115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
|
(cmdln (conc (common:get-megatest-exe)
" -server " (or target-host "-") (if (equal? (configf:lookup *configdat* "server" "daemonize") "yes")
" -daemonize "
"")
;; " -log " logfile
" -m testsuite:" testsuite)) ;; (conc " >> " logfile " 2>&1 &")))))
(log-rotate (make-thread common:rotate-logs "server run, rotate logs thread"))
(load-limit (configf:lookup-number *configdat* "server" "load-limit" default: 0.9)))
;; we want the remote server to start in *toppath* so push there
(push-directory areapath)
(debug:print 0 *default-log-port* "INFO: Trying to start server (" cmdln ") ...")
(thread-start! log-rotate)
;; host.domain.tld match host?
(if (and target-host
;; look at target host, is it host.domain.tld or ip address and does it
;; match current ip or hostname
(not (string-match (conc "("curr-host "|" curr-host"\\..*)") target-host))
(not (equal? curr-ip target-host)))
(begin
(debug:print-info 0 *default-log-port* "Starting server on " target-host ", logfile is " logfile)
(setenv "TARGETHOST" target-host)))
(setenv "TARGETHOST_LOGF" logfile)
(common:wait-for-normalized-load load-limit " delaying server start due to load" remote-host: (get-environment-variable "TARGETHOST")) ;; do not try starting servers on an already overloaded machine, just wait forever
(system (conc "nbfake " cmdln))
(unsetenv "TARGETHOST_LOGF")
(if (get-environment-variable "TARGETHOST")(unsetenv "TARGETHOST"))
(thread-join! log-rotate)
(pop-directory)))
;; given a path to a server log return: host port startseconds
|
|
|
|
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
|
(cmdln (conc (common:get-megatest-exe)
" -server " (or target-host "-") (if (equal? (configf:lookup *configdat* "server" "daemonize") "yes")
" -daemonize "
"")
;; " -log " logfile
" -m testsuite:" testsuite)) ;; (conc " >> " logfile " 2>&1 &")))))
(log-rotate (make-thread common:rotate-logs "server run, rotate logs thread"))
(load-limit (configf:lookup-number *configdat* "jobtools" "maxhomehostload" default: 3.0)))
;; we want the remote server to start in *toppath* so push there
(push-directory areapath)
(debug:print 0 *default-log-port* "INFO: Trying to start server (" cmdln ") ...")
(thread-start! log-rotate)
;; host.domain.tld match host?
(if (and target-host
;; look at target host, is it host.domain.tld or ip address and does it
;; match current ip or hostname
(not (string-match (conc "("curr-host "|" curr-host"\\..*)") target-host))
(not (equal? curr-ip target-host)))
(begin
(debug:print-info 0 *default-log-port* "Starting server on " target-host ", logfile is " logfile)
(setenv "TARGETHOST" target-host)))
(setenv "TARGETHOST_LOGF" logfile)
(common:wait-for-normalized-load load-limit " delaying server start due to load" target-host) ;; do not try starting servers on an already overloaded machine, just wait forever
(system (conc "nbfake " cmdln))
(unsetenv "TARGETHOST_LOGF")
(if (get-environment-variable "TARGETHOST")(unsetenv "TARGETHOST"))
(thread-join! log-rotate)
(pop-directory)))
;; given a path to a server log return: host port startseconds
|
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
|
(> (length rec) 2))
(let ((start-time (list-ref rec 3))
(mod-time (list-ref rec 0)))
;; (print "start-time: " start-time " mod-time: " mod-time)
(and start-time mod-time
(> (- now start-time) 0) ;; been running at least 0 seconds
(< (- now mod-time) 16) ;; still alive - file touched in last 16 seconds
(< (- now start-time)
(+ (- (string->number (or (configf:lookup *configdat* "server" "runtime") "3600"))
180)
(random 360))) ;; under one hour running time +/- 180
))
#f))
srvlst)
(lambda (a b)
|
|
|
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
|
(> (length rec) 2))
(let ((start-time (list-ref rec 3))
(mod-time (list-ref rec 0)))
;; (print "start-time: " start-time " mod-time: " mod-time)
(and start-time mod-time
(> (- now start-time) 0) ;; been running at least 0 seconds
(< (- now mod-time) 16) ;; still alive - file touched in last 16 seconds
(< (- now start-time)
(+ (- (string->number (or (configf:lookup *configdat* "server" "runtime") "3600"))
180)
(random 360))) ;; under one hour running time +/- 180
))
#f))
srvlst)
(lambda (a b)
|
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
|
(define (server:login toppath)
(lambda (toppath)
(set! *db-last-access* (current-seconds)) ;; might not be needed.
(if (equal? *toppath* toppath)
#t
#f)))
;; timeout is in hours
(define (server:get-timeout)
(let ((tmo (configf:lookup *configdat* "server" "timeout")))
(if (and (string? tmo)
(string->number tmo))
(* 60 60 (string->number tmo))
;; (* 3 24 60 60) ;; default to three days
;;(* 60 60 1) ;; default to one hour
(* 60 5) ;; default to five minutes
)))
;; moving this here as it needs access to db and cannot be in common.
;;
(define (server:writable-watchdog dbstruct)
(thread-sleep! 0.05) ;; delay for startup
(let ((legacy-sync (common:run-sync?))
(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.")
|
|
>
|
|
>
|
<
<
<
<
|
>
|
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
|
(define (server:login toppath)
(lambda (toppath)
(set! *db-last-access* (current-seconds)) ;; might not be needed.
(if (equal? *toppath* toppath)
#t
#f)))
;; timeout is hms string: 1h 5m 3s, default is 1 minute
;;
(define (server:expiration-timeout)
(let ((tmo (configf:lookup *configdat* "server" "timeout")))
(if (and (string? tmo)
(common:hms-string->seconds tmo)) ;; BUG: hms-string->seconds is broken, if given "10" returns 0. Also, it doesn't belong in this logic unless the string->number is changed below
(* 3600 (string->number tmo))
60)))
;; moving this here as it needs access to db and cannot be in common.
;;
(define (server:writable-watchdog dbstruct)
(thread-sleep! 0.05) ;; delay for startup
(let ((legacy-sync (common:run-sync?))
(sync-stale-seconds (configf:lookup-number *configdat* "server" "sync-stale-seconds" default: 300))
(debug-mode (debug:debug-mode 1))
(last-time (current-seconds))
(no-sync-db (db:open-no-sync-db))
(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.")
|
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
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
|
(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*)
(should-sync (and (not *time-to-exit*)
(> (- (current-seconds) *db-last-sync*) 5))) ;; sync every five seconds minimum
(start-time (current-seconds))
(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))
(recently-synced (and (< (- start-time mt-mod-time) 4) ;; not useful if sync didn't modify megatest.db!
(< mt-mod-time last-sync-start)))
(sync-done (<= last-sync-start last-sync-end))
(will-sync (and (or need-sync should-sync)
sync-done
(not sync-in-progress)
(not recently-synced))))
(debug:print-info 13 *default-log-port* "WD writable-watchdog top of loop. need-sync="need-sync" sync-in-progress=" sync-in-progress
" should-sync="should-sync" start-time="start-time" mt-mod-time="mt-mod-time" recently-synced="recently-synced" will-sync="will-sync
" sync-done=" sync-done)
;; (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 ((sync-start (current-milliseconds)))
(with-output-to-file start-file (lambda ()(print (current-process-id))))
;; put lock here
(if (< sync-duration 300)
(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))))
|
>
|
>
>
|
>
>
|
|
|
>
>
>
>
|
|
|
>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
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
|
(debug:print-info 0 *default-log-port* "Server running, periodic sync started.")
(let loop ()
;; sync for filesystem local db writes
;;
(mutex-lock! *db-multi-sync-mutex*)
(let* ((need-sync (>= *db-last-access* *db-last-sync*)) ;; no sync since last write
(sync-in-progress *db-sync-in-progress*)
(min-intersync-delay (configf:lookup-number *configdat* "server" "minimum-intersync-delay" default: 5))
(should-sync (and (not *time-to-exit*)
(> (- (current-seconds) *db-last-sync*) min-intersync-delay))) ;; sync every five seconds minimum, deprecated logic, can probably be removed
(start-time (current-seconds))
(cpu-load-adj (alist-ref 'adj-proc-load (common:get-normalized-cpu-load #f)))
(mt-mod-time (file-modification-time mtpath))
(last-sync-start (if (common:file-exists? start-file)
(file-modification-time start-file)
0))
(last-sync-end (if (common:file-exists? end-file)
(file-modification-time end-file)
10))
(sync-period (+ 3 (* cpu-load-adj 30))) ;; as adjusted load increases increase the sync period
(recently-synced (and (< (- start-time mt-mod-time) sync-period) ;; not useful if sync didn't modify megatest.db!
(< mt-mod-time last-sync-start)))
(sync-done (<= last-sync-start last-sync-end))
(sync-stale (> start-time (+ last-sync-start sync-stale-seconds)))
(will-sync (and (not *time-to-exit*) ;; do not start a sync if we are in the process of exiting
(or need-sync should-sync)
(or sync-done sync-stale)
(not sync-in-progress)
(not recently-synced))))
(debug:print-info 13 *default-log-port* "WD writable-watchdog top of loop. need-sync="need-sync" sync-in-progress=" sync-in-progress
" should-sync="should-sync" start-time="start-time" mt-mod-time="mt-mod-time" recently-synced="recently-synced" will-sync="will-sync
" sync-done=" sync-done " sync-period=" sync-period)
(if (and (> sync-period 5)
(common:low-noise-print 30 "sync-period"))
(debug:print-info 0 *default-log-port* "Increased sync period due to long sync times, sync took: " sync-period " seconds."))
;; (if recently-synced (debug:print-info 0 *default-log-port* "Skipping sync due to recently-synced flag=" recently-synced))
;; (debug:print-info 0 *default-log-port* "need-sync: " need-sync " sync-in-progress: " sync-in-progress " should-sync: " should-sync " will-sync: " will-sync)
(if will-sync (set! *db-sync-in-progress* #t))
(mutex-unlock! *db-multi-sync-mutex*)
(if will-sync
(let (;; (max-sync-duration (configf:lookup-number *configdat* "server" "max-sync-duration")) ;; KEEPING THIS AVAILABLE BUT SHOULD NOT USE, I'M PRETTY SURE IT DOES NOT WORK!
(sync-start (current-milliseconds)))
(with-output-to-file start-file (lambda ()(print (current-process-id))))
;; put lock here
;; (if (or (not max-sync-duration)
;; (< sync-duration max-sync-duration)) ;; NOTE: db:sync-to-megatest.db keeps track of time of last sync and syncs incrementally
(let ((res (db:sync-to-megatest.db dbstruct no-sync-db: no-sync-db))) ;; did we sync any data? If so need to set the db touched flag to keep the server alive
(set! sync-duration (- (current-milliseconds) sync-start))
(if (> res 0) ;; some records were transferred, keep the db alive
(begin
(mutex-lock! *heartbeat-mutex*)
(set! *db-last-access* (current-seconds))
(mutex-unlock! *heartbeat-mutex*)
(debug:print-info 0 *default-log-port* "sync called, " res " records transferred."))
(debug:print-info 2 *default-log-port* "sync called but zero records transferred")))))
;; ;; TODO: factor this next routine out into a function
;; (with-input-from-pipe ;; this should not block other threads but need to verify this
;; (conc "megatest -sync-to-megatest.db -m testsuite:" (common:get-area-name) ":" *toppath*)
;; (lambda ()
;; (let loop ((inl (read-line))
;; (res #f))
;; (if (eof-object? inl)
;; (begin
;; (set! sync-duration (- (current-milliseconds) sync-start))
;; (cond
;; ((not res)
;; (debug:print 0 *default-log-port* "ERROR: sync from /tmp db to megatest.db appears to have failed. Recommended that you stop your runs and run \"megatest -cleanup-db\""))
;; ((> res 0)
;; (mutex-lock! *heartbeat-mutex*)
;; (set! *db-last-access* (current-seconds))
;; (mutex-unlock! *heartbeat-mutex*))))
;; (let ((num-synced (let ((matches (string-match "^Synced (\\d+).*$" inl)))
;; (if matches
;; (string->number (cadr matches))
;; #f))))
;; (loop (read-line)
;; (or num-synced res))))))))))
(if will-sync
(begin
(mutex-lock! *db-multi-sync-mutex*)
(set! *db-sync-in-progress* #f)
(set! *db-last-sync* start-time)
(with-output-to-file end-file (lambda ()(print (current-process-id))))
|