209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
|
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
|
-
+
+
+
|
;; (tests:set-full-meta-info test-id run-id (calc-minutes) work-area)
(tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area 10)
(let loop ((minutes (calc-minutes))
(cpu-load (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
(disk-free (get-df (current-directory)))
(last-sync (current-seconds)))
(common:telemetry-log "zombie" (conc "launch:monitor-job - top of loop encountered at "(current-seconds)" with last-sync="last-sync))
;; (common:telemetry-log "zombie" (conc "launch:monitor-job -
;; top of loop encountered at "(current-seconds)" with
;; last-sync="last-sync))
(let* ((over-time (> (current-seconds) (+ last-sync update-period)))
(new-cpu-load (let* ((load (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
(delta (abs (- load cpu-load))))
(if (> delta 0.1) ;; don't bother updating with small changes
load
#f)))
(new-disk-free (let* ((df (if over-time ;; only get df every 30 seconds
|
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
|
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
|
-
+
|
(do-sync (or new-cpu-load new-disk-free over-time))
(test-info (rmt:get-test-info-by-id run-id test-id))
(state (db:test-get-state test-info))
(status (db:test-get-status test-info))
(kill-reason "no kill reason specified")
(kill-job? #f))
(common:telemetry-log "zombie" (conc "launch:monitor-job - decision time encountered at "(current-seconds)" with last-sync="last-sync" do-sync="do-sync" over-time="over-time" update-period="update-period))
#;(common:telemetry-log "zombie" (conc "launch:monitor-job - decision time encountered at "(current-seconds)" with last-sync="last-sync" do-sync="do-sync" over-time="over-time" update-period="update-period))
(cond
((test-get-kill-request run-id test-id)
(set! kill-reason "KILLING TEST since received kill request (KILLREQ)")
(set! kill-job? #t))
((and runtlim (> (- (current-seconds) start-seconds) runtlim))
(set! kill-reason (conc "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" (- (current-seconds) start-seconds) " seconds, limit=" runtlim))
(set! kill-job? #t))
|
463
464
465
466
467
468
469
470
471
472
473
474
475
476
|
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
|
+
+
+
|
(db:test-get-host test-info)
(begin
(debug:print 0 *default-log-port* "ERROR: failed to find a record for test-id " test-id ", exiting.")
(exit))))
(test-pid (db:test-get-process_id test-info)))
(cond
;; -mrw- I'm removing KILLREQ from this list so that a test in KILLREQ state is treated as a "do not run" flag.
;;((or (member (db:test-get-state test-info) '("INCOMPLETE" "KILLED" "UNKNOWN" "STUCK")) ;; prior run of this test didn't complete, go ahead and try to rerun
;; (and (equal? (db:test-get-state test-info) "COMPLETED") ;; completed/abort => rerun if asked
;; (member (db:test-get-status test-info) '("ABORT"))))
((member (db:test-get-state test-info) '("INCOMPLETE" "KILLED" "UNKNOWN" "STUCK")) ;; prior run of this test didn't complete, go ahead and try to rerun
(debug:print 0 *default-log-port* "INFO: test is INCOMPLETE or KILLED, treat this execute call as a rerun request")
;; (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")
(rmt:general-call 'set-test-start-time #f test-id)
(rmt:test-set-state-status run-id test-id "REMOTEHOSTSTART" "n/a" #f)
) ;; prime it for running
|