
    res))

;; we don't reuse run-id's (except possibly *after* a db cleanup) so it is safe
;; to cache the resuls in a hash
;;
(define (rmt:get-key-vals run-id)
  (or (hash-table-ref/default *keyvals* run-id #f)
      (let ((res (rmt:send-receive 'get-key-vals #f (list run-id))))

;; Copyright 2006-2016, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

(declare (uses common))
(declare (uses db))
(declare (uses tests))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.

(include "common_records.scm")
(include "db_records.scm")

(define *heartbeat-mutex* (make-mutex))
(define *server-loop-heart-beat* (current-seconds))


;; procstr is the name of the procedure to be called as a string
(define (rpc-transport:autoremote procstr params)  ;; may be unused, I think api-exec deprecates this one.
  (let* ((procsym (if (symbol? procstr)
                     procstr
                     (string->symbol (->string procstr))))
        (res
         (begin
           (apply (eval procsym) params))))
    res))


;; rpc receiver
(define (rpc-transport:api-exec cmd params)
  (let* ( (resdat  (api:execute-requests *dbstruct-db* (vector cmd params))) ;; #( flag result )
          (flag    (vector-ref resdat 0))
          (res     (vector-ref resdat 1)))

    (mutex-lock! *heartbeat-mutex*)

    (set! *last-db-access* (current-seconds)) ;; bump *last-db-access*; this will renew keep-running thread's lease on life for another (server:get-timeout) seconds
    ;;(BB> "in api-exec; last-db-access updated to "*last-db-access*)
    (mutex-unlock! *heartbeat-mutex*)

    res))


;; retry an operation (depends on srfi-18)
;; ==================
;; idea here is to avoid spending time on coding retrying something.  Trying to be generic here.
;;
;; Exception handling:
;; -------------------
;; if evaluating the thunk results in exception, it will be retried.
;; on last try, if final-failure-returns-actual is true, the exception will be re-thrown to caller.
;;
;; look at options below #!key to see how to configure behavior
;;
;;
(define (retry-thunk
         the-thunk
         #!key ;;;; options below
         (accept-result?   (lambda (x) x)) ;; retry if predicate applied to thunk's result is false 
         (retries                       4) ;; how many tries
         (failure-value                #f) ;; return this on final failure, unless following option is enabled:
         (final-failure-returns-actual #f) ;; on failure, on the last try, just return the result, not failure-value

         (retry-delay                 0.1) ;; delay between tries
         (back-off-factor               1) ;; multiply retry-delay by this factor on retry
         (random-delay                0.1) ;; add a random portion of this value to wait

         (chatty                       #f) ;; print status as we go, for debugging.
         )
  
  (when chatty (print) (print "Entered retry-thunk") (print "-=-=-=-=-=-"))
  (let* ((guarded-thunk ;; we are guarding the thunk against exceptions.  We will record whether result of evaluation is an exception or a regular result.
          (lambda ()
           (let* ((EXCEPTION (gensym)) ;; using gensym to avoid potential collision
                  (res
                   (condition-case
                    (the-thunk) ;; this is what we are guarding the execution of
                    [x () (cons EXCEPTION x)]
                    )))
             (cond
              ((and (pair? res) (eq? (car res) EXCEPTION))
               (if chatty
                   (print " - the-thunk threw exception >"(cdr res)"<"))
               (cons 'exception (cdr res)))
               (else
                (if chatty
                    (print " - the-thunk returned result >"res"<"))
                (cons 'regular-result res)))))))
    
    (let loop ((guarded-res (guarded-thunk))
               (retries-left retries)
               (fail-wait retry-delay))
      (if chatty (print "   =========="))
      (let* ((wait-time (+ fail-wait (+ (* fail-wait back-off-factor)
                                        (* random-delay
                                           (/ (random 1024) 1024) ))))
             (res-type (car guarded-res))
             (res-value (cdr guarded-res)))
        (cond
         ((and (eq? res-type 'regular-result) (accept-result? res-value))
                   (if chatty (print " + return result that satisfied accept-result? >"res-value"<"))
                   res-value)

         ((> retries-left 0)
          (if chatty (print " - sleep "wait-time))
          (thread-sleep! wait-time)
          (if chatty (print " + retry ["retries-left" tries left]"))
          (loop (guarded-thunk)
                (sub1 retries-left)
                wait-time))
         
         ((eq? res-type 'regular-result)
          (if final-failure-returns-actual
              (begin
                (if chatty (print " + last try failed- return the result >"res-value"<"))
                res-value)
              (begin
                (if chatty (print " + last try failed- return canned failure value >"failure-value"<"))
              failure-value)))
         
         (else ;; no retries left; result was not accepted and res-type can only be 'exception
          (if final-failure-returns-actual 
              (begin
                (if chatty (print " + last try failed with exception- re-throw it >"res-value"<"))
                (abort res-value)); re-raise the exception. TODO: find a way for call-history to show as though from entry to this function
              (begin
                (if chatty (print " + last try failed with exception- return canned failure value >"failure-value"<"))
                failure-value))))))))


(define (rpc-transport:server-shutdown server-id rpc:listener ) ;;#!key (from-on-exit #f))
  ;;(on-exit (lambda () #t)) ;; turn off on-exit stuff
  ;;(tcp-close rpc:listener) ;; gotta exit nicely
  ;;(tasks:server-set-state! (db:delay-if-busy (tasks:open-db)) server-id "stopped")


  ;; TODO: (low) the following is extraordinaritly slow.  Maybe we don't even need portlogger for rpc anyway??  the exception-based failover when ports are taken is fast!
  ;;(portlogger:open-run-close portlogger:set-port (rpc:default-server-port) "released")
  
  (set! *time-to-exit* #t)
  ;;(if *dbstruct-db* (db:sync-touched *dbstruct-db* *run-id* force-sync: #t))
  
  (server:remove-dotserver-file *toppath* "anyhost:anyport" force: #t)
  (tasks:server-delete-record (db:delay-if-busy (tasks:open-db)) server-id " rpc-transport:keep-running complete")
  
  (rpc:close-all-connections!)
  ;;(BB> "Before (exit) (from-on-exit="from-on-exit")")
  ;;(unless from-on-exit (exit))  ;; sometimes we hang (around) here with 100% cpu.
  ;;(BB> "After")
  ;; strace reveals endless:
  ;; getrusage(RUSAGE_SELF, {ru_utime={413, 917868}, ru_stime={0, 60003}, ...}) = 0
  ;; getrusage(RUSAGE_SELF, {ru_utime={414, 9874}, ru_stime={0, 60003}, ...}) = 0
  ;; getrusage(RUSAGE_SELF, {ru_utime={414, 13874}, ru_stime={0, 60003}, ...}) = 0
  ;; getrusage(RUSAGE_SELF, {ru_utime={414, 105880}, ru_stime={0, 60003}, ...}) = 0
  ;; getrusage(RUSAGE_SELF, {ru_utime={414, 109880}, ru_stime={0, 60003}, ...}) = 0
  ;; getrusage(RUSAGE_SELF, {ru_utime={414, 201886}, ru_stime={0, 60003}, ...}) = 0
  ;; getrusage(RUSAGE_SELF, {ru_utime={414, 205886}, ru_stime={0, 60003}, ...}) = 0
  ;; getrusage(RUSAGE_SELF, {ru_utime={414, 297892}, ru_stime={0, 60003}, ...}) = 0
  ;; getrusage(RUSAGE_SELF, {ru_utime={414, 301892}, ru_stime={0, 60003}, ...}) = 0
  ;; getrusage(RUSAGE_SELF, {ru_utime={414, 393898}, ru_stime={0, 60003}, ...}) = 0
  ;; getrusage(RUSAGE_SELF, {ru_utime={414, 397898}, ru_stime={0, 60003}, ...}) = 0
  ;; make a post to chicken-users w/ http://paste.call-cc.org/paste?id=60a4b66a29ccf7d11359ea866db642c970735978


  ;; (if from-on-exit
  ;;     ;; avoid above condition!  End current process externally  since 1 in 20 (exit)'s result in hung, 100% cpu zombies. (see above)
  
  (system (conc  "kill -9  "(current-process-id)))
  )


;; all routes though here end in exit ...
;;
;; start_server? 
;;
(define (rpc-transport:launch run-id)


  (set! *run-id*   run-id)

  ;; ;; send to background if requested
  ;; (when (args:get-arg "-daemonize")
  ;;     (daemon:ize)
  ;;     (when *alt-log-file* ;; we should re-connect to this port, I think daemon:ize disrupts it
  ;;       (current-error-port *alt-log-file*)
  ;;       (current-output-port *alt-log-file*)))

  ;; double check we dont alrady have a running server for this run-id
  (when (and (server:read-dotserver *toppath*)
             (server:check-if-running run-id))

    (debug:print 0 *default-log-port* "INFO: Server for run-id " run-id " already running")
    (exit 0))

  ;; did not find server running, let's clean up the table of dead servers
  (tasks:server-force-clean-running-records-for-run-id  (db:delay-if-busy (tasks:open-db)) run-id "notresponding")  

  (server:dotserver-starting)

  


  
  ;; let's get a server-id for this server
  ;;   if at first we do not suceed, try 3 more times.
  (let ((server-id (retry-thunk
                    (lambda () (tasks:server-lock-slot (db:delay-if-busy (tasks:open-db)) run-id 'rpc))
                    chatty: #f
                    final-failure-returns-actual: #t
                    retries: 4)))
    (when (not server-id) ;; dang we couldn't get a server-id.
      ;; since we didn't get the server lock we are going to clean up and bail out


      (debug:print-info 2 *default-log-port* "INFO: server pid=" (current-process-id) ", hostname=" (get-host-name) " not starting due to other candidates ahead in start queue")
      (tasks:server-delete-records-for-this-pid (db:delay-if-busy (tasks:open-db)) " rpc-transport:launch")
      (server:dotserver-starting-remove)
      (exit 1))

    ;; we got a server-id (and a corresponding entry in servers table in globally shared mdb)
    ;; all systems go.  Proceed to setup rpc server.  
    (rpc-transport:run
     (if (args:get-arg "-server")
         (args:get-arg "-server")
         "-")
     run-id
     server-id)
    (exit)))

(define *rpc-listener-port* #f)
(define *rpc-listener-port-bind-timestamp* #f)

(define *on-exit-flag #f)

(define (rpc-transport:server-dat-get-iface         vec)    (vector-ref  vec 0))
(define (rpc-transport:server-dat-get-port          vec)    (vector-ref  vec 1))
(define (rpc-transport:server-dat-get-last-access   vec)    (vector-ref  vec 5))
(define (rpc-transport:server-dat-get-transport     vec)    (vector-ref  vec 6))
(define (rpc-transport:server-dat-update-last-access vec)
  (if (vector? vec)
      (vector-set! vec 5 (current-seconds))
      (begin
	(print-call-chain (current-error-port))
	(debug:print-error 0 *default-log-port* "call to rpc-transport:server-dat-update-last-access with non-vector!!"))))


(define *api-exec-ht* (make-hash-table))
(define *api-exec-mutex* (make-mutex))
;; let's see if caching the rpc stub curbs thread-profusion on server side
(define (rpc-transport:get-api-exec iface port)
  (mutex-lock! *api-exec-mutex*)
  (let* ((lu (hash-table-ref/default *api-exec-ht* (cons iface  port) #f)))
    (if lu
        (begin
          (mutex-unlock! *api-exec-mutex*)
          lu)
        (let ((res (rpc:procedure 'api-exec iface port)))
          (hash-table-set! *api-exec-ht* (cons iface port) res)
          (mutex-unlock! *api-exec-mutex*)
          res))))

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; this client-side procedure makes rpc call to server and returns result
;;
(define (rpc-transport:client-api-send-receive run-id serverdat cmd params #!key (numretries 3))
  ;;(BB> "entered rpc-transport:client-api-send-receive with run-id="run-id " serverdat="serverdat" cmd="cmd" params="params" numretries="numretries)
  (if (not (vector? serverdat))
      (begin
        (BB> "WHAT?? for run-id="run-id", serverdat="serverdat)
        (print-call-chain)
        (rpc:close-all-connections!)
        (exit 1)))
  (let* ((iface (rpc-transport:server-dat-get-iface serverdat))
         (port  (rpc-transport:server-dat-get-port serverdat))
         (res #f)
         (api-exec (rpc-transport:get-api-exec iface port))  ;; chached by host/port. may need to clear...
         (send-receive (lambda ()
                         (tcp-buffer-size 0)
                         (set! res (retry-thunk
                                    (lambda ()
                                      (condition-case
                                       ;;(vector #t (run-remote cmd params))
                                       (vector 'success (api-exec cmd params))
                                       [x (exn i/o net) (vector 'comms-fail (conc "communications fail ["(->string x)"]") x)]
                                       [x () (vector 'other-fail "other fail ["(->string x)"]" x)]))
                                    chatty: #f
                                    accept-result?: (lambda(x)
                                                      (and (vector? x) (vector-ref x 0)))
                                    retries: 8
                                    back-off-factor: 1.5
                                    random-wait: 0.2
                                    retry-delay: 0.1
                                    final-failure-returns-actual: #t))
                         ;;(BB> "HEY res="res)
                         res
                         ))
         (th1 (make-thread send-receive "send-receive"))
         (time-out-reached #f)
         (time-out     (lambda ()
			      (thread-sleep! 45)
                              (set! time-out-reached #t)
                              (thread-terminate! th1)
                              
			      #f))

         (th2 (make-thread time-out     "time out")))
	 (thread-start! th1)
	 (thread-start! th2)
	 (thread-join! th1)
	 (thread-terminate! th2)
         ;;(BB> "alt got res="res)
	 (debug:print-info 11 *default-log-port* "got res=" res)
	 (if (vector? res)
             (case (vector-ref res 0)
               ((success) (vector #t (vector-ref res 1)))
               (
                (comms-fail other-fail)
                ;;(comms-fail) 
                (debug:print 0 *default-log-port* "WARNING: comms failure for rpc request >>"res"<<")

                ;;(debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
                (vector #f (vector-ref res 1)))
               (else
                (debug:print-error 0 *default-log-port* "error occured at server, info=" (vector-ref res 1))
                (debug:print 0 *default-log-port* " client call chain:")
                (print-call-chain (current-error-port))
                (debug:print 0 *default-log-port* " server call chain:")
                (pp (vector-ref res 1) (current-error-port))
                (signal (vector-ref res 2))))
             (signal (make-composite-condition
                      (make-property-condition 
             	       'timeout
             	       'message "nmsg-transport:client-api-send-receive-raw timed out talking to server"))))))
        


(define (rpc-transport:run hostn run-id server-id)
  (debug:print 2 *default-log-port* "Attempting to start the rpc server ...")
   ;; (trace rpc:publish-procedure!)

  ;;======================================================================
  ;;	  start of publish-procedure section
  ;;======================================================================
  (rpc:publish-procedure! 'server:login server:login) ;; this allows client to validate it is the same megatest instance as the server.  No security here, just making sure we're in the right room.
  (rpc:publish-procedure!
   'testing
   (lambda ()
     "Just testing"))

  ;; procedure to receive arbitrary API request from client's rpc:send-receive/rpc-transport:client-api-send-receive 
  (rpc:publish-procedure! 'rpc-transport:autoremote rpc-transport:autoremote)
  ;; can use this to run most anything at the remote
  (rpc:publish-procedure! 'api-exec rpc-transport:api-exec)
  
  
  ;;======================================================================
  ;;	  end of publish-procedure section
  ;;======================================================================


  (let* ((db              #f)
	 (hostname        (let ((res (get-host-name)))  res))
         (server-start-time (current-seconds))
         (server-timeout (server:get-timeout))
	 (ipaddrstr       (let* ((ipstr (if (string=? "-" hostn)
					   ;; (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
					   (server:get-best-guess-address hostname)
					   #f))
                                 (res (if ipstr ipstr hostn)))
                            res)) ;; hostname))) 
	 (start-port      (let ((res (portlogger:open-run-close portlogger:find-port)))      ;; BB> TODO: remove portlogger!
                            res))
	 (link-tree-path  (configf:lookup *configdat* "setup" "linktree"))

         ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
         ;; rpc:listener is the tcp-listen result from inside the find-free-port-and-open complex.
         ;;   It is our handle on the listening tcp port
         ;;   We will attach this to our rpc server with rpc:make-server in thread th1 .
	 (rpc:listener    (rpc-transport:find-free-port-and-open start-port)) 
	 (th1             (make-thread
			   (lambda ()
                             ;;(BB> "BEFORE rpc:make-server")
			     ((rpc:make-server rpc:listener) #t)
                             ;;(BB> "BEFORE rpc:make-server")
                             )
			   "rpc:server"))


         (hostname        (if (string=? "-" hostn)
			      (get-host-name) 
			      hostn))
	 (ipaddrstr       (if (string=? "-" hostn)
			      (server:get-best-guess-address hostname) ;; (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
                              (string-intersperse 
                               (map number->string
                                    (u8vector->list
                                      (hostname->ip hostn))) ".")
                              ))
	 (portnum         (let ((res (rpc:default-server-port)))  res))
	 (host:port       (conc (if ipaddrstr ipaddrstr hostname) ":" portnum)))

    (when (not (equal? ipaddrstr (server:get-best-guess-address (get-host-name))))
        
      (debug:print 0 *default-log-port* "Error: This host "(ip->string (hostname->ip (get-host-name)))" ("(get-host-name)") is not the homehost "ipaddrstr" ("(ip->hostname (string->ip ipaddrstr))"; Cannot proceed.")
      (server:dotserver-starting-remove)
      (tcp-close rpc:listener) ;; gotta exit nicely and free up that tcp port
      (exit))
    
    (tasks:server-set-interface-port (db:delay-if-busy (tasks:open-db)) server-id ipaddrstr portnum)

    ;;============================================================
    ;;  activate thread th1 to attach opened tcp port to rpc server
    ;;=============================================================
    (thread-start! th1)
    (set! db *dbstruct-db*)




    (debug:print 0 *default-log-port* "Server started on " host:port)
    ;;(BB> "before SELF-TEST")
    (if (retry-thunk (lambda ()
                       (rpc-transport:self-test run-id ipaddrstr portnum))
                     final-failure-returns-actual: #t ;; TODO: remove this line
                     )
        (debug:print 0 *default-log-port* "INFO: rpc self test passed!")
        (begin
          (debug:print 0 *default-log-port* "Error: rpc listener did not pass self test.  Shutting down.  On: " host:port)
          (tasks:server-set-state! (db:delay-if-busy (tasks:open-db)) server-id "dead")
          (tcp-close rpc:listener) ;; gotta exit nicely and free up that tcp port
          (rpc-transport:server-shutdown server-id rpc:listener)
          (server:dotserver-starting-remove)
          (exit)))




    ;;(on-exit (lambda ()
    ;;           (rpc-transport:server-shutdown server-id rpc:listener from-on-exit: #t)))
    
    ;; check again for running servers for this run-id in case one has snuck in since we checked last in rpc-transport:launch
    (if (not (equal? server-id (tasks:server-am-i-the-server? (db:delay-if-busy (tasks:open-db)) run-id)));; try to ensure no double registering of servers
        (begin ;; i am not the server, another server snuck in and beat this one to the punch
          (tcp-close rpc:listener) ;; gotta exit nicely and free up that tcp port
          (tasks:server-set-state! (db:delay-if-busy (tasks:open-db)) server-id "collision")
          (server:dotserver-starting-remove))

        (begin ;; i am the server
          ;; setup the in-memory db
          (set! *dbstruct-db*  (db:setup run-id))
          (db:get-db *dbstruct-db* run-id)

          ;; at this point, satisfied server has started
          ;; let's make it official
          (server:write-dotserver *toppath* (conc ipaddrstr ":" portnum))
          (mutex-lock! *heartbeat-mutex*)
          (set! *last-db-access* (current-seconds))
          (mutex-unlock! *heartbeat-mutex*)
          (set! *rpc:listener* rpc:listener) 
          (tasks:server-set-state! (db:delay-if-busy (tasks:open-db)) server-id "running") ;; update our mdb servers entry

          
          
          ;; this let loop will hold open this thread until we want the server to shut down.
          ;;   if no requests received within the last 20 seconds :
          ;;   database hasnt changed in ??
          ;;


          
          ;; keep-running loop: polls last-db-access to see if we have timed out.  keep running if not. 
          (let loop ((count          0)
                     (bad-sync-count 0))
            (BB> "keep running: count = "count)
            ;; Use this opportunity to sync the inmemdb to db

            (let ((start-time (current-milliseconds))
                  (sync-time  #f)
                  (rem-time   #f))

              ;; following is now done in common:watchdog, commenting out.  sync-time will now be 0; can live with that.
              ;; ;; inmemddb is a dbstruct
              ;; (condition-case
              ;;  (db:sync-touched *dbstruct-db* *run-id* force-sync: #t)
              ;;  ((sync-failed)(cond
              ;;                 ((> bad-sync-count 10) ;; time to give up
              ;;                  (rpc-transport:server-shutdown server-id rpc:listener))
              ;;                 (else ;; (> bad-sync-count 0)  ;; we've had a fail or two, delay and loop
              ;;                  (thread-sleep! 5)
              ;;                  (loop count (+ bad-sync-count 1)))))
              ;;  ((exn)
              ;;   (debug:print-error 0 *default-log-port* "error from sync code other than 'sync-failed. Attempting to gracefully shutdown the server ")
              ;;   (rpc-transport:server-shutdown server-id rpc:listener)))
              
              (set! sync-time  (- (current-milliseconds) start-time))

              (set! rem-time (quotient (- 4000 sync-time) 1000))
              (debug:print 4 *default-log-port* "SYNC: time= " sync-time ", rem-time=" rem-time)
              

              (if (and (<= rem-time 4)
                       (> rem-time 0))
                  (thread-sleep! rem-time)
                  (thread-sleep! 4))) ;; fallback for if the math is changed ...
            
            (if (< count 1) ;; 3x3 = 9 secs aprox
                (loop (+ count 1) bad-sync-count))

            ;; BB: don't see how this is possible with RPC
            ;; ;; Check that iface and port have not changed (can happen if server port collides)
            ;; (mutex-lock! *heartbeat-mutex*)
            ;; (set! sdat *server-info*)
            ;; (mutex-unlock! *heartbeat-mutex*)
            
            ;; (if (or (not (equal? sdat (list iface port)))
            ;;         (not server-id))
            ;;     (begin 
            ;;       (debug:print-info 0 *default-log-port* "interface changed, refreshing iface and port info")
            ;;       (set! iface (car sdat))
            ;;       (set! port  (cadr sdat))))
            
            ;; Transfer *last-db-access* to last-access to use in checking that we are still alive
            (mutex-lock! *heartbeat-mutex*)
            (set! last-access *last-db-access*)
            (mutex-unlock! *heartbeat-mutex*)
            
            ;; (debug:print 11 *default-log-port* "last-access=" last-access ", server-timeout=" server-timeout)
            ;;
            ;; no_traffic, no running tests, if server 0, no running servers
            ;;
            ;; (let ((wait-on-running (configf:lookup *configdat* "server" b"wait-on-running"))) ;; wait on running tasks (if not true then exit on time out)
            ;;
            (let* ((hrs-since-start  (/ (- (current-seconds) server-start-time) 3600))
                   (adjusted-timeout (if (> hrs-since-start 1)
                                         (- server-timeout (inexact->exact (round (* hrs-since-start 60))))  ;; subtract 60 seconds per hour
                                         server-timeout)))
              (if (common:low-noise-print 120 "server timeout")
                  (debug:print-info 0 *default-log-port* "Adjusted server timeout: " adjusted-timeout))
              (if (and *server-run*
                       (> (+ last-access server-timeout)
                          (current-seconds)))
                  (begin
                    (if (common:low-noise-print 120 "server continuing")
                        (debug:print-info 0 *default-log-port* "Server continuing, seconds since last db access: " (- (current-seconds) last-access)))
                    ;;
                    ;; Consider implementing some smarts here to re-insert the record or kill self is
                    ;; the db indicates so
                    ;;
                    (if (tasks:server-am-i-the-server? (db:delay-if-busy (tasks:open-db)) run-id)
                        (tasks:server-set-state! (db:delay-if-busy (tasks:open-db)) server-id "running"))
                    ;;
                    (loop 0 bad-sync-count))
                  (begin
                    ;;(BB> "SERVER SHUTDOWN CALLED!  last-access="last-access" current-seconds="(current-seconds)" server-timeout="server-timeout)
                    
                    (rpc-transport:server-shutdown server-id rpc:listener)))))
          ;; end new loop
          ))))


(define (rpc-transport:find-free-port-and-open port #!key )
  (handle-exceptions
   exn
   (begin
     (print "Failed to bind to port " (rpc:default-server-port) ", trying next port")
     (rpc-transport:find-free-port-and-open (add1 port)))
   (rpc:default-server-port port)
   (set! *rpc-listener-port* port) ;; a bit paranoid about rpc:default-server-port parameter not changing across threads (as params are wont to do).  keeping this global in my back pocket in case this causes problems
   (set! *rpc-listener-port-bind-timestamp* (current-milliseconds)) ;; may want to test how long it has been since the last bind attempt happened...
   (tcp-read-timeout 240000)
   (tcp-buffer-size 0) ;; gotta do this because http-transport undoes it.
   (tcp-listen (rpc:default-server-port) 10000)
   ))
  
(define (rpc-transport:ping run-id host port)
  (handle-exceptions
   exn
   (begin
     (print "SERVER_NOT_FOUND exn="exn)
     (exit 1))
   (let ((login-res ((rpc:procedure 'server:login host port) *toppath*)))

     (if login-res
	 (begin
	   (print "LOGIN_OK")
	   (exit 0))
	 (begin
	   (print "LOGIN_FAILED")
	   (exit 1))))))

(define (rpc-transport:self-test run-id host port)
  (if (not host)
      (abort "host not set."))



  (if (not port)

      (abort "port not set."))
  (tcp-buffer-size 0) ;; gotta do this because http-transport undoes it.
  (let* ((testing-res ((rpc:procedure 'testing host port)))
         (login-res ((rpc:procedure 'server:login host port) *toppath*))
         (res (and login-res (equal? testing-res "Just testing"))))
    


    (if login-res











        (begin

          ;;(BB> "Self test PASS.  login-res="login-res" testing-res="testing-res" *toppath*="*toppath*)
          #t)
        (begin
          (BB> "Self test fail.  login-res="login-res" testing-res="testing-res" *toppath*="*toppath*)
           
          #f))
    res))

(define (rpc-transport:client-setup run-id server-dat #!key (remaining-tries 10))
  ;;(BB> "entered rpc-transport:client-setup with run-id="run-id" and server-dat="server-dat" and retries="remaining-tries)

  (tcp-buffer-size 0)
  (debug:print-info 0 *default-log-port* "rpc-transport:client-setup run-id="run-id" server-dat=" server-dat ", remaining-tries=" remaining-tries)







  (let* ((iface     (tasks:hostinfo-get-interface server-dat))
         (hostname  (tasks:hostinfo-get-hostname  server-dat))
         (port      (tasks:hostinfo-get-port      server-dat))



         (runremote-server-dat (vector iface port #f #f #f (current-seconds) 'rpc)) ;; http version := (vector iface port api-uri api-url api-req (current-seconds) 'http  )
         (ping-res (retry-thunk (lambda ()  ;; make 3 attempts to ping.




                                  ((rpc:procedure 'server:login iface port) *toppath*))
                                chatty: #f
                                retries: 3)))
    ;; we got here from rmt:get-connection-info on the condition that *runremote* has no entry for run-id...
    (if ping-res
        (begin
          (debug:print-info 0 *default-log-port* "rpc-transport:client-setup CONNECTION ESTABLISHED run-id="run-id" server-dat=" server-dat)
          runremote-server-dat)
        (begin ;; login failed but have a server record, clean out the record and try again
          (debug:print-info 0 *default-log-port* "rpc-transport:client-setup UNABLE TO CONNECT run-id="run-id" server-dat=" server-dat)
          (tasks:kill-server-run-id run-id)
          (tasks:server-force-clean-run-record  (db:delay-if-busy (tasks:open-db)) run-id iface port
                                                   " rpc-transport:client-setup (server-dat = #t)")
          (if (> remaining-tries 2)
              (thread-sleep! (+ 1 (random 5))) ;; spread out the starts a little
              (thread-sleep! (+ 15 (random 20)))) ;; it isn't going well. give it plenty of time
          (server:try-running run-id)
          (thread-sleep! 5)   ;; give server a little time to start up
          (client:setup run-id remaining-tries: (sub1 remaining-tries))))))

	      (val (config-lookup test-conf "test_meta" fld)))
	 ;; (debug:print 5 *default-log-port* "idx: " idx " fld: " fld " val: " val)
	 (if (and val (not (equal? (vector-ref currrecord idx) val)))
	     (begin
	       (print "Updating " test-name " " fld " to " val)
	       (rmt:testmeta-update-field test-name fld val)))))
     '(("author" 2)("owner" 3)("description" 4)("reviewed" 5)("tags" 9)("jobgroup" 10)))))

;; find tests with matching tags, tagpatt is a string "tagpatt1,tagpatt2%, ..."
;;
(define (runs:get-tests-matching-tags tagpatt)
  (let* ((tagdata (rmt:get-tests-tags))
         (res     '())) ;; list of tests that match one or more tags
    (for-each
     (lambda (tag)
       (if (patt-list-match tag tagpatt)
           (set! res (append (hash-table-ref tagdata tag)))))
     (hash-table-keys tagdata))
    res))
    

;; Update test_meta for all tests
(define (runs:update-all-test_meta db)
  (let ((test-names (tests:get-all))) ;; (tests:get-valid-tests)))
    (for-each 
     (lambda (test-name)
       (let* ((test-conf    (mt:lazy-read-test-config test-name)))

;; Call this to start the actual server
;;

;; all routes though here end in exit ...
;;
;; start_server
;;
(define (server:launch run-id transport-type-raw)
  (let ((transport-type
         (cond
          ((string? transport-type-raw) (string->symbol transport-type-raw))
          (else transport-type-raw))))
         
    ;;(BB> "server:launch fired for run-id="run-id" transport-type="transport-type)

    (case transport-type
      ((http)(http-transport:launch run-id))
      ;;((nmsg)(nmsg-transport:launch run-id))
      ((rpc)  (rpc-transport:launch run-id))
      (else (debug:print-error 0 *default-log-port* "unknown server type " transport-type)))))


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

;; Get the transport
(define (server:get-transport)
  (if *transport-type*
      *transport-type*
      (let ((ttype (string->symbol
		    (or (args:get-arg "-transport")
			(configf:lookup *configdat* "server" "transport")
			*DEFAULT-TRANSPORT*))))
	(set! *transport-type* ttype)
	ttype)))
	    
;; Generate a unique signature for this server
(define (server:mk-signature)
  (message-digest-string (md5-primitive) 
			 (with-output-to-string

	 (curr-ip     (server:get-best-guess-address curr-host))
	 (curr-pid    (current-process-id))
	 (homehost    (common:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" ))
	 (target-host (car homehost))
	 (testsuite   (common:get-testsuite-name))
	 (logfile     (conc *toppath* "/logs/server.log"))
	 (cmdln (conc (common:get-megatest-exe)
		      " -server " (or target-host "-") " -run-id " 0
                      (if (equal? (configf:lookup *configdat* "server" "daemonize") "yes")
                          (conc " -daemonize -log " logfile)
                          "")
                      " -transport " (server:get-transport)
		      " -m testsuite:" testsuite)) ;; (conc " >> " logfile " 2>&1 &")))))
	 (log-rotate  (make-thread common:rotate-logs  "server run, rotate logs thread")))
    ;; we want the remote server to start in *toppath* so push there
    (push-directory *toppath*)
    (debug:print 0 *default-log-port* "INFO: Trying to start server (" cmdln ") ...")
    (thread-start! log-rotate)

	      (file-read-access? dotfile))
	 (with-input-from-file
	     dotfile
	   (lambda ()
	     (read-line)))
	 #f))))


(define (server:dotserver-starting)
  (with-output-to-file
      (conc *toppath* "/.starting-server")
    (lambda ()
      (print (current-process-id) " on " (get-host-name)))))

(define (server:dotserver-starting-remove)
  (delete-file* (conc *toppath* "/.starting-server")))
  

  
;; write a .server file in *toppath* with hostport
;; return #t on success, #f otherwise
;;
(define (server:write-dotserver areapath hostport)
  (let ((lock-file   (conc areapath "/.server.lock"))
	(server-file (conc areapath "/.server")))
    (if (common:simple-file-lock lock-file)
	(let ((res (handle-exceptions
		    exn
		    #f ;; failed for some reason, for the moment simply return #f
		    (with-output-to-file server-file
		      (lambda ()
			(print hostport)))
		    #t)))
	  (debug:print-info 0 *default-log-port* "server file " server-file " for " hostport " created")
	  (common:simple-file-release-lock lock-file)
	  res)
	#f)))

(define (server:remove-dotserver-file areapath hostport #!key (force #f))
  (let ((dotserver   (server:read-dotserver areapath))
	(server-file (conc areapath "/.server"))
	(lock-file   (conc areapath "/.server.lock")))
    (if (or force (and dotserver (string-match (conc ".*:" hostport "$") dotserver))) ;; port matches, good enough info to decide to remove the file
	(if (common:simple-file-lock lock-file)
	    (begin
	      (handle-exceptions
	       exn
	       #f
	       (delete-file* server-file))
	      (debug:print-info 0 *default-log-port* "server file " server-file " for " hostport " removed")
	      (common:simple-file-release-lock lock-file))))))

;; no longer care if multiple servers are started by accident. older servers will drop off in time.
;;
(define (server:check-if-running areapath)
  (let* ((dotserver (server:read-dotserver areapath))) ;; tdbdat (tasks:open-db)))
    (if dotserver
	(let* ((res (case *transport-type*
		      ((http rpc)(server:ping-server dotserver))
		      ;; ((nmsg)(nmsg-transport:ping (tasks:hostinfo-get-interface server)
		      )))
	  (if res
	      dotserver
	      #f))
	#f)))

	  (begin
	    (if host-port-in
		(debug:print 0 *default-log-port*  "ERROR: bad host:port"))
	    (if do-exit (exit 1))
	    #f)
	  (let* ((iface      (car host-port))
		 (port       (cadr host-port))
		 (server-dat
                  (case (remote-transport *runremote*)
                    ((http) (http-transport:client-connect iface port))
                    ((rpc) (rpc-transport:client-connect iface port))
                    (else
                     (debug:print 0 *default-log-port* "ERROR: transport " (remote-transport *runremote*) " not supported (4)")
                     (exit))))
		 (login-res  (rmt:login-no-auto-client-setup server-dat)))
	    (if (and (list? login-res)
		     (car login-res))
		(begin
		  (print "LOGIN_OK")
		  (if do-exit (exit 0)))
		(begin

(define (tasks:hostinfo-get-interface   vec)    (vector-ref  vec 1))
(define (tasks:hostinfo-get-port        vec)    (vector-ref  vec 2))
(define (tasks:hostinfo-get-pubport     vec)    (vector-ref  vec 3))
(define (tasks:hostinfo-get-transport   vec)    (vector-ref  vec 4))
(define (tasks:hostinfo-get-pid         vec)    (vector-ref  vec 5))
(define (tasks:hostinfo-get-hostname    vec)    (vector-ref  vec 6))

(define (tasks:server-lock-slot mdb run-id transport-type)
  (tasks:server-clean-out-old-records-for-run-id mdb run-id " tasks:server-lock-slot")
  (if (< (tasks:num-in-available-state mdb run-id) 4)
      (begin 
	(tasks:server-set-available mdb run-id transport-type)
	(thread-sleep! (/ (random 1500) 1000)) ;; (thread-sleep! 2) ;; Try removing this. It may not be needed.
	(tasks:server-am-i-the-server? mdb run-id))
      #f))
	
;; register that this server may come online (first to register goes though with the process)
(define (tasks:server-set-available mdb run-id transport-type)
  (sqlite3:execute 
   mdb 
   "INSERT INTO servers (pid,hostname,port,pubport,start_time,      priority,state,mt_version,heartbeat,   interface,transport,run_id)
                   VALUES(?, ?,       ?,   ?, strftime('%s','now'), ?,       ?,    ?,-1,?,        ?,        ?);"
   (current-process-id)          ;; pid
   (get-host-name)               ;; hostname
   -1                            ;; port
   -1                            ;; pubport
   (random 1000)                 ;; priority (used a tiebreaker on get-available)
   "available"                   ;; state
   (common:version-signature)    ;; mt_version
   -1                            ;; interface
   ;; (conc (server:get-transport)) ;; transport
   (symbol->string transport-type)    ;; transport
   run-id
   ))

(define (tasks:num-in-available-state mdb run-id)
  (let ((res 0))
    (sqlite3:for-each-row
     (lambda (num-in-queue)

   (let* ((header (list "id" "hostname" "pid" "interface" "port" "pubport" "state" "run_id" "priority" "start_time"))
	  (selstr (string-intersperse header ","))
	  (res    '()))
    (sqlite3:for-each-row
     (lambda (a . b)
       (set! res (cons (apply vector a b) res)))
     mdb
     (conc "SELECT " selstr " FROM servers WHERE state in ('available','running','dbprep') ORDER BY start_time DESC;")

     )
    (vector header res)))

(define (tasks:get-server mdb run-id #!key (retries 10))
  (let ((res  #f)
	(best #f))
    (handle-exceptions
     exn

[server]

# force use of server always
# required yes

# Use http instead of direct filesystem access
transport rpc
# transport http
# transport fs
# transport nmsg

synchronous 0

# If the server can't be started on this port it will try the next port until
# it succeeds

[ezsteps]
calcresults megatest -list-runs $MT_RUNNAME -target $MT_TARGET
check_triggers  cat $MT_RUN_AREA_HOME/triggers_$MT_RUNNAME.dat

[logpro]
check_triggers ;;
  (expect:error in "LogFileBody" = 0 "No errors" #/error/i)

[requirements]
waiton  #{getenv ALL_TOPLEVEL_TESTS}

(use srfi-18)


;; wrap a proc with a mutex so that two threads may not call proc simultaneously.
;; will catch exceptions to ensure mutex is unlocked even if exception is thrown.
;; will generate a unique mutex for proc unless one is specified with canned-mutex: option
;;
;; example 1: (define thread-safe-+ (make-synchronized-proc +))
;; example 2: (define thread-safe-plus
;;               (make-synchronized-proc
;;                  (lambda (x y)
;;                      (+ x y))))

(define (make-synchronized-proc proc
                                #!key (canned-mutex #f))
  (let* ((guard-mutex (if canned-mutex canned-mutex (make-mutex)))
         (guarded-proc ;; we are guarding the thunk against exceptions.  We will record whether result of evaluation is an exception or a regular result.
          (lambda args
            (mutex-lock! guard-mutex)
            (let* ((EXCEPTION (gensym)) ;; using gensym to avoid potential collision with a proc that returns a pair having the first element be our flag.  gensym guarantees the symbol is unique.
                   (res
                    (condition-case
                     (apply proc args) ;; this is what we are guarding the execution of
                     [x () (cons EXCEPTION x)]
                     )))
              (mutex-unlock! guard-mutex)
              (cond
               ((and (pair? res) (eq? (car res) EXCEPTION))
                (raise (cdr res)))
               (else
                res))))))
    guarded-proc))


;; retry an operation (depends on srfi-18)
;; ==================
;; idea here is to avoid spending time on coding retrying something.  Trying to be generic here.
;;
;; Exception handling:
;; -------------------
;; if evaluating the thunk results in exception, it will be retried.
;; on last try, if final-failure-returns-actual is true, the exception will be re-thrown to caller.
;;
;; look at options below #!key to see how to configure behavior
;;
;;

(define (retry-thunk
         the-thunk
         #!key ;;;; options below
         (accept-result?   (lambda (x) x)) ;; retry if predicate applied to thunk's result is false 
         (retries                       4) ;; how many tries
         (failure-value                #f) ;; return this on final failure, unless following option is enabled:
         (final-failure-returns-actual #f) ;; on failure, on the last try, just return the result, not failure-value

         (retry-delay                 0.1) ;; delay between tries
         (back-off-factor               1) ;; multiply retry-delay by this factor on retry
         (random-delay                0.1) ;; add a random portion of this value to wait

         (chatty                       #f) ;; print status as we go, for debugging.
         )
  
  (when chatty (print) (print "Entered retry-thunk") (print "-=-=-=-=-=-"))
  (let* ((guarded-thunk ;; we are guarding the thunk against exceptions.  We will record whether result of evaluation is an exception or a regular result.
          (lambda ()
           (let* ((EXCEPTION (gensym)) ;; using gensym to avoid potential collision
                  (res
                   (condition-case
                    (the-thunk) ;; this is what we are guarding the execution of
                    [x () (cons EXCEPTION x)]
                    )))
             (cond
              ((and (pair? res) (eq? (car res) EXCEPTION))
               (if chatty
                   (print " - the-thunk threw exception >"(cdr res)"<"))
               (cons 'exception (cdr res)))
               (else
                (if chatty
                    (print " - the-thunk returned result >"res"<"))
                (cons 'regular-result res)))))))
    
    (let loop ((guarded-res (guarded-thunk))
               (retries-left retries)
               (fail-wait retry-delay))
      (if chatty (print "   =========="))
      (let* ((wait-time (+ fail-wait (+ (* fail-wait back-off-factor)
                                        (* random-delay
                                           (/ (random 1024) 1024) ))))
             (res-type (car guarded-res))
             (res-value (cdr guarded-res)))
        (cond
         ((and (eq? res-type 'regular-result) (accept-result? res-value))
                   (if chatty (print " + return result that satisfied accept-result? >"res-value"<"))
                   res-value)

         ((> retries-left 0)
          (if chatty (print " - sleep "wait-time))
          (thread-sleep! wait-time)
          (if chatty (print " + retry ["retries-left" tries left]"))
          (loop (guarded-thunk)
                (sub1 retries-left)
                wait-time))
         
         ((eq? res-type 'regular-result)
          (if final-failure-returns-actual
              (begin
                (if chatty (print " + last try failed- return the result >"res-value"<"))
                res-value)
              (begin
                (if chatty (print " + last try failed- return canned failure value >"failure-value"<"))
              failure-value)))
         
         (else ;; no retries left; result was not accepted and res-type can only be 'exception
          (if final-failure-returns-actual 
              (begin
                (if chatty (print " + last try failed with exception- re-throw it >"res-value"<"))
                (abort res-value)); re-raise the exception. TODO: find a way for call-history to show as though from entry to this function
              (begin
                (if chatty (print " + last try failed with exception- return canned failure value >"failure-value"<"))
                failure-value))))))))

#  greater. See the accompanying file COPYING for details.
# 
#  This program is distributed WITHOUT ANY WARRANTY; without even the
#  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
#  PURPOSE.

echo You may need to do the following first:
echo sudo apt-get install libreadline-dev libsqlite3-dev libwebkitgtk-dev 

echo sudo apt-get install libpangox-1.0-0 zlib1g-dev libfreetype6-dev cmake
echo sudo apt-get install libssl-dev uuid-dev libglu1-mesa-dev
echo sudo apt-get install libmotif3 -OR- set KTYPE=26g4
echo
echo Set OPTION to std, currently OPTION=$OPTION
echo
echo Additionally, if you want mysql-client, you will need to make sure
echo mysql_config is in your path
echo
echo You are using PREFIX=$PREFIX
echo You are using proxy="$proxy"
echo 
echo "Set additional_libpath to help find gtk or other libraries, don't forget a leading :"

if [[ "$OPTION"x == "x" ]];then
  OPTION=std
fi

SYSTEM_TYPE=$(lsb_release -irs |tr ' ' '_' |tr '\n' '-')$(uname -i)-$OPTION

# default chicken version variables. Override in case statement as appropriate
CHICKEN_VERSION=4.10.0
CHICKEN_BASEVER=4.10.0

# Set up variables
#
case $SYSTEM_TYPE in
Ubuntu-16.04-x86_64-std)
	KTYPE=32
	CDVER=5.10
	IUPVER=3.17
	IMVER=3.11
	;;
Ubuntu-16.04-x86_64-new)
	KTYPE=32
	CDVER=5.10
	IUPVER=3.17
	IMVER=3.11
	CHICKEN_VERSION=4.10.0
	CHICKEN_BASEVER=4.10.0
	;;
Ubuntu-16.04-i686-std)
	KTYPE=32
	CDVER=5.10
	IUPVER=3.17
	IMVER=3.11
	;;

  echo Using KTYPE=$KTYPE
fi

# Put all the downloaded tar files in tgz
mkdir -p tgz

# http://code.call-cc.org/releases/4.8.0/chicken-4.8.0.5.tar.gz


chicken_targz=chicken-${CHICKEN_VERSION}.tar.gz
if ! [[ -e tgz/$chicken_targz ]]; then 
    wget http://code.call-cc.org/releases/${CHICKEN_BASEVER}/${chicken_targz}
    mv $chicken_targz tgz
fi 

BUILDHOME=$PWD

	fi
fi
cd $BUILDHOME

# Some eggs are quoted since they are reserved to Bash
# for f in matchable readline apropos base64 regex-literals format "regex-case" "test" coops trace csv dot-locking posix-utils posix-extras directory-utils hostinfo tcp rpc csv-xml fmt json md5; do
# $CHICKEN_INSTALL $PROX -keep-installed matchable readline apropos base64 regex-literals format "regex-case" "test" coops trace csv dot-locking posix-utils posix-extras directory-utils hostinfo tcp rpc csv-xml fmt json md5 awful http-client spiffy uri-common intarweb http-client spiffy-request-vars md5 message-digest http-client spiffy-directory-listing
for egg in matchable readline apropos dbi base64 regex-literals format "regex-case" "test" \
	coops trace csv dot-locking posix-utils posix-extras directory-utils hostinfo \
	tcp rpc csv-xml fmt json md5 awful http-client spiffy uri-common intarweb http-client \
	spiffy-request-vars s md5 message-digest spiffy-directory-listing ssax sxml-serializer \
	sxml-modifications logpro z3 call-with-environment-variables \
	pathname-expand typed-records simple-exceptions numbers crypt parley srfi-42 \
	alist-lib ansi-escape-sequences args basic-sequences bindings chicken-doc chicken-doc-cmd \
	cock condition-utils debug define-record-and-printer easyffi easyffi-base \

	cd opensrc
	#fossil open ../opensrc.fossil
	cd histstore
	$PREFIX/bin/csc histstore.scm -o hs 
	cp -f hs $PREFIX/bin/hs 
	cd ../mutils
	$PREFIX/bin/chicken-install
	# cd ../dbi 
	# $PREFIX/bin/chicken-install
	cd ../margs
	$PREFIX/bin/chicken-install
fi
cd $BUILDHOME

if ! [[ -e $PREFIX/bin/stmlrun ]] ; then
	#fossil clone http://www.kiatoa.com/fossils/stml stml.fossil