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
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
|
;; ;; get registered dashboards
;; ;;
;; (define (mddb:get-dashboards)
;; (let ((db (mddb:open-db)))
;; (query fetch-column
;; (sql db "SELECT ipaddr || ':' || portnum FROM dashboards;"))))
;;======================================================================
;; T E S T L A U N C H I N G P E R I T E M W I T H H O S T T Y P E S
;;======================================================================
;;
;; [hosts]
;; arm cubie01 cubie02
;; x86_64 zeus xena myth01
;; allhosts #{g hosts arm} #{g hosts x86_64}
;;
;; [host-types]
;; general #MTLOWESTLOAD #{g hosts allhosts}
;; arm #MTLOWESTLOAD #{g hosts arm}
;; nbgeneral nbjob run JOBCOMMAND -log $MT_LINKTREE/$MT_TARGET/$MT_RUNNAME.$MT_TESTNAME-$MT_ITEM_PATH.lgo
;;
;; [host-rules]
;; # maxnload => max normalized load
;; # maxnjobs => max jobs per cpu
;; # maxjobrate => max jobs per second
;; general maxnload=1.1; maxnjobs=1.2; maxjobrate=0.1
;;
;; [launchers]
;; envsetup general
;; xor/%/n 4C16G
;; % nbgeneral
;;
;; [jobtools]
;; # if defined and not "no" flexi-launcher will bypass "launcher" unless no match.
;; flexi-launcher yes
;; launcher nbfake
;;
(define (common:get-launcher configdat testname itempath)
(let ((fallback-launcher (configf:lookup configdat "jobtools" "launcher")))
(if (and (configf:lookup configdat "jobtools" "flexi-launcher") ;; overrides launcher
(not (equal? (configf:lookup configdat "jobtools" "flexi-launcher") "no")))
(let* ((launchers (hash-table-ref/default configdat "launchers" '())))
(if (null? launchers)
fallback-launcher
(let loop ((hed (car launchers))
(tal (cdr launchers)))
(let ((patt (car hed))
(host-type (cadr hed)))
(if (tests:match patt testname itempath)
(begin
(debug:print-info 2 *default-log-port* "Have flexi-launcher match for " testname "/" itempath " = " host-type)
(let ((launcher (configf:lookup configdat "host-types" host-type)))
(if launcher
(let* ((launcher-parts (string-split launcher))
(launcher-exe (car launcher-parts)))
(if (equal? launcher-exe "#MTLOWESTLOAD") ;; this is our special case, we will find the lowest load and craft a nbfake commandline
(let host-loop ((targ-host (common:get-least-loaded-host (cdr launcher-parts) host-type configdat))
(count 100))
(if targ-host
(conc "remrun " targ-host)
(if (> count 0)
(begin
(debug:print 0 *default-log-port* "INFO: Waiting for a host for host-type " host-type)
(thread-sleep! (- 101 count))
(host-loop (common:get-least-loaded-host (cdr launcher-parts) host-type configdat)
(- count 1)))
(begin
(debug:print 0 *default-log-port* "FATAL: Failed to find a host from #MTLOWESTLOAD for host-type " host-type)
(exit)))))
launcher))
(begin
(debug:print-info 0 *default-log-port* "WARNING: no launcher found for host-type " host-type)
(if (null? tal)
fallback-launcher
(loop (car tal)(cdr tal)))))))
;; no match, try again
(if (null? tal)
fallback-launcher
(loop (car tal)(cdr tal))))))))
fallback-launcher)))
;;======================================================================
;; faux-lock is deprecated. Please use simple-lock below
;;
(define (common:faux-lock keyname #!key (wait-time 8)(allow-lock-steal #t))
(if (rmt:no-sync-get/default keyname #f) ;; do not be tempted to compare to pid. locking is a one-shot action, if already locked for this pid it doesn't actually count
(if (> wait-time 0)
(begin
|
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
|
168
169
170
171
172
173
174
175
176
177
178
179
180
181
|
;; ;; get registered dashboards
;; ;;
;; (define (mddb:get-dashboards)
;; (let ((db (mddb:open-db)))
;; (query fetch-column
;; (sql db "SELECT ipaddr || ':' || portnum FROM dashboards;"))))
;;======================================================================
;; faux-lock is deprecated. Please use simple-lock below
;;
(define (common:faux-lock keyname #!key (wait-time 8)(allow-lock-steal #t))
(if (rmt:no-sync-get/default keyname #f) ;; do not be tempted to compare to pid. locking is a one-shot action, if already locked for this pid it doesn't actually count
(if (> wait-time 0)
(begin
|
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
|
(define (common:faux-unlock keyname #!key (force #f))
(if (or force (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))
(begin
(if (rmt:no-sync-get/default keyname #f) (rmt:no-sync-del! keyname))
#t)
#f))
;;======================================================================
;; simple lock. improve and converge on this one.
;;
(define (common:simple-lock keyname)
(rmt:no-sync-get-lock keyname))
(define (common:simple-unlock keyname #!key (force #f))
(rmt:no-sync-del! keyname))
;;======================================================================
;; ideally put all this info into the db, no need to preserve it across moving homehost
;;
;; return list of
;; ( reachable? cpuload update-time )
(define (common:get-host-info hostname)
(let* ((loadinfo (rmt:get-latest-host-load hostname)) ;; if this host happens to have been recently used by a test reuse the load data
(load (car loadinfo))
(load-sample-time (cdr loadinfo))
(load-sample-age (- (current-seconds) load-sample-time))
(loadinfo-timeout-seconds 6) ;; this was 20 seconds, seems way too lax. Switch to 6 seconds
(host-last-update-timeout-seconds 4)
(host-rec (hash-table-ref/default *host-loads* hostname #f))
)
(cond
((< load-sample-age loadinfo-timeout-seconds)
(list #t
load-sample-time
load))
((and host-rec
(< (current-seconds) (+ (host-last-update host-rec) host-last-update-timeout-seconds)))
(list #t
(host-last-update host-rec)
(host-last-cpuload host-rec )))
((common:unix-ping hostname)
(list #t
(current-seconds)
(alist-ref 'adj-core-load (common:get-normalized-cpu-load hostname)))) ;; this is cheaper than you might think. get-normalized-cpu-load is cached for up to 5 seconds
(else
(list #f 0 -1) ;; bad host, don't use!
))))
;;======================================================================
;; see defstruct host at top of file.
;; host: reachable last-update last-used last-cpuload
;;
(define (common:update-host-loads-table hosts-raw)
(let* ((hosts (filter (lambda (x)
(string-match (regexp "^\\S+$") x))
hosts-raw)))
(for-each
(lambda (hostname)
(let* ((rec (let ((h (hash-table-ref/default *host-loads* hostname #f)))
(if h
h
(let ((h (make-host)))
(hash-table-set! *host-loads* hostname h)
h))))
(host-info (common:get-host-info hostname))
(is-reachable (car host-info))
(last-reached-time (cadr host-info))
(load (caddr host-info)))
(host-reachable-set! rec is-reachable)
(host-last-update-set! rec last-reached-time)
(host-last-cpuload-set! rec load)))
hosts)))
;;======================================================================
;; go through the hosts from least recently used to most recently used, pick the first that meets the load criteral from the
;; [host-rules] section.
;;
(define (common:get-least-loaded-host hosts-raw host-type configdat)
(let* ((rdat (configf:lookup configdat "host-rules" host-type))
(rules (common:val->alist (or rdat "") convert: #t)) ;; maxnload, maxnjobs, maxjobrate
(maxnload (common:alist-ref/default 'maxnload rules 1.5)) ;; max normalized load
(maxnjobs (common:alist-ref/default 'maxnjobs rules 1.5)) ;; max normalized number of jobs
(maxjobrate (common:alist-ref/default 'maxjobrate rules (/ 1 6))) ;; max rate of submitting jobs to a given host in jobs/second
(hosts (filter (lambda (x)
(string-match (regexp "^\\S+$") x))
hosts-raw))
;; (best-host #f)
(get-rec (lambda (hostname)
;; (print "get-rec hostname=" hostname)
(let ((h (hash-table-ref/default *host-loads* hostname #f)))
(if h
h
(let ((h (make-host)))
(hash-table-set! *host-loads* hostname h)
h)))))
(best-load 99999)
(curr-time (current-seconds))
(get-hosts-sorted (lambda (hosts)
(sort hosts (lambda (a b)
(let ((a-rec (get-rec a))
(b-rec (get-rec b)))
;; (print "a=" a " a-rec=" a-rec " host-last-used=" (host-last-used a-rec))
;; (print "b=" b " b-rec=" b-rec " host-last-used=" (host-last-used b-rec))
(< (host-last-used a-rec)
(host-last-used b-rec))))))))
(debug:print 0 *default-log-port* "INFO: hosts-sorted=" (get-hosts-sorted hosts))
(if (null? hosts)
#f ;; no hosts to select from. All done and giving up now.
(let ((hosts-sorted (get-hosts-sorted hosts)))
(common:update-host-loads-table hosts)
(let loop ((hostname (car hosts-sorted))
(tal (cdr hosts-sorted))
(best-host #f))
(let* ((rec (get-rec hostname))
(reachable (host-reachable rec))
(load (host-last-cpuload rec))
(last-used (host-last-used rec))
(delta (- curr-time last-used))
(job-rate (if (> delta 0)
(/ 1 delta)
999)) ;; jobs per second
(new-best
(cond
((not reachable)
(debug:print 0 *default-log-port* "Skipping host " hostname " as it cannot be reached.")
best-host)
((and (< load maxnload) ;; load is acceptable
(< job-rate maxjobrate)) ;; job rate is acceptable
(set! best-load load)
hostname)
(else best-host))))
(debug:print 0 *default-log-port* "INFO: Trying host " hostname " with load " load ", last used " delta " seconds ago, with job-rate " job-rate " for running a test." )
(if new-best
(begin ;; found a host, return it
(debug:print 0 *default-log-port* "INFO: Found host: " new-best " load: " load " last-used: " delta " seconds ago, with job-rate: " job-rate)
(host-last-used-set! rec curr-time)
new-best)
(if (null? tal) #f (loop (car tal)(cdr tal) best-host)))))))))
(define (std-exit-procedure)
;;(common:telemetry-log-close)
(on-exit (lambda () 0))
;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
(let ((no-hurry (if *time-to-exit* ;; hurry up
#f
(begin
|
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
|
193
194
195
196
197
198
199
200
201
202
203
204
205
206
|
(define (common:faux-unlock keyname #!key (force #f))
(if (or force (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))
(begin
(if (rmt:no-sync-get/default keyname #f) (rmt:no-sync-del! keyname))
#t)
#f))
(define (std-exit-procedure)
;;(common:telemetry-log-close)
(on-exit (lambda () 0))
;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
(let ((no-hurry (if *time-to-exit* ;; hurry up
#f
(begin
|
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
|
(thread-start! th2)
(thread-join! th1)
)
)
0)
(define (make-and-init-remote)
(make-remote hh-dat: (common:get-homehost)
server-info: (if *toppath* (server:check-if-running *toppath*) #f)
server-timeout: (server:expiration-timeout)))
;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(set! *watchdog* (make-thread
(lambda ()
(handle-exceptions
exn
(begin
|
<
<
<
<
|
241
242
243
244
245
246
247
248
249
250
251
252
253
254
|
(thread-start! th2)
(thread-join! th1)
)
)
0)
;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(set! *watchdog* (make-thread
(lambda ()
(handle-exceptions
exn
(begin
|