| ︙ | | | ︙ | |
47
48
49
50
51
52
53
54
55
56
57
58
59
60
|
runname max-concurrent-jobs run-id
test-patts required-tests test-registry
registry-mutex flags keyvals run-info all-tests-registry
can-run-more-tests
((can-run-more-tests-count 0) : fixnum)
(last-fuel-check 0) ;; time when we last checked fuel
(beginning-of-time (current-seconds))
)
(defstruct runs:testdat
hed tal reg reruns test-record
test-name item-path jobgroup
waitons testmode newtal itemmaps prereqs-not-met)
|
>
|
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
|
runname max-concurrent-jobs run-id
test-patts required-tests test-registry
registry-mutex flags keyvals run-info all-tests-registry
can-run-more-tests
((can-run-more-tests-count 0) : fixnum)
(last-fuel-check 0) ;; time when we last checked fuel
(beginning-of-time (current-seconds))
(load-mgmt-function #f)
)
(defstruct runs:testdat
hed tal reg reruns test-record
test-name item-path jobgroup
waitons testmode newtal itemmaps prereqs-not-met)
|
| ︙ | | | ︙ | |
310
311
312
313
314
315
316
317
318
319
320
321
322
323
|
(define (runs:can-run-more-tests runsdat run-id jobgroup max-concurrent-jobs)
;; Take advantage of a good place to exit if running the one-pass methodology
(if (and (> (runs:dat-can-run-more-tests-count runsdat) 20)
(args:get-arg "-one-pass"))
(exit 0))
;; (thread-sleep! (cond ;; BB: check with Matt. Should this sleep move
;; ;; to cond clauses below where we determine we
;; ;; have too many jobs running rather than each
;; ;; time the and condition above is true (which
;; ;; seems like always)?
;; ((< (- (current-seconds)(runs:dat-beginning-of-time runsdat)) 30) ;; for the first 30 seconds do not throttle in any way
|
>
>
|
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
|
(define (runs:can-run-more-tests runsdat run-id jobgroup max-concurrent-jobs)
;; Take advantage of a good place to exit if running the one-pass methodology
(if (and (> (runs:dat-can-run-more-tests-count runsdat) 20)
(args:get-arg "-one-pass"))
(exit 0))
(if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
;; (thread-sleep! (cond ;; BB: check with Matt. Should this sleep move
;; ;; to cond clauses below where we determine we
;; ;; have too many jobs running rather than each
;; ;; time the and condition above is true (which
;; ;; seems like always)?
;; ((< (- (current-seconds)(runs:dat-beginning-of-time runsdat)) 30) ;; for the first 30 seconds do not throttle in any way
|
| ︙ | | | ︙ | |
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
|
(null? non-completed))
(debug:print-info 4 *default-log-port* "cond branch - " "ei-4")
(if (runs:can-keep-running? hed 20)
(begin
(runs:inc-cant-run-tests hed)
(debug:print-info 0 *default-log-port* "no fails in prerequisites for " hed " but also none running, keeping " hed " for now. Try count: " (hash-table-ref/default *seen-cant-run-tests* hed 0)) ;;
;; getting here likely means the system is way overloaded, kill a full minute before continuing
(thread-sleep! 5) ;; TODO: gate by normalized server load > 1.0 (maxload config thing)
;; num-retries code was here
;; we use this opportunity to move contents of reg to tal
(list (car newtal)(append (cdr newtal) reg) '() reruns)) ;; an issue with prereqs not yet met?
(begin
(debug:print-info 1 *default-log-port* "no fails in prerequisites for " hed " but nothing seen running in a while, dropping test " hed " from the run queue")
(let ((test-id (rmt:get-test-id run-id hed "")))
|
>
>
>
>
|
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
|
(null? non-completed))
(debug:print-info 4 *default-log-port* "cond branch - " "ei-4")
(if (runs:can-keep-running? hed 20)
(begin
(runs:inc-cant-run-tests hed)
(debug:print-info 0 *default-log-port* "no fails in prerequisites for " hed " but also none running, keeping " hed " for now. Try count: " (hash-table-ref/default *seen-cant-run-tests* hed 0)) ;;
;; getting here likely means the system is way overloaded, kill a full minute before continuing
;;
;; No runsdat, can't do this yet
;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
;;
(thread-sleep! 5) ;; TODO: gate by normalized server load > 1.0 (maxload config thing)
;; num-retries code was here
;; we use this opportunity to move contents of reg to tal
(list (car newtal)(append (cdr newtal) reg) '() reruns)) ;; an issue with prereqs not yet met?
(begin
(debug:print-info 1 *default-log-port* "no fails in prerequisites for " hed " but nothing seen running in a while, dropping test " hed " from the run queue")
(let ((test-id (rmt:get-test-id run-id hed "")))
|
| ︙ | | | ︙ | |
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
|
(if (vector? t)
(conc (db:test-get-state t) "/" (db:test-get-status t))
(conc " WARNING: t is not a vector=" t )))
prereqs-not-met)
", ") ") fails: " fails
"\nregistered? " (hash-table-ref/default test-registry (db:test-make-full-name test-name item-path) #f))
(if (and (not (null? prereqs-not-met))
(runs:lownoise (conc "waiting on tests " prereqs-not-met hed) 60))
(debug:print-info 2 *default-log-port* "waiting on tests; " (string-intersperse (runs:mixed-list-testname-and-testrec->list-of-strings prereqs-not-met) ", ")))
;; Don't know at this time if the test have been launched at some time in the past
;; i.e. is this a re-launch?
|
>
>
>
>
>
>
>
>
>
>
>
>
|
>
>
>
>
>
>
|
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
|
(if (vector? t)
(conc (db:test-get-state t) "/" (db:test-get-status t))
(conc " WARNING: t is not a vector=" t )))
prereqs-not-met)
", ") ") fails: " fails
"\nregistered? " (hash-table-ref/default test-registry (db:test-make-full-name test-name item-path) #f))
;; well, first lets see if cpu load throttling is enabled. If so wait around until the
;; average cpu load is under the threshold before continuing
;;
(if (runs:dat-load-mgmt-function runsdat)
((runs:dat-load-mgmt-function runsdat))
(runs:dat-load-mgmt-function-set!
runsdat
(lambda ()
;; jobtools maxload is useful for where the full Megatest run is done on one machine
(if (and (not (common:on-homehost?))
maxload) ;; only gate if maxload is specified, NOTE: maxload is normalized, i.e. load=1 means all cpus fully utilized
(common:wait-for-normalized-load maxload "Waiting for load to drop before starting more tests" #f))
;; jobtools maxhomehostload is intended to prevent overloading on the homehost which can cause database corruption issues
(if maxhomehostload
(common:wait-for-homehost-load maxhomehostload
(conc "Waiting for homehost load to drop below normalized value of " maxhomehostload))))))
(if (and (not (null? prereqs-not-met))
(runs:lownoise (conc "waiting on tests " prereqs-not-met hed) 60))
(debug:print-info 2 *default-log-port* "waiting on tests; " (string-intersperse (runs:mixed-list-testname-and-testrec->list-of-strings prereqs-not-met) ", ")))
;; Don't know at this time if the test have been launched at some time in the past
;; i.e. is this a re-launch?
|
| ︙ | | | ︙ | |
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
|
(null? non-completed)
(not (member 'exclusive testmode)))))
;; (hash-table-delete! *max-tries-hash* (db:test-make-full-name test-name item-path))
;; we are going to reset all the counters for test retries by setting a new hash table
;; this means they will increment only when nothing can be run
(set! *max-tries-hash* (make-hash-table))
;; well, first lets see if cpu load throttling is enabled. If so wait around until the
;; average cpu load is under the threshold before continuing
;; jobtools maxload is useful for where the full Megatest run is done on one machine
(if (and (not (common:on-homehost?))
maxload) ;; only gate if maxload is specified, NOTE: maxload is normalized, i.e. load=1 means all cpus fully utilized
(common:wait-for-normalized-load maxload "Waiting for load to drop before starting more tests" #f))
;; jobtools maxhomehostload is intended to prevent overloading on the homehost which can cause database corruption issues
(if maxhomehostload
(common:wait-for-homehost-load maxhomehostload
(conc "Waiting for homehost load to drop below normalized value of " maxhomehostload)))
(run:test run-id run-info keyvals runname test-record flags #f test-registry all-tests-registry)
|
<
<
<
<
<
<
<
<
<
<
<
<
<
<
|
|
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
|
(null? non-completed)
(not (member 'exclusive testmode)))))
;; (hash-table-delete! *max-tries-hash* (db:test-make-full-name test-name item-path))
;; we are going to reset all the counters for test retries by setting a new hash table
;; this means they will increment only when nothing can be run
(set! *max-tries-hash* (make-hash-table))
(run:test run-id run-info keyvals runname test-record flags #f test-registry all-tests-registry runsdat)
(runs:incremental-print-results run-id)
(hash-table-set! test-registry (db:test-make-full-name test-name item-path) 'running)
(runs:shrink-can-run-more-tests-count runsdat) ;; DELAY TWEAKER (still needed?)
;; (thread-sleep! *global-delta*)
(if (or (not (null? tal))(not (null? reg)))
(runs:loop-values tal reg reglen regfull reruns)
#f))
|
| ︙ | | | ︙ | |
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
|
prereqs-not-met) ", ")))
(if (or (null? fails)
(member 'toplevel testmode))
(begin
;; couldn't run, take a breather
(if (runs:lownoise "Waiting for more work to do..." 60)
(debug:print-info 0 *default-log-port* "Waiting for more work to do..."))
(thread-sleep! 1)
(list (car newtal)(cdr newtal) reg reruns))
;; the waiton is FAIL so no point in trying to run hed ever again
(begin
(let ((my-test-id (rmt:get-test-id run-id test-name item-path)))
(mt:test-set-state-status-by-id-unless-completed run-id my-test-id "COMPLETED" "PREQ_FAIL" "Failed to run due to failed prerequisites2"))
(if (or (not (null? reg))(not (null? tal)))
(if (vector? hed)
(begin
(debug:print 1 *default-log-port* "WARNING: Dropping test " test-name "/" item-path
" from the launch list as it has prerequistes that are FAIL")
(let ((test-id (rmt:get-test-id run-id hed "")))
(if test-id (mt:test-set-state-status-by-id-unless-completed run-id test-id "COMPLETED" "PREQ_FAIL" "Failed to run due to failed prerequisites")))
|
>
>
|
<
<
<
|
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
|
prereqs-not-met) ", ")))
(if (or (null? fails)
(member 'toplevel testmode))
(begin
;; couldn't run, take a breather
(if (runs:lownoise "Waiting for more work to do..." 60)
(debug:print-info 0 *default-log-port* "Waiting for more work to do..."))
(if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
(thread-sleep! 1)
(list (car newtal)(cdr newtal) reg reruns))
;; the waiton is FAIL so no point in trying to run hed ever again
(begin
(let ((my-test-id (rmt:get-test-id run-id test-name item-path)))
(mt:test-set-state-status-by-id-unless-completed run-id my-test-id "COMPLETED" "PREQ_FAIL" "Failed to run due to failed prerequisites2"))
(if (or (not (null? reg))(not (null? tal)))
(if (vector? hed)
(begin
(debug:print 1 *default-log-port* "WARNING: Dropping test " test-name "/" item-path
" from the launch list as it has prerequistes that are FAIL")
(let ((test-id (rmt:get-test-id run-id hed "")))
(if test-id (mt:test-set-state-status-by-id-unless-completed run-id test-id "COMPLETED" "PREQ_FAIL" "Failed to run due to failed prerequisites")))
|
| ︙ | | | ︙ | |
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
|
(debug:print-info 4 *default-log-port* "Exiting loop with...\n hed=" hed "\n tal=" tal "\n reruns=" reruns))
))) ;; end loop on sorted test names
;; this is the point where everything is launched and now you can mark the run in metadata table as all launched
(rmt:set-var (conc "lunch-complete-" run-id) "yes")
;; now *if* -run-wait we wait for all tests to be done
;; Now wait for any RUNNING tests to complete (if in run-wait mode)
(thread-sleep! 10) ;; I think there is a race condition here. Let states/statuses settle
(let wait-loop ((num-running (rmt:get-count-tests-running-for-run-id run-id #t)) ;; fastmode=yes
(prev-num-running 0))
;; (debug:print-info 13 *default-log-port* "num-running=" num-running ", prev-num-running=" prev-num-running)
(if (and (or (args:get-arg "-run-wait")
(equal? (configf:lookup *configdat* "setup" "run-wait") "yes"))
(> num-running 0))
(begin
;; Here we mark any old defunct tests as incomplete. Do this every fifteen minutes
;; (debug:print 0 *default-log-port* "Got here eh! num-running=" num-running " (> num-running 0) " (> num-running 0))
(if (> (current-seconds)(+ last-time-incomplete 900))
(let ((actual-num-running (rmt:get-count-tests-running-for-run-id run-id #f))) ;; fastmode=no
(debug:print-info 0 *default-log-port* "Marking stuck tests as INCOMPLETE while waiting for run " run-id
". Running as pid " (current-process-id) " on " (get-host-name))
(set! last-time-incomplete (current-seconds)) ;; FIXME, this might be causing slow down - use of set!
(rmt:find-and-mark-incomplete run-id #f)
(debug:print-info 0 *default-log-port* "run-wait specified, waiting on " actual-num-running
" tests in RUNNING, REMOTEHOSTSTART or LAUNCHED state at "
(time->string (seconds->local-time (current-seconds))))))
(thread-sleep! 1) ;; (if (>= num-running max-concurrent-jobs) 5 1))
(wait-loop (rmt:get-count-tests-running-for-run-id run-id #t) ;; fastmode=yes
num-running))))
;; LET* ((test-record
;; we get here on "drop through". All done!
;; this is moved to runs:run-testes since this function is getting called twice to ensure everthing is completed.
;; (debug:print-info 0 *default-log-port* "Calling Post Hook")
|
>
>
>
|
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
|
(debug:print-info 4 *default-log-port* "Exiting loop with...\n hed=" hed "\n tal=" tal "\n reruns=" reruns))
))) ;; end loop on sorted test names
;; this is the point where everything is launched and now you can mark the run in metadata table as all launched
(rmt:set-var (conc "lunch-complete-" run-id) "yes")
;; now *if* -run-wait we wait for all tests to be done
;; Now wait for any RUNNING tests to complete (if in run-wait mode)
(if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
(thread-sleep! 10) ;; I think there is a race condition here. Let states/statuses settle
(let wait-loop ((num-running (rmt:get-count-tests-running-for-run-id run-id #t)) ;; fastmode=yes
(prev-num-running 0))
;; (debug:print-info 13 *default-log-port* "num-running=" num-running ", prev-num-running=" prev-num-running)
(if (and (or (args:get-arg "-run-wait")
(equal? (configf:lookup *configdat* "setup" "run-wait") "yes"))
(> num-running 0))
(begin
;; Here we mark any old defunct tests as incomplete. Do this every fifteen minutes
;; (debug:print 0 *default-log-port* "Got here eh! num-running=" num-running " (> num-running 0) " (> num-running 0))
(if (> (current-seconds)(+ last-time-incomplete 900))
(let ((actual-num-running (rmt:get-count-tests-running-for-run-id run-id #f))) ;; fastmode=no
(debug:print-info 0 *default-log-port* "Marking stuck tests as INCOMPLETE while waiting for run " run-id
". Running as pid " (current-process-id) " on " (get-host-name))
(set! last-time-incomplete (current-seconds)) ;; FIXME, this might be causing slow down - use of set!
(rmt:find-and-mark-incomplete run-id #f)
(debug:print-info 0 *default-log-port* "run-wait specified, waiting on " actual-num-running
" tests in RUNNING, REMOTEHOSTSTART or LAUNCHED state at "
(time->string (seconds->local-time (current-seconds))))))
(if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
(thread-sleep! 1) ;; (if (>= num-running max-concurrent-jobs) 5 1))
(wait-loop (rmt:get-count-tests-running-for-run-id run-id #t) ;; fastmode=yes
num-running))))
;; LET* ((test-record
;; we get here on "drop through". All done!
;; this is moved to runs:run-testes since this function is getting called twice to ensure everthing is completed.
;; (debug:print-info 0 *default-log-port* "Calling Post Hook")
|
| ︙ | | | ︙ | |
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
|
(if (not (vector? t))
(conc t)
(conc (db:test-get-testname t) ":" (db:test-get-state t) "/" (db:test-get-status t))))
lst))
;; parent-test is there as a placeholder for when parent-tests can be run as a setup step
;;
(define (run:test run-id run-info keyvals runname test-record flags parent-test test-registry all-tests-registry)
;; All these vars might be referenced by the testconfig file reader
;;
;; NEED to reprocess testconfig here, ensuring that item variables are available.
;; This is for Tal's issue with item-specific env vars not being set for use in skip.
;; HSD https://hsdes.intel.com/appstore/icf/index.html#/article?articleId=1408763273
;;
(let* ((test-name (tests:testqueue-get-testname test-record))
|
|
|
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
|
(if (not (vector? t))
(conc t)
(conc (db:test-get-testname t) ":" (db:test-get-state t) "/" (db:test-get-status t))))
lst))
;; parent-test is there as a placeholder for when parent-tests can be run as a setup step
;;
(define (run:test run-id run-info keyvals runname test-record flags parent-test test-registry all-tests-registry runsdat)
;; All these vars might be referenced by the testconfig file reader
;;
;; NEED to reprocess testconfig here, ensuring that item variables are available.
;; This is for Tal's issue with item-specific env vars not being set for use in skip.
;; HSD https://hsdes.intel.com/appstore/icf/index.html#/article?articleId=1408763273
;;
(let* ((test-name (tests:testqueue-get-testname test-record))
|
| ︙ | | | ︙ | |
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
|
(rmt:register-test run-id test-name item-path)
(set! test-id (rmt:get-test-id run-id test-name item-path))))
(debug:print-info 4 *default-log-port* "test-id=" test-id ", run-id=" run-id ", test-name=" test-name ", item-path=\"" item-path "\"")
(set! testdat (rmt:get-test-info-by-id run-id test-id))
(if (not testdat)
(begin
(debug:print-info 0 *default-log-port* "WARNING: server is overloaded, trying again in one second")
(thread-sleep! 1)
(loop)))))
(if (not testdat) ;; should NOT happen
(debug:print-error 0 *default-log-port* "failed to get test record for test-id " test-id))
(set! test-id (db:test-get-id testdat))
(if (common:file-exists? test-path)
(change-directory test-path)
|
>
|
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
|
(rmt:register-test run-id test-name item-path)
(set! test-id (rmt:get-test-id run-id test-name item-path))))
(debug:print-info 4 *default-log-port* "test-id=" test-id ", run-id=" run-id ", test-name=" test-name ", item-path=\"" item-path "\"")
(set! testdat (rmt:get-test-info-by-id run-id test-id))
(if (not testdat)
(begin
(debug:print-info 0 *default-log-port* "WARNING: server is overloaded, trying again in one second")
(if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
(thread-sleep! 1)
(loop)))))
(if (not testdat) ;; should NOT happen
(debug:print-error 0 *default-log-port* "failed to get test record for test-id " test-id))
(set! test-id (db:test-get-id testdat))
(if (common:file-exists? test-path)
(change-directory test-path)
|
| ︙ | | | ︙ | |