Megatest

Check-in [1b9511af26]
Login
Overview
Comment:ok, new syncer works, but -run-wait is not waiting.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | v1.65-tweak-bfsync
Files: files | file ages | folders
SHA1: 1b9511af2613357b5ebbe3d52977a1318d2a5ac6
User & Date: bjbarcla on 2019-03-15 18:38:41
Other Links: branch diff | manifest | tags
Context
2019-03-15
20:32
fix fork to background for syncer check-in: b90b2aeb10 user: bjbarcla tags: v1.65-tweak-bfsync
18:38
ok, new syncer works, but -run-wait is not waiting. check-in: 1b9511af26 user: bjbarcla tags: v1.65-tweak-bfsync
2019-03-14
18:28
wip check-in: 89d013c708 user: bjbarcla tags: v1.65-tweak-bfsync
Changes

Modified common.scm from [8e98e8852f] to [b1d85b703a].

316
317
318
319
320
321
322
323






324
325
326
327
328
329
330
(define (common:version-changed?)
  (not (equal? (common:get-last-run-version)
               (common:version-signature))))

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






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







|
>
>
>
>
>
>







316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
(define (common:version-changed?)
  (not (equal? (common:get-last-run-version)
               (common:version-signature))))

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


(define (common:get-sync-lock-filepath)
  (let* ((tmp-area     (common:get-db-tmp-area))
         (lockfile     (conc tmp-area "/megatest.db.sync-lock")))
    lockfile))
    
;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma
;;
(define (common:cleanup-db dbstruct #!key (full #f))
  (apply db:multi-db-sync 
   dbstruct
   'schema
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622


623
624
625
626
627
628
629
   (read (open-input-string (z3:decode-buffer (base64:base64-decode instr))))))

;; dot-locking egg seems not to work, using this for now
;; if lock is older than expire-time then remove it and try again
;; to get the lock
;;
(define (common:simple-file-lock fname #!key (expire-time 300))
  (handle-exceptions
      exn
      #f ;; don't really care what went wrong right now. NOTE: I have not seen this one actually fail.
    (if (common:file-exists? fname)
	(if (> (- (current-seconds)(file-modification-time fname)) expire-time)
	    (begin
	      (delete-file* fname)
	      (common:simple-file-lock fname expire-time: expire-time))
	    #f)
	(let ((key-string (conc (get-host-name) "-" (current-process-id))))
	  (with-output-to-file fname
	    (lambda ()
	      (print key-string)))
	  (thread-sleep! 0.25)
	  (if (common:file-exists? fname)
	      (with-input-from-file fname
		(lambda ()
		  (equal? key-string (read-line))))
	      #f)))))



(define (common:simple-file-lock-and-wait fname #!key (expire-time 300))
  (let ((end-time (+ expire-time (current-seconds))))
    (let loop ((got-lock (common:simple-file-lock fname expire-time: expire-time)))
      (if got-lock
	  #t
	  (if (> end-time (current-seconds))







|
|
|















|
>
>







603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
   (read (open-input-string (z3:decode-buffer (base64:base64-decode instr))))))

;; dot-locking egg seems not to work, using this for now
;; if lock is older than expire-time then remove it and try again
;; to get the lock
;;
(define (common:simple-file-lock fname #!key (expire-time 300))
;  (handle-exceptions
;      exn
;      #f ;; don't really care what went wrong right now. NOTE: I have not seen this one actually fail.
    (if (common:file-exists? fname)
	(if (> (- (current-seconds)(file-modification-time fname)) expire-time)
	    (begin
	      (delete-file* fname)
	      (common:simple-file-lock fname expire-time: expire-time))
	    #f)
	(let ((key-string (conc (get-host-name) "-" (current-process-id))))
	  (with-output-to-file fname
	    (lambda ()
	      (print key-string)))
	  (thread-sleep! 0.25)
	  (if (common:file-exists? fname)
	      (with-input-from-file fname
		(lambda ()
		  (equal? key-string (read-line))))
	      #f)))
;    )
  )

(define (common:simple-file-lock-and-wait fname #!key (expire-time 300))
  (let ((end-time (+ expire-time (current-seconds))))
    (let loop ((got-lock (common:simple-file-lock fname expire-time: expire-time)))
      (if got-lock
	  #t
	  (if (> end-time (current-seconds))

Modified db.scm from [b2f54e4e77] to [597e7ddad8].

1047
1048
1049
1050
1051
1052
1053
1054




1055
1056
1057
1058
1059
1060
1061
	 ;; kill servers
	 ((killservers)
	  (for-each
	   (lambda (server)
	     (match-let (((mod-time host port start-time pid) server))
	       (if (and host pid)
		   (tasks:kill-server host pid))))
	   servers))




	 
	 ;; clear out junk records
	 ;;
	 ((dejunk)
	  (db:delay-if-busy mtdb) ;; ok to delay on mtdb
	  (db:clean-up mtdb)
	  (db:clean-up tmpdb)







|
>
>
>
>







1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
	 ;; kill servers
	 ((killservers)
	  (for-each
	   (lambda (server)
	     (match-let (((mod-time host port start-time pid) server))
	       (if (and host pid)
		   (tasks:kill-server host pid))))
	   servers)

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

Modified megatest.scm from [8e2abc3809] to [8964e71ae0].

424
425
426
427
428
429
430

431
432
433
434
435
436
437
			"-create-megatest-area"
			"-mark-incompletes"

			"-convert-to-norm"
			"-convert-to-old"
			"-import-megatest.db"
			"-sync-to-megatest.db"

			"-logging"
			"-v" ;; verbose 2, more than normal (normal is 1)
			"-q" ;; quiet 0, errors/warnings only

                        "-diff-rep"
                        )
		 args:arg-hash







>







424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
			"-create-megatest-area"
			"-mark-incompletes"

			"-convert-to-norm"
			"-convert-to-old"
			"-import-megatest.db"
			"-sync-to-megatest.db"
                        "-sync-brute-force"
			"-logging"
			"-v" ;; verbose 2, more than normal (normal is 1)
			"-q" ;; quiet 0, errors/warnings only

                        "-diff-rep"
                        )
		 args:arg-hash
2294
2295
2296
2297
2298
2299
2300




2301
2302
2303
2304
2305
2306
2307
       'killservers
       'dejunk
       'adj-testids
       'old2new
       ;; 'new2old
       )
      (set! *didsomething* #t)))





(if (args:get-arg "-sync-to-megatest.db")
    (let* ((dbstruct (db:setup #f))
	   (tmpdbpth (cdr (dbr:dbstruct-tmpdb dbstruct)))
	   (lockfile (conc tmpdbpth ".lock"))
	   (locked   (common:simple-file-lock lockfile)) 
	   (res      (if locked







>
>
>
>







2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
       'killservers
       'dejunk
       'adj-testids
       'old2new
       ;; 'new2old
       )
      (set! *didsomething* #t)))

(when (args:get-arg "-sync-brute-force")
  ((server:get-bruteforce-syncer (db:setup #t) persist-until-sync: #t))
  (set! *didsomething* #t))

(if (args:get-arg "-sync-to-megatest.db")
    (let* ((dbstruct (db:setup #f))
	   (tmpdbpth (cdr (dbr:dbstruct-tmpdb dbstruct)))
	   (lockfile (conc tmpdbpth ".lock"))
	   (locked   (common:simple-file-lock lockfile)) 
	   (res      (if locked

Modified server.scm from [98cdb6b75b] to [219dea7dfd].

499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517

518
519
520
521
522
523
524
525
526



527
528

529


530



531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554

555
556
557
558
559
560
561
562



563

564






565
566
567




568
569
570
571
572
573

574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822

823
824
825


826
;;       (server:release-sync-lock)
;;       (server:have-sync-lock?))
;;      (else #f))))

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

(define (server:get-bruteforce-syncer dbstruct #!key (fork-to-background #f))
  (let* ((sqlite-exe   (or (get-environment-variable "MT_SQLITE3_EXE"))) ;; defined in cfg.sh
         (sync-log     (or (args:get-arg "-sync-log") (conc *toppath* "/logs/sync-" (current-process-id) "-" (get-host-name) ".log")))
	 (tmp-area     (common:get-db-tmp-area))
	 (tmp-db       (conc tmp-area "/megatest.db"))
	 (staging-file (conc *toppath* "/.megatest.db"))
	 (mtdbfile     (conc *toppath* "/megatest.db"))
	 (lockfile     (conc tmp-db ".lock"))
         (sync-cmd-core     (conc sqlite-exe" " tmp-db " .dump | "sqlite-exe" " staging-file "&>"sync-log))
         (sync-cmd     (if fork-to-background 
                           (conc "/usr/bin/env NBFAKE_LOG="*toppath*"/logs/last-server-sync-"(current-process-id)".log nbfake "sync-cmd-core)
                           sync-cmd-core))

	 (min-intersync-delay (configf:lookup-number *configdat* "server" "minimum-intersync-delay" default: 30))
         (default-duty-cycle 0.1)
         (duty-cycle   (configf:lookup-number *configdat* "server" "sync-duty-cycle" default: default-duty-cycle))
         (last-sync-seconds 10) ;; we will adjust this to a measurement and delay last-sync-seconds * (1 - duty-cycle)
         (calculate-off-time (lambda (work-duration duty-cycle)
                                  (* (/ (- 1 duty-cycle) duty-cycle) last-sync-seconds)))
         (off-time min-intersync-delay)
         (do-a-sync
          (lambda ()



            (if (common:simple-file-lock lockfile)
	        (begin

                  (if (not fork-to-background)


                      (thread-sleep! (max off-time min-intersync-delay)))



                  (if (not (configf:lookup *configdat* "server" "disable-db-snapshot"))
                      (common:snapshot-file mtdbfile subdir: ".db-snapshot"))
		  (delete-file* staging-file)
		  (let* ((start-time (current-milliseconds))
                         (res (system sync-cmd))
                         (res2 
                          (cond
                           ((eq? 0 res)
		            (delete-file* (conc mtdbfile ".backup"))
                            (if (eq? 0 (file-size sync-log))
                                (delete-file sync-log))
		            (system (conc "/bin/mv " staging-file " " mtdbfile))
                            
                            (set! last-sync-seconds (/ (- (current-milliseconds) start-time) 1000))
                            (set! off-time (calculate-off-time
                                            last-sync-seconds
                                            (cond
                                             ((and (number? duty-cycle) (> duty-cycle 0) (< duty-cycle 1))
                                              duty-cycle)
                                             (else
                                              (debug:print 0 *default-log-port* "WARNING: ["(common:human-time)"] server.sync-duty-cycle is invalid.  Should be a number between 0 and 1, but "duty-cycle" was specified.  Using default value: "default-duty-cycle)
                                              default-duty-cycle))))
                            
                            (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" SYNC took "last-sync-seconds" sec")

                            'sync-completed)
                           (else
                            (system (conc "/bin/cp "sync-log" "sync-log".fail"))
                            (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)"] Sync failed. See log at "sync-log".fail")
                            (if (file-exists? (conc mtdbfile ".backup"))
                                (system (conc "/bin/cp "mtdbfile ".backup " mtdbfile)))
                            #f)))
                         (common:simple-file-release-lock lockfile)



                         res2)))

                ;; else






                (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" other SYNC in progress; not syncing.")
                'parallel-sync
                ) ;; end if got lockfile




            ) ;; end lambda
          ))
    do-a-sync))
    
(define (server:writable-watchdog dbstruct)
  (thread-sleep! 10) ;; delay for startup

  (let* ((do-a-sync (server:get-bruteforce-syncer dbstruct)))
    (if (and (not (args:get-arg "-sync-to-megatest.db")) ;; conditions under which we do not run the sync
	     (args:get-arg "-server"))
	(let loop ()
          (debug:print 0 *default-log-port* "INFO: syncer thread sleeping for server.minimum-intersync-delay seconds ["min-intersync-delay"]")

	  (do-a-sync)
	  
	  ;; keep going unless time to exit
	  ;;
	  (if (not *time-to-exit*)
	      (let delay-loop ((count 0))
		;;(debug:print-info 13 *default-log-port* "delay-loop top; count="count" pid="(current-process-id)" this-wd-num="this-wd-num" *time-to-exit*="*time-to-exit*)
		(if (eq? count 0)
                    ((server:get-bruteforce-syncer dbstruct fork-to-background: #t)))
		(if (and (not *time-to-exit*)
			 (< count 6)) ;; was 11, changing to 4. 
		    (begin
		      (thread-sleep! 1)
		      (delay-loop (+ count 1))))
		(if (not *time-to-exit*) (loop))))
	  ;; time to exit, close the no-sync db here
	  ;; (db:no-sync-close-db no-sync-db)
	  (if (common:low-noise-print 30)
	      (debug:print-info 0 *default-log-port* "Exiting watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)
				))))))


(define (server:old-writable-watchdog dbstruct)
  (thread-sleep! 10) ;; delay for startup
  (let* ((sqlite-exe   (or (get-environment-variable "MT_SQLITE3_EXE"))) ;; defined in cfg.sh
         (sync-log     (or (args:get-arg "-sync-log") (conc *toppath* "/logs/sync-" (current-process-id) "-" (get-host-name) ".log")))
	 (tmp-area     (common:get-db-tmp-area))
	 (tmp-db       (conc tmp-area "/megatest.db"))
	 (staging-file (conc *toppath* "/.megatest.db"))
	 (mtdbfile     (conc *toppath* "/megatest.db"))
	 (lockfile     (conc tmp-db ".lock"))
         (sync-cmd     (conc sqlite-exe" " tmp-db " .dump | "sqlite-exe" " staging-file "&>"sync-log))
	 (min-intersync-delay (configf:lookup-number *configdat* "server" "minimum-intersync-delay" default: 30))
         (default-duty-cycle 0.9)
         (duty-cycle   (configf:lookup-number *configdat* "server" "sync-duty-cycle" default: default-duty-cycle))
         (last-sync-seconds 10) ;; we will adjust this to a measurement and delay last-sync-seconds * (1 - duty-cycle)
         (calculate-off-time (lambda (work-duration duty-cycle)
                                  (* (/ (- 1 duty-cycle) duty-cycle) last-sync-seconds)))
         (off-time min-intersync-delay)
         (do-a-sync
          (lambda ()
            (if (common:simple-file-lock lockfile)
	        (begin

                  (if (not (configf:lookup *configdat* "server" "disable-db-snapshot"))
                      (common:snapshot-file mtdbfile subdir: ".db-snapshot"))
		  (delete-file* staging-file)
		  (let* ((start-time (current-milliseconds))
                         (res (system sync-cmd)))
                    (cond
                     ((eq? 0 res)
		      (delete-file* (conc mtdbfile ".backup"))
                      (if (eq? 0 (file-size sync-log))
                          (delete-file sync-log))
		      (system (conc "/bin/mv " staging-file " " mtdbfile))

                      (set! last-sync-seconds (/ (- (current-milliseconds) start-time) 1000))
                      (set! off-time (calculate-off-time
                                         last-sync-seconds
                                         (cond
                                          ((and (number? duty-cycle) (> duty-cycle 0) (< duty-cycle 1))
                                           duty-cycle)
                                          (else
                                           (debug:print 0 *default-log-port* "WARNING: ["(common:human-time)"] server.sync-duty-cycle is invalid.  Should be a number between 0 and 1, but "duty-cycle" was specified.  Using default value: "default-duty-cycle)
                                           default-duty-cycle))))

                      (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" SYNC took "last-sync-seconds" sec")
                      #t)
                     (else
                      (system (conc "/bin/cp "sync-log" "sync-log".fail"))
                      (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)"] Sync failed. See log at "sync-log".fail")
                      (if (file-exists? (conc mtdbfile ".backup"))
                          (system (conc "/bin/cp "mtdbfile ".backup " mtdbfile)))))
                    (common:simple-file-release-lock lockfile)))
                ;; else
                (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" other SYNC in progress; not syncing.")
                ) ;; end if got lockfile
            ) ;; end lambda
          ))
    
    (if (and (not (args:get-arg "-sync-to-megatest.db")) ;; conditions under which we do not run the sync
	     (args:get-arg "-server"))
	(let loop ()
          (debug:print 0 *default-log-port* "INFO: syncer thread sleeping for server.minimum-intersync-delay seconds ["min-intersync-delay"]")
          (thread-sleep! (max off-time min-intersync-delay))
	  (do-a-sync)
	  
	  ;; keep going unless time to exit
	  ;;
	  (if (not *time-to-exit*)
	      (let delay-loop ((count 0))
		;;(debug:print-info 13 *default-log-port* "delay-loop top; count="count" pid="(current-process-id)" this-wd-num="this-wd-num" *time-to-exit*="*time-to-exit*)
		(if (eq? count 0)
                    (do-a-sync))
		(if (and (not *time-to-exit*)
			 (< count 6)) ;; was 11, changing to 4. 
		    (begin
		      (thread-sleep! 1)
		      (delay-loop (+ count 1))))
		(if (not *time-to-exit*) (loop))))
	  ;; time to exit, close the no-sync db here
	  ;; (db:no-sync-close-db no-sync-db)
	  (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)
				))))))

;;   (let ((legacy-sync  (common:run-sync?)))
;;         (sync-stale-seconds (configf:lookup-number *configdat* "server" "sync-stale-seconds" default: 300))
;; 	(debug-mode   (debug:debug-mode 1))
;; 	(last-time    (current-seconds))
;; 	(no-sync-db   (db:open-no-sync-db))
;;         (sync-duration 0) ;; run time of the sync in milliseconds
;;         ;;(this-wd-num  (begin (mutex-lock! *wdnum*mutex) (let ((x *wdnum*)) (set! *wdnum* (add1 *wdnum*)) (mutex-unlock! *wdnum*mutex) x)))
;;         )
;;     (set! *no-sync-db* no-sync-db) ;; make the no sync db available to api calls
;;     (debug:print-info 2 *default-log-port* "Periodic sync thread started.")
;;     (debug:print-info 3 *default-log-port* "watchdog starting. legacy-sync is " legacy-sync" pid="(current-process-id)  );;  " this-wd-num="this-wd-num)
;;     (if (and legacy-sync (not *time-to-exit*))
;; 	(let* (;;(dbstruct (db:setup))
;; 	       (mtdb       (dbr:dbstruct-mtdb dbstruct))
;; 	       (mtpath     (db:dbdat-get-path mtdb))
;; 	       (tmp-area   (common:get-db-tmp-area))
;; 	       (lockfile   (conc tmp-area "/megatest.db.lock"))
;; 	       (start-file (conc tmp-area "/.start-sync"))
;; 	       (end-file   (conc tmp-area "/.end-sync")))
;; 	  (debug:print-info 0 *default-log-port* "Server running, periodic sync started.")
;; 	  (let loop ()
;; 	    ;; sync for filesystem local db writes
;; 	    ;;
;; 	    
;; 	    (mutex-lock! *db-multi-sync-mutex*)
;; 		  (let* ((need-sync        (>= *db-last-access* *db-last-sync*)) ;; no sync since last write
;; 			 (sync-in-progress *db-sync-in-progress*)
;; 			 
;; 			 (min-intersync-delay (configf:lookup-number *configdat* "server" "minimum-intersync-delay" default: 5))
;; 			 (should-sync      (and (not *time-to-exit*)
;; 						(> (- (current-seconds) *db-last-sync*) min-intersync-delay))) ;; sync every five seconds minimum, deprecated logic, can probably be removed
;; 			 (start-time       (current-seconds))
;; 			 (cpu-load-adj     (alist-ref 'adj-proc-load (common:get-normalized-cpu-load #f)))
;; 			 (mt-mod-time      (file-modification-time mtpath))
;; 			 (last-sync-start  (if (common:file-exists? start-file)
;; 					       (file-modification-time start-file)
;; 					       0))
;; 			 (last-sync-end    (if (common:file-exists? end-file)
;; 					       (file-modification-time end-file)
;; 					       10))
;; 			 (sync-period      (+ 3 (* cpu-load-adj 30))) ;; as adjusted load increases increase the sync period
;; 			 (recently-synced  (and (< (- start-time mt-mod-time) sync-period) ;; not useful if sync didn't modify megatest.db!
;; 						(< mt-mod-time last-sync-start)))
;; 			 (sync-done        (<= last-sync-start last-sync-end))
;; 			 (sync-stale       (> start-time (+ last-sync-start sync-stale-seconds)))
;; 			 (will-sync-pre        (and (not *time-to-exit*)       ;; do not start a sync if we are in the process of exiting
;; 						    have-lock?
;; 						    (or need-sync should-sync)
;; 						    (or sync-done sync-stale)
;; 						    (not sync-in-progress)
;; 						    (not recently-synced)))
;; 			 (will-sync        (if will-sync-pre
;; 					       ;; delay get lock until we decide to sync
;; 					       #t ;; (server:have-sync-lock?) 
;; 					       #f)))
;; 		    ;; if another server is syncing, postpone sync
;; 		    (if (and will-sync-pre (not will-sync))
;; 			(set! *db-last-sync* start-time))
;; 		    (debug:print-info 13 *default-log-port* "WD writable-watchdog top of loop.  need-sync="need-sync" sync-in-progress=" sync-in-progress
;; 				      " should-sync="should-sync" start-time="start-time" mt-mod-time="mt-mod-time" recently-synced="recently-synced" will-sync="will-sync
;; 				      " sync-done=" sync-done " sync-period=" sync-period)
;; 		    (if (and (> sync-period 5)
;; 			     (common:low-noise-print 30 "sync-period"))
;; 			(debug:print-info 0 *default-log-port* "Increased sync period due to long sync times, sync took: " sync-period " seconds."))
;; 		    ;; (if recently-synced (debug:print-info 0 *default-log-port* "Skipping sync due to recently-synced flag=" recently-synced))
;; 		    ;; (debug:print-info 0 *default-log-port* "need-sync: " need-sync " sync-in-progress: " sync-in-progress " should-sync: " should-sync " will-sync: " will-sync)
;; 		    (if will-sync (set! *db-sync-in-progress* #t))
;; 		    (mutex-unlock! *db-multi-sync-mutex*)
;; 		    (if will-sync
;; 			(let (;; (max-sync-duration  (configf:lookup-number *configdat* "server" "max-sync-duration")) ;; KEEPING THIS AVAILABLE BUT SHOULD NOT USE, I'M PRETTY SURE IT DOES NOT WORK!
;; 			      (sync-start         (current-milliseconds)))
;; 			  (with-output-to-file start-file (lambda ()(print (current-process-id))))
;; 			  
;; 			  ;; put lock here
;; 			  
;; 			  ;; (if (or (not max-sync-duration)
;; 			  ;;        (< sync-duration max-sync-duration)) ;; NOTE: db:sync-to-megatest.db keeps track of time of last sync and syncs incrementally
;; 			  (let ((res        (db:sync-to-megatest.db dbstruct no-sync-db: no-sync-db))) ;; did we sync any data? If so need to set the db touched flag to keep the server alive
;; 			    (set! sync-duration (- (current-milliseconds) sync-start))
;; 			    (if (> res 0) ;; some records were transferred, keep the db alive
;; 				(begin
;; 				  (mutex-lock! *heartbeat-mutex*)
;; 				  (set! *db-last-access* (current-seconds))
;; 				  (mutex-unlock! *heartbeat-mutex*)
;; 				  (debug:print-info 0 *default-log-port* "sync called, " res " records transferred."))
;; 				(debug:print-info 2 *default-log-port* "sync called but zero records transferred")))))
;; 		    ;;                         ;; TODO: factor this next routine out into a function
;; 		    ;;                         (with-input-from-pipe ;; this should not block other threads but need to verify this
;; 		    ;;                          (conc "megatest -sync-to-megatest.db -m testsuite:" (common:get-area-name) ":" *toppath*)
;; 		    ;;                          (lambda ()
;; 		    ;;                            (let loop ((inl (read-line))
;; 		    ;;                                       (res #f))
;; 		    ;;                              (if (eof-object? inl)
;; 		    ;;                                  (begin
;; 		    ;;                                    (set! sync-duration (- (current-milliseconds) sync-start))
;; 		    ;;                                    (cond
;; 		    ;;                                     ((not res)
;; 		    ;;                                      (debug:print 0 *default-log-port* "ERROR: sync from /tmp db to megatest.db appears to have failed. Recommended that you stop your runs and run \"megatest -cleanup-db\""))
;; 		    ;;                                     ((> res 0)
;; 		    ;;                                      (mutex-lock! *heartbeat-mutex*)
;; 		    ;;                                      (set! *db-last-access* (current-seconds))
;; 		    ;;                                      (mutex-unlock! *heartbeat-mutex*))))
;; 		    ;;                                  (let ((num-synced (let ((matches (string-match "^Synced (\\d+).*$" inl)))
;; 		    ;;                                                      (if matches
;; 		    ;;                                                          (string->number (cadr matches))
;; 		    ;;                                                          #f))))
;; 		    ;;                                    (loop (read-line)
;; 		    ;;                                          (or num-synced res))))))))))
;; 		    (if will-sync
;; 			(begin
;; 			  (mutex-lock! *db-multi-sync-mutex*)
;; 			  (set! *db-sync-in-progress* #f)
;; 			  (set! *db-last-sync* start-time)
;; 			  (with-output-to-file end-file (lambda ()(print (current-process-id))))
;; 
;; 			  ;; release lock here
;; 			  ;; (server:release-sync-lock)
;; 			  (mutex-unlock! *db-multi-sync-mutex*)))
;; 		    (if (and debug-mode
;; 			     (> (- start-time last-time) 60))
;; 			(begin
;; 			  (set! last-time start-time)
;; 			  (debug:print-info 4 *default-log-port* "timestamp -> " (seconds->time-string (current-seconds)) ", time since start -> " (seconds->hr-min-sec (- (current-seconds) *time-zero*))))))
;; 		  
;; 		  ;; keep going unless time to exit
;; 		  ;;
;; 		  (if (not *time-to-exit*)
;; 		      (let delay-loop ((count 0))
;; 			;;(debug:print-info 13 *default-log-port* "delay-loop top; count="count" pid="(current-process-id)" this-wd-num="this-wd-num" *time-to-exit*="*time-to-exit*)
;; 			
;; 			(if (and (not *time-to-exit*)
;; 				 (< count 6)) ;; was 11, changing to 4. 
;; 			    (begin
;; 			      (thread-sleep! 1)
;; 			      (delay-loop (+ count 1))))
;; 			(if (not *time-to-exit*) (loop))))
;; 		  ;; time to exit, close the no-sync db here

;; 		  (db:no-sync-close-db no-sync-db)
;; 		  (if (common:low-noise-print 30)
;; 		      (debug:print-info 0 *default-log-port* "Exiting watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id) ))))))) ;;" this-wd-num="this-wd-num)))))))










|






|


|

>
|





|


>
>
>
|
|
>
|
>
>
|
>
>
>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
>
|
|
|
|
|
|
|
|
>
>
>
|
>
|
>
>
>
>
>
>
|
|
|
>
>
>
>



|


>
|
|
|
<
<
|
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
|
<
<
|
<
<
<
<
<
<
<
<
<
<
<
<
<
|
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
|
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
|
>
|
|
|
>
>

499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602


603


















































































604


605













606




















607

























































































































608
609
610
611
612
613
614
615
;;       (server:release-sync-lock)
;;       (server:have-sync-lock?))
;;      (else #f))))

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

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

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

(define (server:writable-watchdog dbstruct)
  (thread-sleep! 10) ;; delay for startup
  (let* ((do-a-sync  (server:get-bruteforce-syncer dbstruct))
         (final-sync (server:get-bruteforce-syncer dbstruct persist-until-sync: #t)))
    (when (and (not (args:get-arg "-sync-to-megatest.db")) ;; conditions under which we do not run the sync
	       (args:get-arg "-server"))


      


















































































      (let loop ()


	(do-a-sync)













        (if (not *time-to-exit*) (loop))) ;; keep going unless time to exit














































































































































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

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