Megatest

recent-commits.csv : .fslckout
	fossil timeline -n 350 -t ci -F "%h,%a,%b,%t,\"%c\"" > recent-commits.csv


SHELL=/bin/bash
PREFIX=$(PWD)
# CSCOPTS=-lfa2 -specialize -inline-global
CSCOPTS=
INSTALL=install
SRCFILES = common.scm items.scm launch.scm ods.scm runconfig.scm	\
           server.scm configf.scm db.scm keys.scm		\
           process.scm runs.scm tasks.scm tests.scm genexample.scm	\
           tdb.scm mt.scm	\
           ezsteps.scm rmt.scm api.scm		\

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

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

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

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

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

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

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

    tasks-set-state-given-param-key
    ))

(define *db-write-mutexes* (make-hash-table))
(define *server-signature* #f)

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

(define (api:get-thread-command th-in)
  (let ((thread-data (assoc th-in *api-threads*)))
    (if thread-data
        (third thread-data) ; Assuming the command is the third element in the list
        #f))) ; Return #f if the thread is not found

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

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

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

(define (api:get-threads)
  (map (lambda (thdat)
         (let ((thread (first thdat))
               (timestamp (second thdat))
               (command (third thdat)))
           (format "\nThread: ~a, age: ~a, Command: ~a" thread (- (current-seconds) timestamp) command)))
       *api-threads*))


(define *api:last-stats-print* 0)
(define *api-print-db-stats-mutex* (make-mutex))
(define (api:print-db-stats)
  (debug:print-info 0 *default-log-port* "Started periodic db stats printer")
  (let loop ()
    (mutex-lock! *api-print-db-stats-mutex*)

;;          reads/writes to current in/out port
;;
(define (api:tcp-dispatch-request-make-handler dbstruct) ;; cmd run-id params)
  (assert *toppath* "FATAL: api:tcp-dispatch-request-make-handler called but *toppath* not set.")
  (if (not *server-signature*)
      (set! *server-signature* (tt:mk-signature *toppath*)))
  (lambda (indat)
    (api:register-thread (current-thread))
    (api:register-thread (current-thread) (car indat))
    (let* ((result 
	    (let* ((numthreads (api:get-count-threads-alive))
		   (delay-wait (if (> numthreads 10)
				   (- numthreads 10)
				   0))
		   (normal-proc (lambda (cmd run-id params)
				  (case cmd

					 ((ping) #t) ;; we are fine
					 (else
					  (assert ok "FATAL: database file and run-id not aligned.")))))
			     (ttdat   *server-info*)
			     (server-state (tt-state ttdat))
			     (maxthreads   20) ;; make this a parameter?
			     (status  (cond
				       ((and (> numthreads maxthreads)
					     (> (random 100) 70)) ;; allow a 30% probability to go through so we can figure out what is going wrong in main.db server.
					'busy)
				       ;; ((> numthreads 5) 'loaded) ;; this gets transmitted to the client which calls tt:backoff-incr to slow stuff down.
                                       ((> numthreads maxthreads)
					(let* ((testsuite (common:get-testsuite-name))
					       (mtexe     (common:find-local-megatest))
					       (proc      (lambda ()
							    ;; we are overloaded, try to start another server
							    (debug:print 0 *default-log-port* "Too many threads running, starting another server")
							    (tt:server-process-run *toppath* testsuite mtexe run-id))))
					  (set! *server-start-requests* (cons proc *server-start-requests*)))
					;; 'busy
					'loaded ;; not ideal since the client will not backoff
					)
				       (else 'ok)))
			     (errmsg  (case status
					((busy)   (conc "Server overloaded, "numthreads" threads in flight"))
					((busy)   (conc "Server overloaded, "numthreads" threads in flight, current cmd: " cmd "\n current threads: " (api:get-threads)))
					((loaded) (conc "Server loaded, "numthreads" threads in flight"))
					(else     #f)))
			     (result  (case status
					((busy)
					 (if (eq? cmd 'ping)
					     (normal-proc cmd run-id params)
					     ;; numthreads must be greater than 5 for busy
					     (* 0.1 (- numthreads maxthreads)) ;; was 15
					     )) ;; (- numthreads 29)) ;; call back in as many seconds
					((loaded)
					 ;; 			    (if (eq? (rmt:transport-mode) 'tcp)
					 ;; 				(thread-sleep! 0.5))
					 (normal-proc cmd run-id params))
					(else
					 (normal-proc cmd run-id params))))
			     (meta   (case cmd
				       ((ping) `((sstate . ,server-state)))
				       ((ping) `((sstate . ,server-state)(sload . ,numthreads)))
				       (else   `((wait . ,delay-wait)))))
			     (payload (list status errmsg result meta)))
			;; (cmd run-id params meta)
			(db:add-stats cmd run-id params (- (current-milliseconds) start-t))
			payload))
		     (else
		      (assert #f "FATAL: failed to deserialize indat "indat))))))
      ;; (set! *api-process-request-count* (- *api-process-request-count* 1))
      ;; (serialize payload)
     
      (api:unregister-thread (current-thread))
      result)))



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

(define (api:dispatch-request dbstruct cmd run-id params)
  (if (not *no-sync-db*)
      (db:open-no-sync-db))
  (let* ((start-time (current-milliseconds)))

(use posix-extras pathname-expand files)


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

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

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

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

  (- megatest-version (common:get-last-run-version-number)))

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


;; From 1.70 to 1.80, db's are compatible.

;; From 1.70 to 1.81, db's are compatible.
;;
;; BUG: This logic is almost certainly not quite correct.
;;
(define (common:api-changed?)
  (let* (
    (megatest-major-version (substring (->string megatest-version) 0 4))
    (run-major-version (substring (conc (common:get-last-run-version)) 0 4))
  (let* ((megatest-major-version (substring (->string megatest-version) 0 4))
	 (run-major-version (substring (conc (common:get-last-run-version)) 0 4)))
   )
   (and (not (equal? megatest-major-version "1.80"))
     (not (equal? megatest-major-version megatest-run-version)))
    (and (not (member megatest-major-version '("1.81" "1.80")))
	 (not (equal? megatest-major-version run-major-version)))))
  )
)

;;======================================================================
;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma
;;
(define (common:cleanup-db dbstruct #!key (full #f))
  (case (rmt:transport-mode)

			(list "Testname: "
			      "Item path: "
			      "Current state: "
			      "Current status: "
			      "Test comment: "
			      "Test id: "
			      "Test date: "))
		   (list (iup:label "" #:expand "VERTICAL"))))
		   (list (iup:label "" #:expand "VERTICAL"
				    ))))
    (apply iup:vbox  ; #:expand "YES"
	   (list 
	    (store-label "testname"
			 (iup:label (db:test-get-testname  testdat) #:expand "HORIZONTAL")
			 (lambda (testdat)(db:test-get-testname testdat)))
	    (store-label "item-path"
			 (iup:label (db:test-get-item-path testdat) #:expand "HORIZONTAL")

			  (iup:label val ; #:expand "HORIZONTAL"
				     ))
			(list "Author: "
			      "Owner: "
			      "Reviewed: "
			      "Tags: "
			      "Description: "))
		   (list (iup:label "" #:expand "VERTICAL"))))
		   (list (iup:label "" #:expand "VERTICAL"
				    ))))
    (apply iup:vbox  ; #:expand "YES"
	   (list 
	    (store-meta "author"
			 (iup:label (db:testmeta-get-author testmeta) #:expand "HORIZONTAL")
			 (lambda (testmeta)(db:testmeta-get-author testmeta)))
	    (store-meta "owner"
			 (iup:label (db:testmeta-get-owner testmeta) #:expand "HORIZONTAL")
			 (lambda (testmeta)(db:testmeta-get-owner testmeta)))
	    (store-meta "reviewed" 
			 (iup:label (db:testmeta-get-reviewed testmeta) #:expand "HORIZONTAL")
			 (lambda (testmeta)(db:testmeta-get-reviewed testmeta)))
	    (store-meta "tags" 
			 (iup:label (db:testmeta-get-tags testmeta) #:expand "HORIZONTAL")
			 (lambda (testmeta)(db:testmeta-get-tags testmeta)))
	    (store-meta "description" 
			 (iup:label (test-meta-panel-get-description testmeta) #:size "x50"); #:expand "HORIZONTAL")
			(iup:label (test-meta-panel-get-description testmeta) ;; #:wordwrap "YES" ;; #:size "x50"
				   ) ;; #:expand "HORIZONTAL")
			 (lambda (testmeta)
			   (test-meta-panel-get-description testmeta)))
	    )))))


;;======================================================================
;; Run info panel

			    (iup:label (conc (car keyval) " ")))
			  keydat)
		     (list (iup:label "runname ")
			   (iup:label "run-id")
			   (iup:label "run-date"))))
      (apply iup:vbox
	     (append (map (lambda (keyval)
			    (iup:vbox
			    (iup:label (cadr keyval) #:expand "HORIZONTAL"))
			     (iup:label (cadr keyval) #:expand "HORIZONTAL")
			     ;; (iup:label "" #:expand "BOTH")
			     )
			    )
			  keydat)
		     (list (iup:label runname)
			   (iup:label (conc run-id))
			   (iup:label (seconds->year-work-week/day-time event_time))
			   (iup:label "" #:expand "VERTICAL"))))))))
			   (iup:label "" ;;#:expand "VERTICAL"
				      ))))))))
  
;;======================================================================
;; Host info panel
;;======================================================================
(define (host-info-panel testdat store-label)
  (iup:frame
   #:title "Remote host and Test Run Info" ; #:expand "YES"
   (iup:hbox ; #:expand "YES"
    (apply iup:vbox ; #:expand "YES" ;; The heading labels
	   (append (map (lambda (val)
			  (iup:label val ; #:expand "HORIZONTAL"
				     ))
			(list "Hostname: "
			      "Disk free: "
			      "CPU Load: "
			      "Run duration: "
			      "Logfile: "
			      "Top process id: "
			      "Uname -a: "))
		   (iup:label "" #:expand "VERTICAL")))
		   (iup:label "" ;; #:expand "VERTICAL"
			      )))
    (apply iup:vbox ; #:expand "YES"
	   (list
	    ;; NOTE: Yes, the host can change!
	    (store-label "HostName"
			 (iup:label ;; (sdb:qry 'getstr 
			  (db:test-get-host testdat) ;; )
			  #:expand "HORIZONTAL")

;; if there is a submegatest create a button to launch dashboard in that area
;;
(define (submegatest-panel dbstruct keydat testdat runname testconfig)
  (let* ((test-run-dir      (db:test-get-rundir testdat))
	 (subarea           (subrun:get-runarea test-run-dir))
	 (area-exists       (and subarea (common:file-exists? subarea silent: #t))))
    (if subarea
	(iup:frame 
	 #:title "Megatest Run Info" ; #:expand "YES"
    (iup:frame 
     #:title "Megatest Run Info" ;; #:expand "HORIZONTAL"
     (if subarea
	 (iup:button
	  "Launch Dashboard"
	  #:action (lambda (obj)
                     (subrun:launch-dashboard test-run-dir))))
	(iup:vbox))))
                     (subrun:launch-dashboard test-run-dir)))
	 (iup:vbox
	  (iup:label "Not a subrun..." #:expand "HORIZONTAL")
	  )))))

;; use a global for setting the buttons colors
;;                           state status teststeps
(define *state-status* (vector #f #f #f))
(define (update-state-status-buttons testdat)
  (let* ((state  (db:test-get-state  testdat))
	 (status (db:test-get-status testdat))

			    (iup:attribute-set! 
			     command-text-box "VALUE"
			     (conc "megatest -target " keystring " -runname " runname 
				   " -run -testpatt " (conc testname "/" (if (equal? item-path "")
									"%" 
									item-path))
				   " -clean-cache"
				   ))))
	       (rerun-clean  (lambda (x)
			    (iup:attribute-set! 
			     command-text-box "VALUE"
			     (conc "megatest -target " keystring " -runname " runname 
				   " -rerun-clean -testpatt " (conc testname "/" (if (equal? item-path "")
									"%" 
									item-path))
				   " -clean-cache"
				   ))))
	       (remove-test (lambda (x)
			      (iup:attribute-set!
			       command-text-box "VALUE"
			       (conc "megatest -remove-runs -target " keystring " -runname " runname
				     " -testpatt " (conc testname "/" (if (equal? item-path "")
									  "%"

	   ((not testdat)(begin (debug:print 0 *default-log-port* "ERROR: bad test info for " test-id)(exit 1)))
	   ((not rundat)(begin (debug:print 0 *default-log-port* "ERROR: found test info but there is a problem with the run info for " run-id)(exit 1)))
	   (else
	    ;;  (test-set-status! db run-id test-name state status itemdat)
	    (set! self ; 
		  (iup:dialog #:close_cb (lambda (a)(exit)) ; #:expand "YES"
			      #:title testfullname
			      (iup:vbox
			       (iup:hbox
			      (iup:vbox ; #:expand "YES"
			       ;; The run and test info
			       (iup:hbox  ; #:expand "YES"
				(run-info-panel dbstruct keydat testdat runname)
				(test-info-panel testdat store-label widgets)
				(test-meta-panel testmeta store-meta))
			       (iup:hbox
				(host-info-panel testdat store-label)
				(submegatest-panel dbstruct keydat testdat runname testconfig))
				(iup:vbox ; #:expand "YES"
				 ;; The run and test info
				 (iup:hbox  ; #:expand "YES"
				  (run-info-panel dbstruct keydat testdat runname)
				  (test-info-panel testdat store-label widgets))
				 (host-info-panel testdat store-label))
				(iup:vbox
				 (test-meta-panel testmeta store-meta)
				 (submegatest-panel dbstruct keydat testdat runname testconfig)))
			       ;; The controls
			       (iup:frame #:title "Actions" 
			       (iup:hbox ;; frame #:title "Actions" 
					  (iup:vbox
					   (iup:hbox 
					    (iup:button "View Log"      #:action viewlog      #:size "80x")
					    (iup:button "Start Xterm"   #:action xterm        #:size "80x")
					    (iup:button "Run Test"      #:action run-test     #:size "80x")
					    (iup:button "Clean Test"    #:action remove-test  #:size "80x")
					   (iup:hbox
					    (iup:frame
					     #:title "Immediate"
					     (iup:hbox
					      (iup:button "Start Xterm"   #:action xterm        #:size "80x")
					      (iup:button "CleanRunExecute!"    #:action clean-run-execute #:size "80x")
					      (iup:button "View Log"      #:action viewlog      #:size "80x")))
					    (iup:frame
					     #:title "Command line"
					     (iup:hbox
					      (iup:button "Run Test"      #:action run-test     #:size "80x")
					      (iup:button "Rerun-clean"   #:action rerun-clean  #:size "80x")
					      (iup:button "Clean Test"    #:action remove-test  #:size "80x")
					    (iup:button "CleanRunExecute!"    #:action clean-run-execute #:size "80x")
					    (iup:button "Kill All Jobs" #:action kill-jobs    #:size "80x")
					    (iup:button "Archive Test"  #:action archive-test #:size "80x")
					    (iup:button "Close"         #:action (lambda (x)(exit)) #:size "80x"))
					      (iup:button "Kill All Jobs" #:action kill-jobs    #:size "80x")))
					    (iup:label "" #:expand "HORIZONTAL")
					    (iup:frame
					     #:title "Other"
					     (iup:hbox
					      ;; (iup:button "Archive Test"  #:action archive-test #:size "80x")
					      (iup:button "Close"         #:action (lambda (x)(exit)) #:size "80x")
					      )))
					   (apply 
					    iup:hbox
					    (list command-text-box command-launch-button))))
			       (set-fields-panel dbstruct run-id test-id testdat)
			       (let ((tabs 
				      (iup:tabs
				       ;; Replace here with matrix

;;    please-update:        #t
;;    update-mutex:         (make-mutex)
;;    updaters:             (make-hash-table)
;;    updating:             #f
;;    hide-not-hide-tabs:   #f
;;    target:               ""
;;    ))

(set! *journal-stats-enable* #f)

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

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

	fossil add html/*

megatest.pdf : megatest.lyx
	lyx -e pdf2 megatest.lyx

pkts.pdf : pkts.dot
	dot -Tpdf pkts.dot -o pkts.pdf

stepwise.pdf : stepwise-rpc-via-direct-and-tcp-or-http.dot
	dot stepwise-rpc-via-direct-and-tcp-or-http.dot -Tpdf -o stepwise.pdf

		 (lambda ()
		   (print content)
		   ;; (change-file-mode name (bitwise-ior perm/irwxg perm/irwxu))
		   )))
	      (else
	       (debug:print-info 0 "Invalid logpro definiton found in [logpro] section of testconfig. \"" logprodat "\""))))
     logpros)))

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

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

(define (launch:execute encoded-cmd)
  (let* ((cmdinfo    (common:read-encoded-string encoded-cmd))
	 (tconfigreg #f))
    (setenv "MT_CMDINFO" encoded-cmd)
    ;;(bb-check-path msg: "launch:execute incoming")
    (if (list? cmdinfo) ;; ((testpath /tmp/mrwellan/jazzmind/src/example_run/tests/sqlitespeed)

		 (monitorjob   (lambda ()
				 (launch:monitor-job  run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m work-area runtlim misc-flags)))
		 (th1          (make-thread monitorjob "monitor job"))
		 (th2          (make-thread runit "run job"))
                 (tconfig         (tests:get-testconfig test-name item-path tconfigreg #t))
                 (propagate-exit-code (configf:lookup *configdat* "setup" "propagate-exit-code"))
                 (propagate-status-list '("FAIL" "KILLED" "ABORT" "DEAD" "CHECK" "SKIP" "WAIVED"))
                 (test-status "not set")
                 (test-status     "not set")
		 (test-state      "not set")
		 (precmd          (configf:lookup tconfig "setup" "precmd"))
		 (postcmd         (configf:lookup tconfig "setup" "postcmd")))
	    ;; first, if set, run the precmd
	    (if precmd ;; (file-exists? precmd)(file-execute-access? precmd))
		(begin
		  ;; (save-environment-as-files "precmd-envt")
		(system precmd)) ;; up to test author to put nbfake if desired.
		  (system precmd))) ;; up to test author to put nbfake if desired.
	    (set! job-thread th2)
	    (thread-start! th1)
	    (thread-start! th2)
	    (thread-join! th2)
	    (debug:print-info 0 *default-log-port* "Megatest execute of test " test-name ", item path " item-path " complete. Notifying the db ...")
            (debug:print-info 2 *default-log-port* "exit-info = " exit-info)
	    (hash-table-set! misc-flags 'keep-going #f)

	      (rmt:update-run-stats run-id (rmt:get-raw-run-stats run-id))) ;; end of let*

	    (mutex-unlock! m)
            (launch:end-of-run-check run-id )
	    (debug:print 2 *default-log-port* "Output from running " fullrunscript ", pid " (launch:einf-pid exit-info) " in work area " 
			 work-area ":\n====\n exit code " (launch:einf-exit-code exit-info) "\n" "====\n")


            (set! test-status (db:test-get-status (rmt:get-testinfo-state-status run-id test-id)))
	    (let* ((testrec  (rmt:get-testinfo-state-status run-id test-id)))
              (set! test-status (db:test-get-status testrec))
	      (set! test-state  (db:test-get-state  testrec)))

            ;; If the propagate-exit-code option has been set in the megatest config, and the test status matches the list, set the exit code to 1.

	    (if postcmd
		(system postcmd))

            (if (and propagate-exit-code (string=? propagate-exit-code "yes") (member test-status propagate-status-list))
               (begin
                (debug:print 1 *default-log-port* "Setting exit status to 1 because of test status of " test-status) 
                (set! *globalexitstatus* 1)
               )
            )

	    (if postcmd
		(begin
		  (setenv "MT_TEST_STATE" test-state)
		  (setenv "MT_TEST_STATUS" test-status)
		  ;; (save-environment-as-files "postcmd-envt")
		  (system postcmd)))
	    (if (not (launch:einf-exit-status exit-info))
		(exit 4))))
        )))

;; Spec for End of test
;; At end of each test call, after marking self as COMPLETED do run-state-status-rollup
;; At transition to run COMPLETED/X do hooks

;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;; Always use two or four digit decimal
;; 1.01, 1.02...1.10,1.11,1.1101 ... 1.99,2.00..

;; (declare (unit megatest-version))

(define megatest-version 1.8031)
(define megatest-version 1.8101)

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

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

  -mark-incompletes       : find and mark incomplete tests
  -ping run-id|host:port  : ping server, exit with 0 if found
  -debug N|N,M,O...       : enable debug 0-N or N and M and O ...
  -debug-noprop N|M,M,O...: enable debug but do not propagate to subprocesses via MT_DEBUG
  -config fname           : override the megatest.config file with fname
  -append-config fname    : append fname to the megatest.config file
  -import-sexpr fname     : import a sexpr file (use -list-runs % -dumpmode sexpr to create)
  -remove-dbs all         : remove Megatest DBs before importing sexpr. (Use only with -import-sexpr)
  -regen-testfiles        : regenerate scripts and logpro files from testconfig, run in test context
  
Utilities
  -env2file fname         : write the environment to fname.csh and fname.sh
  -envcap a               : save current variables labeled as context 'a' in file envdat.db
  -envdelta a-b           : output enviroment delta from context a to context b to -o fname
                            set the output mode with -dumpmode csh, bash or ini

	   (let* ((db       #f)
		  ;; DO NOT run remote
		  (paths    (tests:test-get-paths-matching keys target (args:get-arg "-test-files"))))
	     (for-each (lambda (path)
			 (print path))
		       paths))))))

;;======================================================================
;; Utils for test areas
;;======================================================================

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

(if (args:get-arg "-regen-testfiles")
    (if (getenv "MT_TEST_RUN_DIR")
	(begin

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

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

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

(module portlogger
*

;; NB// area-dat replaced by ttdat
;; 
(define (rmt:send-receive cmd run-id params #!key (attemptnum 1)(ttdat #f))
  (assert (or (not run-id) (number? run-id)) "FATAL: run-id is required to be a number or #f")
  (assert *toppath* "FATAL: rmt:send-receive called with *toppath* not set.")
  (let* ((areapath      *toppath*) ;; TODO - resolve from dbstruct to be compatible with multiple areas
	 (readonly-mode (rmtmod:calc-ro-mode ttdat *toppath*))
	 (testsuite     (common:get-testsuite-name)))
	 (testsuite     (common:get-testsuite-name))
	 (dbfname       (conc (dbfile:run-id->dbnum run-id)".db"))
	 (dbdir         (conc areapath "/.mtdb")))
    (if (and (not *journal-stats*)
	     (file-exists? dbdir))
	(tt:start-stats dbdir)) ;; fixme - find the right call to get the db directory
    
    ;; check the load on dbfname and add some delay using a droop curve of sorts
    (if *journal-stats*
	(let* ((load  (tt:get-journal-stats dbfname)))
	  (if (> load 0.1) ;; start activating delay at 10% journal load time
	      (let ((dely (* 50 (* load load)))) ;; 100% journal time=50sec delay
		(debug:print 0 *default-log-port* "Journal load "load" on "dbfname" delaying queries "dely"s.")
		(thread-sleep! dely)))))
	
    (case (rmt:transport-mode)
      ((tcp)
       (let* ((start-time    (current-seconds)) ;; snapshot time so all use cases get same value
	      (attemptnum    (+ 1 attemptnum))
	      (mtexe         (common:find-local-megatest))
	      (dbfname       (conc (dbfile:run-id->dbnum run-id)".db"))
	      (ttdat         (rmt:set-ttdat areapath ttdat))
	      (conn          (tt:get-conn ttdat dbfname))
	      (is-main       (equal? dbfname "main.db")) ;; why not (not run-id) ?
	      (server-start-proc (if is-main
				     #f
				     (lambda ()
				       ;; (debug:print-info 0 *default-log-port* "starting server for dbfname: "dbfname)

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

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

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

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

(declare (unit rmtmod))
(declare (uses debugprint))
;; (declare (uses debugprint.import))
(declare (uses commonmod))
;; (declare (uses commonmod.import))
(declare (uses dbfile))    ;; needed for records
(declare (uses dbmod))
;; (declare (uses tcp-transportmod))
;; (declare (uses tcp-transportmod.import))

;; (declare (uses apimod))
;; (declare (uses apimod.import))
;; (declare (uses ulex))

;; (include "ulex/ulex.scm")

(module rmtmod
	*
	
(import scheme chicken data-structures extras matchable srfi-69)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18)
(import commonmod dbfile debugprint) ;; (prefix commonmod cmod:))
(import dbmod
	;; tcp-transportmod
	)

;; (import apimod)
;; (import (prefix ulex ulex:))

(include "db_records.scm")

(defstruct alldat
  (areapath #f)

			      run-id test-id 'foo "COMPLETED" "DEAD"
			      "Test stopped responding while in RUNNING or REMOTEHOSTSTART; presumed dead.")))))))
	     ;; call end of eud of run detection for posthook - from merge, is it needed?
	     ;; (launch:end-of-run-check run-id)
	     all-ids)
	    )))))

;;======================================================================
;; Misc
;;======================================================================

;; (define (rmtmod:wait-on-server-load run-id ttdat)
;;   (let* ((dbfname                 (dbmod:run-id->dbfname run-id))
;; 	 (get-lowest-thread-load
;; 	  (lambda ()
;; 	    (let* ((sdats (tt:get-server-info-sorted ttdat dbfname)))
;; 	      (car (map tt:get-server-threads sdats))))))
;;     (if ttdat
;; 	(let loop ()
;; 	  (if (> (get-lowest-thread-load) 5) ;; load is pretty high
;; 	      (begin
;; 		(debug:print 0 *default-log-port* "Servers appear overloaded, waiting...")
;; 		(thread-sleep! 1)
;; 		(loop))))
;; 	(debug:print 0 *default-log-port* "Can't wait on server load, *ttdat* not set"))))
)

)

(declare (uses tests))
(declare (uses server))
(declare (uses mt))
(declare (uses archive))
(declare (uses mtargs))
(declare (uses rmtmod))
(declare (uses dbfile))
(declare (uses tcp-transportmod))

(use (prefix sqlite3 sqlite3:) srfi-1 posix regex regex-case srfi-69 (srfi 18) 
     posix-extras directory-utils pathname-expand typed-records format  sxml-serializer
     sxml-modifications matchable)



(include "common_records.scm")
(include "key_records.scm")
(include "db_records.scm")
(include "run_records.scm")
(include "test_records.scm")

;; (include "debugger.scm")

(import commonmod
	debugprint
	rmtmod
	dbfile
	tcp-transportmod
	(prefix mtargs args:))

;; use this struct to facilitate refactoring
;;

(defstruct runs:dat
  reglen regfull

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

	  (if (list? items-list)
	      (begin
		(if (null? items-list)
		    (let ((test-id   (rmt:get-test-id run-id test-name ""))
			  (num-items (rmt:test-toplevel-num-items run-id test-name)))
		      (if (and test-id
			       (not (> num-items 0)))
			  (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "ZERO_ITEMS" "Failed to run due to failed prerequisites"))))
			  (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "ZERO_ITEMS" (conc "Failed to run due to failed prerequisites: " (runs:pretty-string fails))))))
		(tests:testqueue-set-items! test-record items-list)
		(list hed tal reg reruns))
	      (begin
		(debug:print-error 0 *default-log-port* "The proc from reading the items table did not yield a list - please report this")
		(exit 1))))))

    ((and (null? fails)

			    (set! give-up #t)))
		      prereqstrs))

	(if (and give-up
		 (not (and (null? tal)(null? reg))))
	    (let ((trimmed-tal (mt:discard-blocked-tests run-id hed tal test-records))
		  (trimmed-reg (mt:discard-blocked-tests run-id hed reg test-records)))
	      (debug:print 1 *default-log-port* "WARNING: test " hed " has discarded prerequisites, removing it from the queue")
	      (debug:print 1 *default-log-port* "WARNING: test " hed " has discarded prerequisites: " prereqstrs ", removing it from the queue")

	      (let ((test-id (rmt:get-test-id run-id hed "")))
		(if test-id (mt:test-set-state-status-by-id run-id test-id "COMPLETED" "PREQ_DISCARDED" "Failed to run due to discarded prerequisites")))
		(if test-id (mt:test-set-state-status-by-id run-id test-id "COMPLETED" "PREQ_DISCARDED" (conc "Failed to run due to discarded prerequisites: " prereqstrs))))
	      
	      (if (and (null? trimmed-tal)
		       (null? trimmed-reg))
		  #f
                  (runs:loop-values trimmed-tal trimmed-reg reglen regfull reruns)
                  ))
	      (list (car newtal)(append (cdr newtal) reg) '() reruns))))

    ((and (null? fails) ;; have not-started tests, but unable to run them.  everything looks completed with no prospect of unsticking something that is stuck.  we should mark hed as moribund and exit or continue if there are more tests to consider
	   (null? prereq-fails)
	   (null? non-completed))
     (debug:print-info 4 *default-log-port* "cond branch - "  "ei-4")
      (if  (runs:can-keep-running? hed 20)
	  (begin
	    (runs:inc-cant-run-tests hed)
	    (debug:print-info 0 *default-log-port* "no fails in prerequisites for " hed " but also none running, keeping " hed " for now. Try count: " (hash-table-ref/default *seen-cant-run-tests* hed 0) ", going to wait 60 sec.") ;; 
	    (debug:print-info 0 *default-log-port* "no fails in prerequisites (" (runs:pretty-string prereqs-not-met) ") for " hed " but also none running, keeping " hed " for now. Try count: " (hash-table-ref/default *seen-cant-run-tests* hed 0) ", going to wait 60 sec.") ;; 
	    ;; getting here likely means the system is way overloaded, kill a full minute before continuing
	    ;; (thread-sleep! 60) ;; TODO: gate by normalized server load > 1.0 (maxload config thing) CHECKTHIS!!!
	    ;; No runsdat, can't do this yet
	    ;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
	    ;;
	    (thread-sleep! 5) ;; TODO: gate by normalized server load > 1.0 (maxload config thing)
	    ;; num-retries code was here
	    ;; we use this opportunity to move contents of reg to tal
	    (list (car newtal)(append (cdr newtal) reg) '() reruns)) ;; an issue with prereqs not yet met?
	  (begin
	    (debug:print-info 1 *default-log-port* "no fails in prerequisites for " hed " but nothing seen running in a while, dropping test " hed " from the run queue")
	    (debug:print-info 1 *default-log-port* "no fails in prerequisites (" (runs:pretty-string prereqs-not-met) ") for " hed " but nothing seen running in a while, dropping test " hed " from the run queue")
	    (let ((test-id (rmt:get-test-id run-id hed "")))
	      (if test-id (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "TIMED_OUT" "Nothing seen running in a while.")))
	      (if test-id (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "TIMED_OUT" (conc "Prerequisites (" (runs:pretty-string prereqs-not-met) ") not seen running in a while."))))
            (runs:loop-values tal reg reglen regfull reruns)
            )))

    ((and 
       (or (not (null? fails))
	   (not (null? prereq-fails)))
       (member 'normal testmode))
      (debug:print-info 4 *default-log-port* "cond branch - "  "ei-5")
      (debug:print-info 1 *default-log-port* "test "  hed " (mode=" testmode ") has failed prerequisite(s); "
      (debug:print-info 1 *default-log-port* "test "  hed " (mode=" testmode ") has failed prerequisite(s): "
			(string-intersperse (map (lambda (t)(conc (db:test-get-testname t) ":" (db:test-get-state t)"/"(db:test-get-status t))) fails) ", ")
			", removing it from to-do list")
      (let ((test-id (rmt:get-test-id run-id hed "")))
	(if test-id
	    (if (not (null? prereq-fails))
		(mt:test-set-state-status-by-id-unless-completed run-id test-id "COMPLETED" "PREQ_DISCARDED" "Failed to run due to prior failed prerequisites")
		(mt:test-set-state-status-by-id-unless-completed run-id test-id "COMPLETED" "PREQ_FAIL"      "Failed to run due to failed prerequisites"))))
		(mt:test-set-state-status-by-id-unless-completed run-id test-id "COMPLETED" "PREQ_DISCARDED" (conc "Failed to run due to prior failed prerequisites: "(runs:pretty-string prereq-fails)))
		(mt:test-set-state-status-by-id-unless-completed run-id test-id "COMPLETED" "PREQ_FAIL"      (conc "Failed to run due to failed prerequisites: " (runs:pretty-string fails ))))))
      ;; (debug:print 4 *default-log-port*"BB> set PREQ_FAIL on "hed)
      ;; (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "PREQ_FAIL"      "Failed to run due to failed prerequisites")))) ;; BB: this works, btu equivalent for itemwait mode does not work.
      (if (or (not (null? reg))(not (null? tal)))
	  (begin
	    (hash-table-set! test-registry hed 'CANNOTRUN)
            (runs:loop-values tal reg reglen regfull (cons hed reruns))
            )

	 (registry-mutex         (runs:dat-registry-mutex runsdat))
	 (flags                  (runs:dat-flags runsdat))
	 (keyvals                (runs:dat-keyvals runsdat))
	 (run-info               (runs:dat-run-info runsdat))
	 (all-tests-registry     (runs:dat-all-tests-registry runsdat))
	 (run-limits-info        (runs:dat-can-run-more-tests runsdat))
	 ;; (runs:can-run-more-tests run-id jobgroup max-concurrent-jobs)) ;; look at the test jobgroup and tot jobs running
	 (have-resources         (car run-limits-info))
	 (have-resources         (and (if *journal-stats*
					  (let* ((dbfname (conc
							   (dbfile:run-id->dbnum run-id)
							   ".db"))
						 (load (tt:get-journal-stats dbfname)))
					    (if (> load 0.1) ;; dbs too busy to start more tests
						(begin
						  (debug:print-info 0 *default-log-port* "Gating launch due to db load "load" based on journal file observations for "dbfname)
						 #f)
						#t))
					  (begin
					    (debug:print-info 0 *default-log-port* "Journal gating not started for "run-id)
					    #t)) ;; if journal monitoring not started do not gate
				      (car run-limits-info)))
	 (num-running            (list-ref run-limits-info 1))
	 (num-running-in-jobgroup(list-ref run-limits-info 2)) 
	 (max-concurrent-jobs    (list-ref run-limits-info 3))
	 (job-group-limit        (list-ref run-limits-info 4))
	 ;; (prereqs-not-met        (rmt:get-prereqs-not-met run-id waitons hed item-path mode: testmode itemmaps: itemmaps))
	 ;; (prereqs-not-met         (mt:lazy-get-prereqs-not-met run-id waitons item-path mode: testmode itemmap: itemmap))
         (prereqs-running        (runs:calc-prereqs-running prereqs-not-met))
	 (fails                  (if (list? prereqs-not-met) ;; TODO: rename fails to failed-prereqs
				      (runs:calc-fails prereqs-not-met)
				      (begin
					(debug:print-error 0 *default-log-port* "prereqs-not-met is not a list! " prereqs-not-met)
					'())))
	 (non-completed           (filter (lambda (x)             ;; remove hed from not completed list, duh, of course it is not completed!
					    (not (equal? x hed)))

	   (if (and (not (common:on-homehost?))
		    maxload) ;; only gate if maxload is specified, NOTE: maxload is normalized, i.e. load=1 means all cpus fully utilized
	       (common:wait-for-normalized-load maxload "Waiting for load to drop before starting more tests" #f))
	   
	   ;; jobtools maxhomehostload is intended to prevent overloading on the homehost which can cause database corruption issues
	   (if maxhomehostload
	       (common:wait-for-homehost-load maxhomehostload
					      (conc "Waiting for homehost load to drop below normalized value of " maxhomehostload))))))
    
 
					      (conc "Waiting for homehost load to drop below normalized value of " maxhomehostload)))

	   ;; lastly lets check the servers are not overloaded by looking at threads
	   (tt:wait-on-server-load run-id *ttdat*)
	   
	   )))
    
    
    (if (and (not (null? prereqs-not-met))
	     (runs:lownoise (conc "waiting on tests " prereqs-not-met hed) 60))
	(debug:print-info 2 *default-log-port* "waiting on tests; " (string-intersperse (runs:mixed-list-testname-and-testrec->list-of-strings prereqs-not-met) ", ")))

    ;; Don't know at this time if the test have been launched at some time in the past
    ;; i.e. is this a re-launch?

	    
	    ;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
	    (thread-sleep! 5)
	    (list (car newtal)(cdr newtal) reg reruns))
	  ;; the waiton is FAIL so no point in trying to run hed ever again
	  (begin
            (let ((my-test-id (rmt:get-test-id run-id test-name item-path)))
              (mt:test-set-state-status-by-id-unless-completed run-id my-test-id "COMPLETED" "PREQ_FAIL" "Failed to run due to failed prerequisites2"))            
              (mt:test-set-state-status-by-id-unless-completed run-id my-test-id "COMPLETED" "PREQ_FAIL" (conc "Failed to run due to failed prerequisites: " (runs:pretty-string fails))))            
            (if (or (not (null? reg))(not (null? tal)))
                (if (vector? hed)
		  (begin
		    (debug:print 1 *default-log-port* "WARNING: Dropping test " test-name "/" item-path
				 " from the launch list as it has prerequistes that are FAIL")
				 " from the launch list as it has prerequistes that are FAIL: " (runs:pretty-string fails))
		    (let ((test-id (rmt:get-test-id run-id hed "")))
		      (if test-id (mt:test-set-state-status-by-id-unless-completed run-id test-id "COMPLETED" "PREQ_FAIL" "Failed to run due to failed prerequisites")))
		      (if test-id (mt:test-set-state-status-by-id-unless-completed run-id test-id "COMPLETED" "PREQ_FAIL" (conc "Failed to run due to failed prerequisites: " (runs:pretty-string fails)) )))
		    (runs:shrink-can-run-more-tests-count runsdat) ;; DELAY TWEAKER (still needed?)
		    ;; (thread-sleep! *global-delta*)
		    ;; This next is for the items

                    (if (not (null? fails))
                        ;;(mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "PREQ_FAIL" #f)
                        (rmt:set-state-status-and-roll-up-items run-id test-name item-path "NOT_STARTED" "PREQ_FAIL" #f) 
                        (rmt:set-state-status-and-roll-up-items run-id test-name item-path "NOT_STARTED" "PREQ_FAIL" (runs:pretty-string fails)) 
                        ;;(mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "BLOCKED" #f)
                        (rmt:set-state-status-and-roll-up-items run-id test-name item-path "NOT_STARTED" "BLOCKED" #f) )
		    (hash-table-set! test-registry (db:test-make-full-name test-name item-path) 'removed)
		    (runs:loop-values tal reg reglen regfull reruns))
		  (let ((nth-try (hash-table-ref/default test-registry hed 0))) ;; hed not a vector...
                    (debug:print 2 *default-log-port* "nth-try("hed")="nth-try)
		    (cond
		     ((member "RUNNING" (map db:test-get-state prereqs-not-met))
		      (if (runs:lownoise (conc "possible RUNNING prerequistes " hed) 60)
			  (debug:print 0 *default-log-port* "WARNING: test " hed " has possible RUNNING prerequisites, don't give up on it yet."))
		     ;;((member "RUNNING" (map db:test-get-state prereqs-not-met))
                      ((> 0 (length prereqs-running))
		      (if (runs:lownoise (conc "possible RUNNING prerequisites " hed) 60)
			  (debug:print 0 *default-log-port* "WARNING: test " hed " has possible RUNNING prerequisites: " prereqs-running ", don't give up on it yet."))
		      (thread-sleep! 0.1)
		      (runs:loop-values tal reg reglen regfull reruns))
		     ((or (not nth-try) ;; BB: condition on subsequent tries, condition below fires on first try 
			  (and (number? nth-try)
			       (< nth-try 2)))
		      (hash-table-set! test-registry hed (if (number? nth-try)
							     (+ nth-try 1)
							     0))
		      (if (runs:lownoise (conc "not removing test " hed) 60)
			  (debug:print 1 *default-log-port* "WARNING: not removing test " hed " from queue although it may not be runnable due to FAILED prerequisites"))
			  (debug:print 1 *default-log-port* "WARNING: not removing test " hed " from queue although it may not be runnable due to FAILED prerequisites: "(runs:pretty-string fails)))
		      ;; may not have processed correctly. Could be a race condition in your test implementation? Dropping test " hed) ;;  " as it has prerequistes that are FAIL. (NOTE: hed is not a vector)")
		      (runs:shrink-can-run-more-tests-count runsdat) ;; DELAY TWEAKER (still needed?)
		      (runs:loop-values newtal reg reglen regfull reruns))
		     ((symbol? nth-try) ;; BB: 'done matches here in one case where prereq itemwait failed.  This is first "try"
		      (if (eq? nth-try 'removed) ;; removed is removed - drop it NOW
			  (if (null? tal)
			      #f ;; yes, really
			      (list (car tal)(cdr tal) reg reruns))
			  (begin
			    (if (runs:lownoise (conc "FAILED prerequisites or other issue" hed) 60)
				(debug:print 0 *default-log-port* "WARNING: test " hed " has FAILED prerequisites or other issue. Internal state >" nth-try "< will be overridden and we'll retry."))
			    ;; was: (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "KEEP_TRYING" #f)
                            (mt:test-set-state-status-by-testname-unless-completed run-id test-name item-path "COMPLETED" "PREQ_FAIL" #f)
				(debug:print 0 *default-log-port* "WARNING: test " hed " has FAILED prerequisites (" (runs:pretty-string fails)") or other issue. Internal state >" nth-try "< will be overridden and we'll retry."))
			    (let* ((test-id      (rmt:get-test-id run-id hed item-path))
				   (test-info    (rmt:get-testinfo-state-status run-id test-id)) ;; we need *current* info
				   (status       (db:test-status test-info)))
			      (if (equal? status "KEEP_TRYING")
				  (mt:test-set-state-status-by-testname-unless-completed run-id test-name item-path "COMPLETED" "PREQ_FAIL" #f)
				  (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "KEEP_TRYING" #f)))
			    (hash-table-set! test-registry hed 'removed) ;; was 0
                            (if (not (and (null? reg) (null? tal)))
                                (runs:loop-values tal reg reglen regfull reruns)
                                #f))))
		     (else
		      (if (runs:lownoise (conc "FAILED prerequitests and we tried" hed) 60)
			  (debug:print 0 *default-log-port* "WARNING: test " hed " has FAILED prerequitests and we've tried at least 10 times to run it. Giving up now."))
			  (debug:print 0 *default-log-port* "WARNING: test " hed " has FAILED prerequitests: " (runs:pretty-string fails) "  and we've tried at least 10 times to run it. Giving up now."))
		      ;; (debug:print 0 *default-log-port* "         prereqs: " prereqs-not-met)
		      (hash-table-set! test-registry hed 'removed)
		      (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "TEN_STRIKES" #f)
		      ;; I'm unclear on if this roll up is needed - it may be the root cause of the "all set to FAIL" bug.
		      (rmt:set-state-status-and-roll-up-items run-id test-name item-path #f "FAIL" #f) ;; treat as FAIL
		      (list (if (null? tal)(car newtal)(car tal))
			    tal

   (lambda (t)
     (or (not (vector? t))
	 (and (equal? "NOT_STARTED" (db:test-get-state t))
	      (member (db:test-get-status t)
		      '("n/a" "KEEP_TRYING")))
	 (and (equal? "RUNNING" (db:test-get-state t))))) ;; account for a test that is running
   prereqs-not-met))

(define (runs:calc-prereqs-running prereqs-not-met)
  (if (list? prereqs-not-met)
    (filter 
     (lambda (t)
       (or (not (vector? t))
         (member (db:test-get-state t) '("RUNNING" "LAUNCHED" "REMOTE_HOST_START"))
       ))
     prereqs-not-met)
    '()
  )
)

(define (runs:pretty-string lst)
  (map (lambda (t)
	 (if (not (vector? t))
	     (conc t)
	     (conc (db:test-get-testname t) ":" (db:test-get-state t) "/" (db:test-get-status t))))
       lst))

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

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

	  extras
	  hostinfo

	  ports
	  posix
	  files
	  data-structures
	  directory-utils
	  tcp
	  ))
 (chicken-5
  (import chicken.base
	  chicken.condition
	  chicken.file
	  chicken.pathname

;; (max-connections 4096)

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

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

	;; no conn

	;; find server with lowest number of threads running (i.e. lowest load)
	;;
        (let* ((sdats (tt:get-server-info-sorted ttdat dbfname))
	       (sdat  (if (null? sdats)
			  #f
			  ;; choose server with lowest threads count
			  (car sdats))))
	   (debug:print-info 2 *default-log-port* "found sdat " sdat)
           (match sdat
			  (car (sort sdats
				     (lambda (a b)
				       (let* ((load-a (tt:get-server-threads a))
					      (load-b (tt:get-server-threads b)))
					 (< load-a load-b))))))))
				     
	  ;; (let ((indx (max (random (- (length sdats) 1)) 0)))
	  ;;    (list-ref sdats indx)))))
	  ;; (debug:print-info 1 *default-log-port* "found sdat " sdat" from sdats: "sdats)
          (match sdat
	    ((host port start-time server-id pid dbfname2 servinffile)
	     (assert (equal? dbfname dbfname2) "FATAL: read server info from wrong file.")
             (debug:print-info 2 *default-log-port* "no conn - in match servinffile:" servinffile)
	     (let* ((host-port (conc host":"port))
		    (conn (make-tt-conn
			   host: host
			   port: port

	     (tt:client-connect-to-server ttdat dbfname run-id testsuite server-start-proc)))))))

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

;; host:port => ( meta . when-updated)
(define *server-load* (make-hash-table))

(define (tt:save-server-meta host port meta)
  (hash-table-set! *server-load* (conc host":"port) (cons meta (current-seconds))))

(define (tt:get-server-threads dat)
  (let* ((host (car  dat))
	 (port (cadr dat))
	 (dat  (tt:get-server-meta host port #t)))
    ;; (debug:print 0 *default-log-port* "host: "host" port: "port" dat: "dat)
    (if (list? dat)
	(or (alist-ref 'sload dat) 99998)
	99999))) ;; absurd number means don't use this one

;; lazy get, does not auto-refresh meta, this might be a problem
;;
(define (tt:get-server-meta host port #!optional (do-ping #f))
  (let* ((get-meta (lambda ()
		     (let* ((dat  (hash-table-ref/default *server-load* (conc host":"port) #f)))
		       (if dat (car dat) #f))))
	 (meta     (get-meta)))
    (if (and (not meta)
	     do-ping)
	(begin
	  (tt:timed-ping host port #f)
	  (get-meta))
	meta)))

(define (tt:wait-on-server-load run-id ttdat)
  (if ttdat ;; if no server yet just pass on through
      (let* ((dbfname                 (dbmod:run-id->dbfname run-id))
	     (get-lowest-thread-load
	      (lambda ()
		(let* ((sdats (tt:get-server-info-sorted ttdat dbfname)))
		  (car (map tt:get-server-threads sdats))))))
	(if ttdat
	    (let loop ((count 0))
	      (let* ((lowestload (get-lowest-thread-load)))
		(if (> lowestload 5) ;; load is pretty high
		    (begin
		      (debug:print 0 *default-log-port* "Servers appear overloaded with "lowestload" threads, waiting...")
		      (thread-sleep! 1)
		      (if (< count 10)
			  (loop (+ count 1)))))))
	    (debug:print 0 *default-log-port* "Can't wait on server load, *ttdat* not set")))))

(define (tt:ping host port server-id #!optional (tries-left 5))
  (let*  ((res      (tt:send-receive-direct host port `(ping #f #f #f) ping-mode: #t)) ;; please send me your server-id
	  (try-again (lambda ()
		       (if (> tries-left 0)
			   (begin
			     (thread-sleep! 1)
			     (tt:ping host port server-id (- tries-left 1)))
			   #f))))
    ;;
    ;; need two threads, one a 5 second timer
    ;;
    (match res
      ((status errmsg result meta)
       (tt:save-server-meta host port meta)
       (if (equal? result server-id)
	   (let* ((server-state (alist-ref 'sstate meta)))
	     ;; (debug:print 0 *default-log-port* "Ping to "host":"port" successful.")
	     (or server-state 'unk)) ;; then we are good
	   (begin
	     (if server-id
	     (debug:print 0 *default-log-port* "WARNING: server-id does not match, expected: "server-id", got: "result)
		 (debug:print 0 *default-log-port* "WARNING: server-id does not match, expected: "server-id", got: "result))
	     #f)))
      (else
       ;; (debug:print 0 *default-log-port* "res not in form (status errmsg result meta), got: "res)
       (try-again)))))

;; client side handler
;;
;;(tt:handler #<tt> get-keys #f () 2 #f "/home/matt/data/megatest/ext-tests" #f "main.db" "ext-tests" "/home/matt/data/megatest/bin/.22.04/../megatest")
;;
;;g
(define (tt:handler ttdat cmd run-id params attemptnum readonly-mode dbfname testsuite mtexe server-start-proc)
  ;; connect-to-server will start a server if needed.
  (let* ((areapath (tt-areapath ttdat))
	 (conn     (tt:client-connect-to-server ttdat dbfname run-id testsuite server-start-proc))) ;; looks up conn keyed by dbfname
    (if conn
	;; have connection, call the server
	(let* ((res (tt:send-receive ttdat conn cmd run-id params)))

			 (debug:print 0 *default-log-port* "Server is loaded, delaying "delay-wait" seconds")
			 (thread-sleep! delay-wait)))))
	     (case status
	       ((busy) ;; result will be how long the server wants you to delay
		(let* ((raw-dly  (if (number? result) result 0.1))
		       (dly      (+ raw-dly (/ attemptnum 10)))) ;; (* raw-dly (/ attemptnum 2))))
		  (debug:print 0 *default-log-port* "WARNING: server for "dbfname" is busy, cmd is "cmd", will try again in "dly" seconds. This is attempt "(- attemptnum 1))
                  (debug:print 0 *default-log-port* errmsg)
		  (thread-sleep! dly)
		  (tt:handler  ttdat cmd run-id params (+ attemptnum 1) readonly-mode dbfname testsuite mtexe server-start-proc)))
	       ((loaded)
		(debug:print 0 *default-log-port* "WARNING: server for "dbfname" is loaded, slowing queries.")
		(tt:backoff-incr (tt-conn-host conn)(tt-conn-port conn))

		;; this would be a good place to force reconnection and connect to a different server
		
		result) ;; (tt:handler  ttdat cmd run-id params (+ attemptnum 1) readonly-mode dbfname testsuite mtexe))
	       (else
		result)))
	    (else ;; did not receive properly formated result
	     (if (not res) ;; tt:send-receive telling us that communication failed
		 (let* ((host    (tt-conn-host conn))
			(port    (tt-conn-port conn))

		   ;; (tt:handler ttdat cmd run-id params (+ attemptnum 1) readonly-mode dbfname testsuite mtexe)
		   )))))
	(begin
	  (thread-sleep! 1) ;; no conn yet set up, give it a rest and try again
	  (tt:handler ttdat cmd run-id params attemptnum readonly-mode dbfname testsuite mtexe server-start-proc)))))

;; gets server info and appends path to server file
;; sorts by age, oldest first
;; sorts by age, --oldest-- now newest first
;;
;; move the ping here?
;;
;; returns list of (host port startseconds server-id servinfofile)
;;
(define (tt:get-server-info-sorted ttdat dbfname)
  (let* ((areapath (tt-areapath ttdat))
	 (sfiles   (tt:find-server areapath dbfname))
	 (sdats    (filter car (map tt:server-get-info sfiles))) ;; first element is #f if the file disappeared while being read
	 (sorted   (sort sdats (lambda (a b)
				 (let* ((starta (list-ref a 2))
					(startb (list-ref b 2)))
				   (if (eq? starta startb)
				       (string>? (list-ref a 3)(list-ref b 3)) ;; if servers started at same time look at server-id
				       (< starta startb))))))
				       (> starta startb))))))
	 (count    0))
    (for-each
     (lambda (rec)
       (if (or (> (length sorted) 1)
	       (common:low-noise-print 120 "server info sorted"))
	   (debug:print 2 *default-log-port* "SERVER #"count": "(string-intersperse (map conc sorted) ", ")))
       (set! count (+ count 1)))

;;======================================================================
;; server
;;======================================================================

(define (tt:sync-dbs ttdat)
  #f)

(define *server-start-requests* '())

;; start the listener and start responding to requests
;;
;; NOTE: organise by dbfname, not run-id so we don't need
;;       to pull in more modules
;;
;; This is the routine called in megatest.scm to start a server. NOTE: sequence is different for main.db vs. X.db

				     (same-host (or (not prime-host) ;; i.e. this is the first host
						    (equal? prime-host host)))
				     (keep-srv  (and good-ping same-host)))
				(if keep-srv	
				    (loop (cdr servrs)
					  host
					  (cons servdat result))
				    (begin
                                      ;; (debug:print-info 0 *default-log-port* "good-ping: " good-ping " same-host: " same-host "keep-srv: " keep-srv)
				      (handle-exceptions
				       exn
				       (debug:print-info 0 *default-log-port* "Error removing server info file: "servinfofile", "
							 (condition->list exn))
				       (delete-file* servinfofile))
				      (loop (cdr servrs) prime-host result)))))
				    (let* ((modtime (file-modification-time servinfofile)))
				      ;; if the .servinfo hasn't been touched in five min
				      ;; we can be pretty sure the server is truly dead
				      (if (> (- (current-seconds) modtime) 360)
					  (handle-exceptions
					   exn
					   (debug:print-info 0 *default-log-port*
							     "Error removing server info file: "servinfofile", "
							     (condition->list exn))
					   (delete-file* servinfofile))
					  (loop (cdr servrs) prime-host result))))))
			     (else
			      ;; can't delete it as we don't have a filename. NOTE: Should really never get here.
			      ;; can't delete it as we don't have a filename. NOTE: Should never get here.
			      (debug:print-info 0 *default-log-port* "ERROR: bad servinfo record \""servdat"\"")
			      (loop (cdr servrs) prime-host result)) ;; drop 
			     )))))
	       (home-host (if (null? good-srvrs)
			      #f
			      (caar good-srvrs))))
	  ;; by here we have a trustworthy list of servers and we have removed the .servinfo file for any unresponsive servers
	  ;; and the list is in good-srvrs
	  ;;
	  (cond
	   ((not home-host) ;; no servers yet, go ahead and start
	    (debug:print-info 0 *default-log-port* "No servers yet, starting on "(get-host-name)))
	   ((> (length good-srvrs) 2) ;; don't need more, just exit
	   ((> (length good-srvrs) 3) ;; don't need more, just exit
	    (debug:print-info 0 *default-log-port* "Have "(length good-srvrs)", no need for more, exiting.")
	    (exit))
	   ((not (equal? home-host (get-host-name))) ;; there is a home-host and we are not on it
	    (debug:print-info 0 *default-log-port* "Prime main server is on host "home-host", but we are on host "(get-host-name)", exiting.")
	    (exit))
	   (else
	    (debug:print-info 0 *default-log-port* "Starting on host "(get-host-name)", along with "(length good-srvrs)" other servers.")))

		      (thread-sleep! 0.25)
		      (loop (+ count 1)))
		    (begin
		      (debug:print 0 *default-log-port* "ERROR: (tt-port ttdat) no port set! Exiting.")
		      (exit)))))
	  
	  ;; create a servinfo file start keep-running
	  ;; On WSL there seems to be a race condition where the .servinfo file
	  ;; is not created fast enough
          (debug:print 0 *default-log-port* "Creating servinfo file for " dbfname)
	  (tt:create-server-registration-file ttdat dbfname)
	  (procinf-status-set! *procinf* "running")
	  (tt-state-set! ttdat 'running)
	  (dbfile:with-no-sync-db
	   nosyncdbpath
	   (lambda (nsdb)

	   nosyncdbpath
	   (lambda (nsdb)
	     (dbfile:insert-or-update-process nsdb *procinf*)))
	  (debug:print 0 *default-log-port* "Exiting now.")
	  (exit))))))

(define (tt:keep-running ttdat dbfname dbstruct)

  (thread-sleep! 1)
  
  ;; at this point the server is running and responding to calls, we just monitor
  ;; for db calls and exit if there are none.

  ;; if I am not in the first 3 servers, exit
  (let* ((start-time (current-seconds)))
    (let loop ()
      (let* ((servers   (tt:get-server-info-sorted ttdat dbfname))
	     (home-host (if (null? servers)
			    #f
			    (caar servers)))
	     (my-index  (list-index (lambda (x)
				      (equal? (list-ref x 6)
					      (tt-servinf-file ttdat)))
				    servers))
	     (ok         (cond
			  ((not my-index)
			   (debug:print 0 *default-log-port* "WARNING: Apparently I don't exist.")
			   #f) ;; keep trying or give up?
			  ((not *server-run*)
			   (debug:print 0 *default-log-port* "WARNING: received a stop server from client by remote request.")
			   #f)
			  ((null? servers)
			   (debug:print 0 *default-log-port* "WARNING: no servinfo files found, this cannot be.")
			   #f) ;; not ok
			  ((> my-index 2)
			   (debug:print 0 *default-log-port* "WARNING: there are more than two servers ahead of me, I'm not needed, exiting.")
			  ((> my-index 3)
			   (debug:print 0 *default-log-port* "WARNING: there are more than three servers ahead of me, I'm not needed, exiting.")
			   #f) ;; not ok to not be in first three
			  ((eq? (tt-state ttdat) 'running) #t) ;; we are good to keep going
			  ((> (- (current-seconds) start-time) 30)
			   (debug:print 0 *default-log-port* "WARNING: over 30 seconds and not yet in runnning mode. Exiting.")
			   #f)
			  (else #t))))
	(if ok

	      (let* ((sinfo-file (tt-servinf-file ttdat)))
		;; (debug:print 0 *default-log-port* "INFO: touching "sinfo-file)
		(set! (file-modification-time sinfo-file) (current-seconds))
		((dbr:dbstruct-sync-proc dbstruct) last-update)
		(dbr:dbstruct-last-update-set! dbstruct curr-secs))))
	
	(if (< (- (current-seconds) (tt-last-access ttdat)) (tt-server-timeout-param))
	    ;; process any requests to start a new server due to load on this one
	    (let* ((requests *server-start-requests*))
	      (set! *server-start-requests* '())
	      (if (> (length requests) 0)
		  (debug:print-info 0 *default-log-port* "Processing "(length requests)" server start requests"))
	      (for-each (lambda (proc)
	    (begin
			  (proc)
			  (thread-sleep! 1))
			requests)
	      (thread-sleep! 5)
	      (loop)))))
    (tt:shutdown-server ttdat)
    (debug:print 0 *default-log-port* "INFO: Server timed out, exiting from tt:keep-running.")))


(define (tt:shutdown-server ttdat)

	 (servinf (conc servdir"/"host":"port"-"(current-process-id)":"dbfname))
	 (serv-id (tt:mk-signature areapath)))
    (assert (and host port) "FATAL: tt:create-server-registration-file called with no conn, dbfname="dbfname)
    (tt-servinf-file-set! ttdat servinf)
    (with-output-to-file servinf
      (lambda ()
	(print "SERVER STARTED: "host":"port" AT "(current-seconds)" server-id: "serv-id" pid: "(current-process-id)" dbfname: "dbfname)))
    (let loop ((count 0))
      (if (not (file-exists? servinf))
	  (begin
	    (debug:print 0 *default-log-port* "WARNING: file "servinf" was created but it doesn't show up on disk! We'll try again.")
	    (thread-sleep! 1)
	    (if (> count 10)
		(debug:print 0 *default-log-port* "WARNING: file "servinf" was not created.")
		(loop (+ count 1))))))
      serv-id))
    serv-id))

;; find valid server
;; get servers listed, last part of name must match :<dbfname>
;; if more than one, wait one second and look again
;; 
;; NOTE: this only gets the servinfo data, no network activity here
;; future: ping oldest, if alive remove other :<dbfname> files
;;       i.e. no ping etc.
;;
(define (tt:find-server areapath dbfname)
  (let* ((servdir  (tt:get-servinfo-dir areapath))
	 (sfiles   (glob (conc servdir"/*:"dbfname)))
	 (goodfiles '()))

    ;; filter the files here by looking in processes table (if we are not main.db)

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

(define (tt:too-recent-server-start dbfname)
  (let* ((last-run-time (hash-table-ref/default *last-server-start* dbfname #f)))
    (and last-run-time
	 (< (- (current-seconds) last-run-time) 5))))

(define *last-server-start-request-time* 0)
    
;; Given an area path,  start a server process    ### NOTE ### > file 2>&1 
;; if the target-host is set 
;; try running on that host
;;   incidental: rotate logs in logs/ dir.
;;
(define  (tt:server-process-run areapath testsuite mtexe run-id #!key (profile-mode "")) ;; areapath is *toppath* for a given testsuite area
  (assert areapath  "FATAL: tt:server-process-run called without areapath defined.")
  (assert testsuite "FATAL: tt:server-process-run called without testsuite defined.")
  (assert mtexe     "FATAL: tt:server-process-run called without mtexe defined.")
  ;; mtest -server - -m testsuite:ext-tests -db 6.db
  (let* ((dbfname  (dbmod:run-id->dbfname run-id)))
    (if (or (< (- (current-seconds) *last-server-start-request-time*) 5) ;; attempted start less than 5 sec ago
    (if (tt:too-recent-server-start dbfname)
	    (tt:too-recent-server-start dbfname))
	#f
	(let* ((load     (get-normalized-cpu-load))
	       (srvrs    (tt:find-server areapath dbfname))
	       (trying   (length srvrs))
	       (nrun     (number-of-processes-running (conc "mtest.*server.*"testsuite".*"dbfname))))
	  (set! *last-server-start-request-time* (current-seconds))
	  (cond
	   ((> load 2.0)
	    (debug:print 0 *default-log-port* "Normalized load "load" on " (get-host-name) " is over the limit of 2.0. Not starting a server. Please reduce the load on "(get-host-name)" by killing some processes")
	    (thread-sleep! 1)
	   ((> load 10.0)
	    (debug:print 0 *default-log-port* "Normalized load " load " over 10, (load: " (commonmod:get-cpu-load) " cores: " (get-current-host-cores) " exiting...")
	    (thread-sleep! 1) ;; I'm not convinced that a delay here is helpful. -mrw-
	    #f)
	   ((> nrun 100)
	    (debug:print 0 *default-log-port* nrun" servers running on " (get-host-name) ", not starting another.")
	    (thread-sleep! 1)
	    #f)
	   ((> trying 2)
	   ((> trying 3)
	    (debug:print 0 *default-log-port* trying" servers registered in .servinfo dir. not starting another.")
	    (thread-sleep! 1)
	    #f)
	   (else
	    (if (not (file-exists? (conc areapath"/logs")))
		(create-directory (conc areapath"/logs") #t))
	    (let* ((logfile   (conc areapath "/logs/server-"dbfname"-"(current-process-id)".log")) ;; -" curr-pid "-" target-host ".log"))

  (sort (get-all-ips) ip-pref-less?))

(define (get-all-ips)
  (map address-info-host
       (filter (lambda (x)
		 (equal? (address-info-type x) "tcp"))
	       (address-infos (get-host-name)))))

;;======================================================================
;; Other Utils
;;======================================================================
)

(defstruct jstats
  (count 0)
  (jcount (make-hash-table)) ;; 1.db => journal_count
  )

;; timeblk => jstats
(define *journal-stats* #f) ;; (make-hash-table))
(define *journal-stats-enable* #t) ;; change to #f to turn off

;; monte-carlo-esque random sampling of journal files
;; for all the files:
;;   if .journal
;;      update stats +1 +1
;;      update stats +1  0
;;
(define (tt:write-load-tracking dbdir)
  (if *journal-stats-enable*
      (let* ((cs    (current-seconds))
	     (key   (inexact->exact (quotient cs 10)))
	     (old   (- key 5)) ;; 4 x 10 seconds ago
	     (jstat (if (hash-table-exists? *journal-stats* key)
			(hash-table-ref *journal-stats* key )
			(let ((new (make-jstats)))
			  (hash-table-set! *journal-stats* key new)
			  new))))
	;; clear out old records
	(for-each
	 (lambda (key)
	   (if (< key old)
	       (hash-table-delete! *journal-stats* key)))
	 (hash-table-keys *journal-stats*))

	;; increment our count of observations
	(jstats-count-set! jstat (+ (jstats-count jstat) 1))
	
	;; now find and increment journal file counts
	(directory-fold
	 (lambda (fname res)
	   ;; is it a journal file?
	   (let ((parts (string-match "^(.*\\.db)-journal.*" fname)))
	     (match parts
		    ((_ dbfname)
		     (hash-table-set! (jstats-jcount jstat) dbfname
				      (+ (hash-table-ref/default (jstats-jcount jstat) dbfname 0) 1.0)
				      ))
		    (else #f)
		    )))
	 '()
	 dbdir 
     ))))

(define *journal-stats-mutex* (make-mutex))

(define (tt:journal-stats-run dbdir)
  (if (not *journal-stats*)(set! *journal-stats* (make-hash-table)))
  (let loop ()
    (mutex-lock! *journal-stats-mutex*)
    (tt:write-load-tracking dbdir)
    (mutex-unlock! *journal-stats-mutex*)
    (thread-sleep! (/ (random 1000) 100.0))
    (loop)))

;; call this to start a thread that is keeping the journal-stats up to date.
(define (tt:start-stats dbdir)
  
  (thread-start!
   (make-thread
    (lambda ()(tt:journal-stats-run dbdir)) "Journal stats collection thread")))

(define (tt:get-journal-stats #!optional (dbfname #f))
  (let* ((result    (make-jstats))
	 (hitcounts (jstats-jcount result)))
    (if (and *journal-stats*
	     *journal-stats-enable*)
	(begin
	  (mutex-lock! *journal-stats-mutex*)
	  (hash-table-for-each
	   *journal-stats*
	   (lambda (k v) ;; key jstats
	     (let* ((count  (jstats-count v))
		    (jcount (jstats-jcount v))) ;; dbfname => hit count
	       (jstats-count-set! result
				  (+ (jstats-count result)
				     (jstats-count v)))
	       (hash-table-for-each
		jcount
		(lambda (dbfname hit-count)
		  (hash-table-set! hitcounts dbfname
				   (+ hit-count
				      (hash-table-ref/default hitcounts dbfname 0))))))))
	  (mutex-unlock! *journal-stats-mutex*))
	(debug:print 0 *default-log-port* "INFO: *journal-stats* not set."))
    ;; convert to normalized alist
    (let* ((tot  (max (jstats-count result) 1)) ;; avoid divide by zero
	   (hits (jstats-jcount result)) ;; 1.db => count
	   (res  (hash-table-map
		  hits
		  (lambda (fname hitcount)
		    (cons fname (/ hitcount tot))))))
      (if dbfname
	  (or (alist-ref dbfname res equal?) 0)
	  res))))

;; megatest> (import tcp-transportmod)
;; megatest> (tt:write-load-tracking ".mtdb")
;; megatest> (hash-table-keys *journal-stats*)
;; (172060297)
;; megatest> (jstats->alist (hash-table-ref *journal-stats* 172060297))
;; ((count . 1) (jcount . #<hash-table (1)>))
;; megatest> (jstats-jcount (hash-table-ref *journal-stats* 172060297))
;; #<hash-table (1)>
;; megatest> (hash-table->alist (jstats-jcount (hash-table-ref *journal-stats* 172060297)))
;; (("1.db" . 4))

)