| ︙ | | |
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
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
|
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
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
|
+
+
+
+
-
+
+
-
+
-
+
-
+
+
+
-
+
+
-
+
-
+
-
+
+
+
+
|
;; (max-connections 4096)
;; do all the busy work of finding and setting up conn for
;; connecting to a server
;;
(define (tt:client-connect-to-server ttdat dbfname run-id testsuite)
(assert (tt:valid-run-id run-id dbfname) "FATAL: invalid run-id "run-id)
(debug:print-info 2 *default-log-port* "tt:client-connect-to-server " dbfname " " run-id)
(let* ((conn (hash-table-ref/default (tt-conns ttdat) dbfname #f))
(server-start-proc (lambda ()
(print "dbfname: "dbfname)
(assert (equal? dbfname "main.db") ;; only main.db is started here
"FATAL: called server-start-proc for db other than main.db")
(tt:server-process-run
(tt-areapath ttdat)
testsuite ;; (dbfile:testsuite-name)
(common:find-local-megatest)
run-id))))
run-id))
))
(if conn
(begin
; (debug:print-info 0 *default-log-port* "already connected to the server")
(debug:print-info 2 *default-log-port* "already connected to a server")
conn) ;; we are already connected to the server
(let* ((sdat (tt:get-current-server-info ttdat dbfname)))
(match sdat
((host port start-time server-id pid dbfname2 servinffile)
(assert (equal? dbfname dbfname2) "FATAL: read server info from wrong file.")
;(debug:print-info 0 *default-log-port* "in match servinffile:" servinffile)
(debug:print-info 2 *default-log-port* "no conn - in match servinffile:" servinffile)
(let* ((host-port (conc host":"port))
(conn (make-tt-conn
host: host
port: port
host-port: host-port
dbfname: dbfname
servinf-file: servinffile
server-id: server-id
server-start: start-time
pid: pid)))
;; verify we can talk to this server
(let* ((result (tt:timed-ping host port server-id))
(ping-res (car result))
(ping (cdr result)))
(debug:print-info 0 *default-log-port* "ping time: " ping)
(debug:print-info 2 *default-log-port* "host " host " port " port " ping time: " ping " result " ping-res)
(case ping-res
((running)
(debug:print-info 2 *default-log-port* "Setting conn = " conn " in hash table")
(hash-table-set! (tt-conns ttdat) dbfname conn) ;;; is this ok to save before validating that the connection is good?
conn)
((starting)
(thread-sleep! 0.5)
(debug:print-info 0 *default-log-port* "server for " dbfname " is in starting state, retrying connect")
(tt:client-connect-to-server ttdat dbfname run-id testsuite))
(else
(let* ((curr-secs (current-seconds)))
;; rm the (last server) would go here
(if (> (- curr-secs (tt-last-serv-start ttdat)) 10)
(begin
(tt-last-serv-start-set! ttdat curr-secs)
(server-start-proc))) ;; start server if 30 sec since last attempt
(server-start-proc))) ;; start server if 10 sec since last attempt
(thread-sleep! 1)
(debug:print-info 2 *default-log-port* "server ping result was neither running nor starting. Retrying connect")
(tt:client-connect-to-server ttdat dbfname run-id testsuite)))))))
(else ;; no good server found, if haven't started server in > 5 secs, start another
(if (> (- (current-seconds) (tt-last-serv-start ttdat)) 5) ;; BUG - grow this number really do not want to swamp the machine with servers
(if (> (- (current-seconds) (tt-last-serv-start ttdat)) 3) ;; BUG - grow this number really do not want to swamp the machine with servers
(begin
(debug:print-info 0 *default-log-port* "No server found. Starting one for run-id "run-id" in dbfile "dbfname)
(debug:print-info 0 *default-log-port* "Starting server for "dbfname)
(server-start-proc)
(tt-last-serv-start-set! ttdat (current-seconds))))
(tt-last-serv-start-set! ttdat (current-seconds))
(thread-sleep! 3)
))
(thread-sleep! 1)
(debug:print-info 0 *default-log-port* "Connect to server for " dbfname)
(tt:client-connect-to-server ttdat dbfname run-id testsuite)))))))
(define (tt:timed-ping host port server-id)
(let* ((start-time (current-milliseconds))
(result (tt:ping host port server-id)))
(cons result (- (current-milliseconds) start-time))))
|
| ︙ | | |
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
|
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
|
+
+
|
(try-again)))))
;; client side handler
;;
;;(tt:handler #<tt> get-keys #f () 2 #f "/home/matt/data/megatest/ext-tests" #f "main.db" "ext-tests" "/home/matt/data/megatest/bin/.22.04/../megatest")
;;
(define (tt:handler ttdat cmd run-id params attemptnum area-dat areapath readonly-mode dbfname testsuite mtexe)
(debug:print 2 *default-log-port* "tt:handler cmd: " cmd " run-id: " run-id " attemptnum: " attemptnum)
;; NOTE: areapath is passed in and in tt struct. We'll use passed in value for now.
;; connect-to-server will start a server if needed.
(let* ((conn (tt:client-connect-to-server ttdat dbfname run-id testsuite))) ;; (hash-table-ref/default (tt-conns ttdat) dbfname #f)))
(if conn
;; have connection, call the server
(let* ((res (tt:send-receive ttdat conn cmd run-id params)))
;; res is (status errmsg result meta)
; (debug:print 0 *default-log-port* "conn:" conn " res: " res)
(match res
|
| ︙ | | |
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
|
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
|
-
+
|
((loaded)
(debug:print 0 *default-log-port* "WARNING: server for "dbfname" is loaded, slowing queries.")
(tt:backoff-incr (tt-conn-host conn)(tt-conn-port conn))
result) ;; (tt:handler ttdat cmd run-id params (+ attemptnum 1) area-dat areapath readonly-mode dbfname testsuite mtexe))
(else
result)))
(else ;; did not receive properly formated result
(if (not res) ;; tt:handler is telling us that communication failed
(if (not res) ;; tt:send-receive telling us that communication failed
(let* ((host (tt-conn-host conn))
(port (tt-conn-port conn))
;; (dbfname (tt-conn-port conn)) ;; 192.168.0.127:4242-726924:4.db
(pid (tt-conn-pid conn))
;;(servinf (tt-conn-servinf-file conn)))
(servinf (tt-servinf-file ttdat))) ;; (conc areapath"/.servinfo/"host":"port"-"pid":"dbfname))) ;; TODO, use (server:get-servinfo-dir areapath)
(hash-table-set! (tt-conns ttdat) dbfname #f)
|
| ︙ | | |
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
|
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
|
-
+
|
;; start server - addressed in client-connect-to-server
;; delay - addressed in client-connect-to-server
;; try again
(thread-sleep! 0.25) ;; dunno, I think this needs to be here
(tt:handler ttdat cmd run-id params (+ attemptnum 1) area-dat areapath readonly-mode dbfname testsuite mtexe))
))))
(begin ;; no server file, delay and try again
(debug:print 0 *default-log-port* "INFO: connection to server "host":"port" broken for "dbfname", but do not see servinf file "servinf)
(debug:print 2 *default-log-port* "INFO: connection to server "host":"port" broken for "dbfname", no servinf file. Server exited? ")
(thread-sleep! 0.5)
(tt:handler ttdat cmd run-id params (+ attemptnum 1) area-dat areapath readonly-mode dbfname testsuite mtexe))))
(begin ;; this case is where res is malformed. Probably should abort
(assert #f "FATAL: tt:handler received bad data "res)
;; (debug:print 0 *default-log-port* "INFO: got corrupt data from server "host":"port", "res", for "dbfname", will try again.")
;; (tt:handler ttdat cmd run-id params (+ attemptnum 1) area-dat areapath readonly-mode dbfname testsuite mtexe)
)))))
|
| ︙ | | |
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
|
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
|
+
+
-
+
|
;; This is the routine called in megatest.scm to start a server
;;
;; Server viability is checked in keep-running. Blindly start and run here.
;;
(define (tt:start-server areapath run-id dbfname-in handler keys)
(assert areapath "FATAL: areapath not provided for tt:start-server")
;; is there already a server for this dbfile? Then exit.
(debug:print 2 *default-log-port* "tt:start-server: " dbfname-in)
(let* ((ttdat (make-tt areapath: areapath))
(dbfname (or dbfname-in (dbmod:run-id->dbfname run-id)))
(servers (tt:find-server areapath dbfname))) ;; should use tt:get-current-server-info instead
(debug:print 0 *default-log-port* "Found " (length servers) " already running for " dbfname)
(if (> (length servers) 4)
(if (> (length servers) 0)
(begin
(debug:print 0 *default-log-port* "INFO: found server(s) already running for db "dbfname", "(string-intersperse servers ",")" Exiting.")
(exit))
(let* ((dbstruct (dbmod:open-dbmoddb areapath run-id dbfname (dbfile:db-init-proc) keys)))
(tt-handler-set! ttdat (handler dbstruct))
(let* ((tcp-thread (make-thread
(lambda ()
|
| ︙ | | |
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
|
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
|
-
-
+
|
(if (tt-cleanup-proc ttdat)
((tt-cleanup-proc ttdat)))
(dbfile:with-no-sync-db nosyncdbpath
(lambda (db)
(let* ((dbtmpname (dbr:dbstruct-dbtmpname dbstruct)))
(debug:print-info 0 *default-log-port* "Running clean up, including removing db file "dbtmpname)
(db:no-sync-del! db dbfname)
#;(if dbtmpname
(delete-file dbtmpname))))))))
|
| ︙ | | |
574
575
576
577
578
579
580
581
582
583
584
585
586
587
|
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
|
+
|
(debug:print 0 *default-log-port* "Failed to get server lock for "dbfname)
#f))))))
(if (and res (common:low-noise-print 120 "top server message"))
(debug:print-info 0 *default-log-port* "Keep running, I'm the top server for "
dbfname" on "(tt-host ttdat)":"(tt-port ttdat)))
res))
(else
;; wrong servinfo file
(debug:print-info 0 *default-log-port* "I'm not the lead server: "servers)
(let* ((leadsrv (car servers)))
(match leadsrv
((host port startseconds server-id pid dbfname servinfofile)
(let* ((result (tt:timed-ping host port server-id))
(res (car result))
(ping (cdr result)))
|
| ︙ | | |
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
|
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
|
+
+
+
+
+
-
-
+
+
+
|
;; get servers listed, last part of name must match :<dbfname>
;; if more than one, wait one second and look again
;; future: ping oldest, if alive remove other :<dbfname> files
;;
(define (tt:find-server areapath dbfname)
(let* ((servdir (tt:get-servinfo-dir areapath))
(sfiles (glob (conc servdir"/*:"dbfname))))
;; filter the files here by looking in processes table (if we are not main.db)
;; and or look at the time stamp on the servinfo file, a running server will
;; touch the file every minute (again, this will only apply for main.db)
sfiles))
;; given a path to a server info file return: host port startseconds server-id pid dbfname logf
;; example of what it's looking for in the log file:
;; SERVER STARTED: 10.38.175.67:50216 AT 1616502350.0 server-id: 4907e90fc55c7a09694e3f658c639cf4
;;
(define (tt:server-get-info logf)
(let ((server-rx (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+) server-id: (\\S+) pid: (\\d+) dbfname: (\\S+)")) ;; SERVER STARTED: host:port AT timesecs server id
(dbprep-rx (regexp "^SERVER: dbprep"))
(dbprep-found 0)
(bad-dat (list #f #f #f #f #f #f logf)))
(let ((fdat (handle-exceptions
exn
(begin
;; WARNING: this is potentially dangerous to blanket ignore the errors
(debug:print-info 0 *default-log-port* "Unable to get server info from "logf", exn="(condition->list exn))
;; BUG, TODO: add err checking, for now blanket ignore the errors?
(debug:print-info 0 *default-log-port* "Unable to get server info from "logf
", exn="(condition->list exn))
'()) ;; no idea what went wrong, call it a bad server, return empty list
(with-input-from-file logf read-lines))))
(if (null? fdat) ;; bad data, return bad-dat
bad-dat
(let loop ((inl (car fdat))
(tail (cdr fdat))
(lnum 0))
|
| ︙ | | |
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
|
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
|
-
+
+
-
-
+
+
+
-
-
+
+
+
-
+
+
-
+
-
+
|
(define (tt:server-process-run areapath testsuite mtexe run-id #!key (profile-mode "")) ;; areapath is *toppath* for a given testsuite area
(assert areapath "FATAL: tt:server-process-run called without areapath defined.")
(assert testsuite "FATAL: tt:server-process-run called without testsuite defined.")
(assert mtexe "FATAL: tt:server-process-run called without mtexe defined.")
;; mtest -server - -m testsuite:ext-tests -db 6.db
(let* ((dbfname (dbmod:run-id->dbfname run-id))
(load (get-normalized-cpu-load))
(trying (length (tt:find-server areapath dbfname)))
(srvrs (tt:find-server areapath dbfname))
(trying (length srvrs))
(nrun (number-of-processes-running (conc "mtest.*server.*"testsuite".*"dbfname))))
(cond
((> load 2.0)
(debug:print 0 *default-log-port* "Normalized load "load" on " (get-host-name) " is over the limit of 2.0. Not starting a server.")
(thread-sleep! 1))
(debug:print 0 *default-log-port* "Normalized load "load" on " (get-host-name) " is over the limit of 2.0. Not starting a server. Please reduce the load on "(get-host-name)" by killing some processes")
(thread-sleep! 1)
#f)
((> nrun 100)
(debug:print 0 *default-log-port* nrun" servers running on " (get-host-name) ", not starting another.")
(thread-sleep! 1))
((> trying 4)
(thread-sleep! 1)
#f)
((> trying 2)
(debug:print 0 *default-log-port* trying" servers registered in .servinfo dir. not starting another.")
(thread-sleep! 1))
(thread-sleep! 1)
#f)
(else
(if (not (file-exists? (conc areapath"/logs")))
(create-directory (conc areapath"/logs") #t))
(let* ((logfile (conc areapath "/logs/server-"dbfname"-"(current-process-id)".log")) ;; -" curr-pid "-" target-host ".log"))
(cmdln (conc
mtexe
" -startdir "areapath
" -server - ";; (or target-host "-")
" -m testsuite:"testsuite
" -db "dbfname ;; (dbmod:run-id->dbfname run-id)
" " profile-mode
(conc " >> " logfile " 2>&1 &"))))
;; we want the remote server to start in *toppath* so push there
;; (push-directory areapath) ;; use cd in the command line instead
(debug:print 0 *default-log-port* "INFO: Trying to start server in tcp mode (" cmdln ") at "(common:human-time)" for "areapath)
(debug:print 2 *default-log-port* "INFO: Trying to start server in tcp mode (" cmdln ") at "(common:human-time)" for "areapath)
;; (debug:print 0 *default-log-port* "INFO: starting server at " (common:human-time))
(system cmdln)
;; ;; use below to go back to nbfake - nbfake does cause trouble ...
;; (setenv "NBFAKE_QUIET" "yes") ;; BUG: change to with-environment-variable ...
;; (setenv "NBFAKE_LOG" logfile)
;; (system (conc "cd "areapath" ; nbfake " cmdln))
;; (unsetenv "NBFAKE_QUIET")
;; (unsetenv "NBFAKE_LOG")
;;(pop-directory)
)))))
#t)))))
;;======================================================================
;; tcp connection stuff
;;======================================================================
;; find a port and start tcp-server. This only starts the tcp portion of
;; the server, look at (tt:start-server ...) above for the entry point
|
| ︙ | | |