| ︙ | | |
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
|
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
|
+
-
+
+
+
+
+
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
-
+
-
+
+
+
|
(set! ezsteps #t) ;; set the needed flag
(set! ezstepslst
(append (or ezstepslst '())
(list (list "subrun" (conc "{subrun=true} " mt-cmd)))))))
;; process the ezsteps
(if ezsteps
(let* ((envdbf (conc "/tmp/."(current-user-name)"-"(current-process-id)"-"run-id"-"test-id".db"))
(let* ((all-steps-dat (make-hash-table))) ;; keep all the info around as stepname ==> alist; where 'params is the params list (add other stuff as needed)
(all-steps-dat (make-hash-table))) ;; keep all the info around as stepname ==> alist;
;;; where 'params is the params list (add other
;;; stuff as needed)
(if (not (common:file-exists? ".ezsteps"))(create-directory ".ezsteps"))
;; if ezsteps was defined then we are sure to have at least one step but check anyway
(if (not (> (length ezstepslst) 0))
(debug:print-error 0 *default-log-port* "ezsteps defined but ezstepslst is zero length")
(let ((all-step-names (map car ezstepslst)))
(setenv "MT_STEP_NAMES" (string-intersperse all-step-names " "))
(let loop ((ezstep (car ezstepslst))
(tal (cdr ezstepslst))
(prevstep #f))
(debug:print-info 0 *default-log-port* "Processing ezstep \"" (string-intersperse ezstep " ") "\"")
;; check exit-info (vector-ref exit-info 1)
(if (launch:einf-exit-status exit-info) ;; (vector-ref exit-info 1)
(let* ((logpro-used (launch:runstep ezstep run-id test-id exit-info m tal testconfig all-steps-dat))
(stepname (car ezstep))
(stepparms (hash-table-ref all-steps-dat stepname)))
(setenv "MT_STEP_NAME" stepname)
(pp (hash-table->alist all-steps-dat))
;; if logpro-used read in the stepname.dat file
(if (and logpro-used (common:file-exists? (conc stepname ".dat")))
(launch:load-logpro-dat run-id test-id stepname))
(if (steprun-good? logpro-used (launch:einf-exit-code exit-info) stepparms)
(if (not (null? tal))
(loop (car tal) (cdr tal) stepname))
(debug:print 0 *default-log-port* "WARNING: step " (car ezstep) " failed. Stopping")))
(debug:print 0 *default-log-port* "WARNING: a prior step failed, stopping at " ezstep)))))))))
(let loop ((ezstep (car ezstepslst))
(tal (cdr ezstepslst))
(prevstep #f))
(debug:print-info 0 *default-log-port* "Processing ezstep \"" (string-intersperse ezstep " ") "\"")
;; check exit-info (vector-ref exit-info 1)
(if (launch:einf-exit-status exit-info) ;; (vector-ref exit-info 1)
(let* ((logpro-used (launch:runstep ezstep run-id test-id exit-info m
tal testconfig all-steps-dat prevstep envdbf))
(stepname (car ezstep))
(stepparms (hash-table-ref all-steps-dat stepname)))
(setenv "MT_STEP_NAME" stepname)
(pp (hash-table->alist all-steps-dat))
;; if logpro-used read in the stepname.dat file
(if (and logpro-used (common:file-exists? (conc stepname ".dat")))
(launch:load-logpro-dat run-id test-id stepname))
(if (steprun-good? logpro-used (launch:einf-exit-code exit-info) stepparms)
(if (not (null? tal))
(loop (car tal) (cdr tal) stepname))
(debug:print 0 *default-log-port* "WARNING: step " (car ezstep) " failed. Stopping")))
(debug:print 0 *default-log-port* "WARNING: a prior step failed, stopping at " ezstep))
))))))))
(define (launch:monitor-job run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m work-area runtlim misc-flags)
(let* ((update-period (string->number (or (configf:lookup *configdat* "setup" "test-stats-update-period") "30")))
(start-seconds (current-seconds))
(calc-minutes (lambda ()
(inexact->exact
(round
(-
(current-seconds)
start-seconds)))))
(kill-tries 0))
;; (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area)
;; (tests:set-full-meta-info test-id run-id (calc-minutes) work-area)
(tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area 10)
(tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area 10 update-db: #t)
(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
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
|
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
|
-
+
-
+
-
+
-
-
-
-
+
-
-
|
(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))
((equal? status "DEAD")
(tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
(tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f update-db: #t)
(rmt:set-state-status-and-roll-up-items run-id test-id 'foo "RUNNING" "n/a" "was marked dead; really still running.")
;;(set! kill-reason "KILLING TEST because it was marked as DEAD by launch:handle-zombie-tests (might indicate really overloaded server or else overzealous setup.deadtime)") ;; MARK RUNNING
(set! kill-job? #f)))
(debug:print 4 *default-log-port* "cpu: " new-cpu-load " disk: " new-disk-free " last-sync: " last-sync " do-sync: " do-sync)
(launch:handle-zombie-tests run-id)
(when do-sync
(if do-sync ;; save meta data about the running of this test
;;(with-output-to-file (conc (getenv "MT_TEST_RUN_DIR") "/last-loadinfo.log" #:append)
;; (lambda () (pp (list (current-seconds) new-cpu-load new-disk-free (calc-minutes)))))
(common:telemetry-log "zombie" (conc "launch:monitor-job - dosync started at "(current-seconds)))
(tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
(common:telemetry-log "zombie" (conc "launch:monitor-job - dosync finished at "(current-seconds))))
|
| ︙ | | |
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
|
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
|
-
+
|
(begin
(thread-sleep! 3) ;; (+ 3 (random 6))) ;; add some jitter to the call home time to spread out the db accesses
(if (hash-table-ref/default misc-flags 'keep-going #f) ;; keep originals for cpu-load and disk-free unless they change more than the allowed delta
(loop (calc-minutes)
(or new-cpu-load cpu-load)
(or new-disk-free disk-free)
(if do-sync (current-seconds) last-sync)))))))
(tests:update-central-meta-info run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f))) ;; NOTE: Checking twice for keep-going is intentional
(tests:update-central-meta-info run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f update-db: #t))) ;; NOTE: Checking twice for keep-going is intentional
(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")
|
| ︙ | | |
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
|
400
401
402
403
404
405
406
407
408
409
410
411
412
413
|
-
|
(debug:print 0 *default-log-port* "INFO: missing files from " work-area ": " (string-intersperse bad-files ", "))
(launch:test-copy testpath work-area))))
;; one more time, change to the work-area directory
(change-directory work-area)))
) ;; let*
(if contour (setenv "MT_CONTOUR" contour))
|
| ︙ | | |
463
464
465
466
467
468
469
470
471
472
473
474
475
476
|
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
|
+
+
+
|
(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
|
| ︙ | | |
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
|
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
|
-
+
|
;; 100% COMPLETED/ (PASS,FAIL,ABORT etc.) ==> COMPLETED / X where X is same as itemized rollup
;; > 3 RUNNING with not test_dead do nothing (run should already be RUNNING/ na
;; > 0 RUNNING and test_dead then send KILLREQ ==> COMPLETED
;; 0 RUNNING ==> this is actually the first condition, should not get here
(define (launch:end-of-run-check run-id )
(let* ((not-completed-cnt (rmt:get-not-completed-cnt run-id))
(running-cnt (rmt:get-count-tests-running-for-run-id run-id #f)) ;; fastmode=no
(running-cnt (rmt:get-count-tests-running-for-run-id run-id))
(all-test-launched (rmt:get-var (conc "lunch-complete-" run-id)))
(current-state (rmt:get-run-state run-id))
(current-status (rmt:get-run-status run-id)))
;;get-vars run-id to query metadata table to check if all completed. if all-test-launched = yes then only not-completed-cnt = 0 means everyting is completed if no entry found in the table do nothing
(debug:print 0 *default-log-port* "Running test cnt :" running-cnt)
(rmt:set-state-status-and-roll-up-run run-id current-state current-status)
(runs:update-junit-test-reporter-xml run-id)
|
| ︙ | | |