Overview
| Comment: | Get list of servers now works |
|---|---|
| Downloads: | Tarball | ZIP archive | SQL archive |
| Timelines: | family | ancestors | descendants | both | server-log-handshaking |
| Files: | files | file ages | folders |
| SHA1: |
f2556f210b10e1b4877a92e4bc3120b9 |
| User & Date: | matt on 2017-01-25 23:48:54 |
| Other Links: | branch diff | manifest | tags |
Context
|
2017-01-27
| ||
| 22:47 | Most needed changes made for multi-server with log handshaking. However there is a segfault ... check-in: 8a4205f90b user: matt tags: server-log-handshaking | |
|
2017-01-25
| ||
| 23:48 | Get list of servers now works check-in: f2556f210b user: matt tags: server-log-handshaking | |
| 22:02 | Stripped out nearly all server management stuff related to monitor.db server tables. partially replaced with log file based handshaking check-in: 6d32c79c1e user: matt tags: server-log-handshaking | |
Changes
Modified http-transport.scm from [c66a3571fe] to [bbf4c4febb].
| ︙ | ︙ | |||
334 335 336 337 338 339 340 | ;; run http-transport:keep-running in a parallel thread to monitor that the db is being ;; used and to shutdown after sometime if it is not. ;; (define (http-transport:keep-running) ;; if none running or if > 20 seconds since ;; server last used then start shutdown ;; This thread waits for the server to come alive | | | 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 |
;; run http-transport:keep-running in a parallel thread to monitor that the db is being
;; used and to shutdown after sometime if it is not.
;;
(define (http-transport:keep-running)
;; if none running or if > 20 seconds since
;; server last used then start shutdown
;; This thread waits for the server to come alive
(debug:print-info 0 *default-log-port* "Starting the sync-back, keep alive thread in server")
(let* ((server-start-time (current-seconds))
(server-info (let loop ((start-time (current-seconds))
(changed #t)
(last-sdat "not this"))
(let ((sdat #f))
(thread-sleep! 0.01)
(debug:print-info 0 *default-log-port* "Waiting for server alive signature")
|
| ︙ | ︙ | |||
356 357 358 359 360 361 362 |
(debug:print-info 0 *default-log-port* "Received server alive signature")
sdat)
(begin
(debug:print-info 0 *default-log-port* "Still waiting, last-sdat=" last-sdat)
(sleep 4)
(if (> (- (current-seconds) start-time) 120) ;; been waiting for two minutes
(begin
| | < | 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 |
(debug:print-info 0 *default-log-port* "Received server alive signature")
sdat)
(begin
(debug:print-info 0 *default-log-port* "Still waiting, last-sdat=" last-sdat)
(sleep 4)
(if (> (- (current-seconds) start-time) 120) ;; been waiting for two minutes
(begin
(debug:print-error 0 *default-log-port* "transport appears to have died, exiting server")
(exit))
(loop start-time
(equal? sdat last-sdat)
sdat)))))))
(iface (car server-info))
(port (cadr server-info))
(last-access 0)
|
| ︙ | ︙ | |||
396 397 398 399 400 401 402 |
(loop (+ count 1) 'running bad-sync-count (current-milliseconds)))
;; Check that iface and port have not changed (can happen if server port collides)
(mutex-lock! *heartbeat-mutex*)
(set! sdat *server-info*)
(mutex-unlock! *heartbeat-mutex*)
| | < | > > | > | | | 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 |
(loop (+ count 1) 'running bad-sync-count (current-milliseconds)))
;; Check that iface and port have not changed (can happen if server port collides)
(mutex-lock! *heartbeat-mutex*)
(set! sdat *server-info*)
(mutex-unlock! *heartbeat-mutex*)
(if (not (equal? sdat (list iface port)))
(let ((new-iface (car sdat))
(new-port (cadr sdat)))
(debug:print-info 0 *default-log-port* "WARNING: interface changed, refreshing iface and port info")
(set! iface new-iface)
(set! port new-port)
(debug:print 0 *default-log-port* "SERVER STARTED: " iface ":" port " AT " (current-seconds))
(flush-output *default-log-port*)))
;; Transfer *db-last-access* to last-access to use in checking that we are still alive
(mutex-lock! *heartbeat-mutex*)
(set! last-access *db-last-access*)
(mutex-unlock! *heartbeat-mutex*)
(if (common:low-noise-print 120 (conc "server running on " iface ":" port))
(begin
(debug:print 0 *default-log-port* "SERVER STARTED: " iface ":" port " AT " (current-seconds))
(flush-output *default-log-port*)))
(let* ((hrs-since-start (/ (- (current-seconds) server-start-time) 3600))
(adjusted-timeout (if (> hrs-since-start 1)
(- server-timeout (inexact->exact (round (* hrs-since-start 60)))) ;; subtract 60 seconds per hour
server-timeout)))
(if (common:low-noise-print 120 "server timeout")
(debug:print-info 0 *default-log-port* "Adjusted server timeout: " adjusted-timeout))
(cond
((and *server-run*
(> (+ last-access server-timeout)
(current-seconds)))
(if (common:low-noise-print 120 "server continuing")
(debug:print-info 0 *default-log-port* "Server continuing, seconds since last db access: " (- (current-seconds) last-access)))
(loop 0 server-state bad-sync-count (current-milliseconds)))
(else
(debug:print-info 0 *default-log-port* "Server timeed out. seconds since last db access: " (- (current-seconds) last-access))
(http-transport:server-shutdown port)))))))
(define (http-transport:server-shutdown port)
(let ((tdbdat (tasks:open-db)))
;;(BB> "http-transport:server-shutdown called")
(debug:print-info 0 *default-log-port* "Starting to shutdown the server. pid="(current-process-id))
;;
;; start_shutdown
;;
;; (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "shutting-down")
|
| ︙ | ︙ | |||
461 462 463 464 465 466 467 |
;; (/ *total-non-write-delay*
;; *number-non-write-queries*))
;; " ms")
(db:print-current-query-stats)
(debug:print-info 0 *default-log-port* "Server shutdown complete. Exiting")
| < < < < | 462 463 464 465 466 467 468 469 470 471 472 473 474 475 |
;; (/ *total-non-write-delay*
;; *number-non-write-queries*))
;; " ms")
(db:print-current-query-stats)
(debug:print-info 0 *default-log-port* "Server shutdown complete. Exiting")
(exit)))
;; all routes though here end in exit ...
;;
;; start_server?
;;
(define (http-transport:launch)
|
| ︙ | ︙ |
Modified server.scm from [d398b938c6] to [e489317c03].
| ︙ | ︙ | |||
45 46 47 48 49 50 51 | ;; ;; all routes though here end in exit ... ;; ;; start_server ;; (define (server:launch run-id transport-type) | < < < < < < < < < < < < < | < < < < < < < | 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 |
;;
;; all routes though here end in exit ...
;;
;; start_server
;;
(define (server:launch run-id transport-type)
(case transport-type
((http)(http-transport:launch))
;;((nmsg)(nmsg-transport:launch run-id))
((rpc) (rpc-transport:launch run-id))
(else (debug:print-error 0 *default-log-port* "unknown server type " transport-type))))
;;======================================================================
;; S E R V E R U T I L I T I E S
;;======================================================================
;; Get the transport
(define (server:get-transport)
|
| ︙ | ︙ | |||
129 130 131 132 133 134 135 | (curr-ip (server:get-best-guess-address curr-host)) (curr-pid (current-process-id)) (homehost (common:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" )) (target-host (car homehost)) (testsuite (common:get-testsuite-name)) (logfile (conc areapath "/logs/server-" curr-pid "-" target-host ".log")) (cmdln (conc (common:get-megatest-exe) | | | | > < < < < < < | | | | | | | | | | | | | | | | | | | | | > > > | > > | | > > > > > > > > > > > > > > > > > > > > > > > > > > > | 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 |
(curr-ip (server:get-best-guess-address curr-host))
(curr-pid (current-process-id))
(homehost (common:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" ))
(target-host (car homehost))
(testsuite (common:get-testsuite-name))
(logfile (conc areapath "/logs/server-" curr-pid "-" target-host ".log"))
(cmdln (conc (common:get-megatest-exe)
" -server " (or target-host "-") (if (equal? (configf:lookup *configdat* "server" "daemonize") "yes")
" -daemonize "
"")
" -log " logfile
" -m testsuite:" testsuite)) ;; (conc " >> " logfile " 2>&1 &")))))
(log-rotate (make-thread common:rotate-logs "server run, rotate logs thread")))
;; we want the remote server to start in *toppath* so push there
(push-directory areapath)
(debug:print 0 *default-log-port* "INFO: Trying to start server (" cmdln ") ...")
(thread-start! log-rotate)
;; host.domain.tld match host?
(if (and target-host
;; look at target host, is it host.domain.tld or ip address and does it
;; match current ip or hostname
(not (string-match (conc "("curr-host "|" curr-host"\\..*)") target-host))
(not (equal? curr-ip target-host)))
(begin
(debug:print-info 0 *default-log-port* "Starting server on " target-host ", logfile is " logfile)
(setenv "TARGETHOST" target-host)))
(setenv "TARGETHOST_LOGF" "server.log") ;; logfile)
(common:wait-for-normalized-load 4 " delaying server start due to load" remote-host: (get-environment-variable "TARGETHOST")) ;; do not try starting servers on an already overloaded machine, just wait forever
(system (conc "nbfake " cmdln))
(unsetenv "TARGETHOST_LOGF")
(if (get-environment-variable "TARGETHOST")(unsetenv "TARGETHOST"))
(thread-join! log-rotate)
(pop-directory)))
;; given a path to a server log return: host port startseconds
;;
(define (server:logf-get-start-info logf)
(let ((rx (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+)"))) ;; SERVER STARTED: host:port AT timesecs
(with-input-from-file
logf
(lambda ()
(let loop ((inl (read-line))
(lnum 0))
(if (not (eof-object? inl))
(let ((mlst (string-match rx inl)))
(if (not mlst)
(if (< lnum 500) ;; give up if more than 500 lines of server log read
(loop (read-line)(+ lnum 1))
(list #f #f #f))
(let ((dat (cdr mlst)))
(list (car dat) ;; host
(string->number (cadr dat)) ;; port
(string->number (caddr dat))))))
(list #f #f #f)))))))
;; get a list of servers with all relevant data
;; ( mod-time host port start-time )
;;
(define (server:get-list areapath)
(if (directory-exists? areapath)
(let ((server-logs (glob (conc areapath "/logs/server-*.log"))))
(if (null? server-logs)
'()
(let loop ((hed (car server-logs))
(tal (cdr server-logs))
(res '()))
(let* ((mod-time (file-modification-time hed))
(serv-dat (server:logf-get-start-info hed))
(serv-rec (cons mod-time serv-dat))
(new-res (cons serv-rec res)))
(if (null? tal)
new-res
(loop (car tal)(cdr tal) new-res))))))))
;; given a list of servers get a list of valid servers, i.e. at least
;; 10 seconds old, has started and is less than 1 hour old and is
;; active (i.e. mod-time < 10 seconds
;;
;; mod-time host port start-time
;;
;; sort by start-time descending. I.e. get the oldest first. Young servers will thus drop off
;; and servers should stick around for about two hours or so.
;;
(define (server:get-best srvlst)
(let ((now (current-seconds)))
(sort
(filter (lambda (rec)
(let ((start-time (list-ref rec 3))
(mod-time (list-ref rec 0)))
(print "start-time: " start-time " mod-time: " mod-time)
(and start-time mod-time
(> (- now start-time) 1) ;; been running at least 1 seconds
(< (- now mod-time) 10) ;; still alive - file touched in last 10 seconds
(< (- now start-time) 3600) ;; under one hour running time
)))
srvlst)
(lambda (a b)
(< (list-ref a 3)
(list-ref b 3))))))
(define (server:get-client-signature) ;; BB> why is this proc named "get-"? it returns nothing -- set! has not return value.
(if *my-client-signature* *my-client-signature*
(let ((sig (server:mk-signature)))
(set! *my-client-signature* sig)
*my-client-signature*)))
|
| ︙ | ︙ |
Modified tasks.scm from [af79560081] to [6f2c907335].
| ︙ | ︙ | |||
168 169 170 171 172 173 174 | (define (tasks:hostinfo-get-interface vec) (vector-ref vec 1)) (define (tasks:hostinfo-get-port vec) (vector-ref vec 2)) (define (tasks:hostinfo-get-pubport vec) (vector-ref vec 3)) (define (tasks:hostinfo-get-transport vec) (vector-ref vec 4)) (define (tasks:hostinfo-get-pid vec) (vector-ref vec 5)) (define (tasks:hostinfo-get-hostname vec) (vector-ref vec 6)) | < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < | 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 | (define (tasks:hostinfo-get-interface vec) (vector-ref vec 1)) (define (tasks:hostinfo-get-port vec) (vector-ref vec 2)) (define (tasks:hostinfo-get-pubport vec) (vector-ref vec 3)) (define (tasks:hostinfo-get-transport vec) (vector-ref vec 4)) (define (tasks:hostinfo-get-pid vec) (vector-ref vec 5)) (define (tasks:hostinfo-get-hostname vec) (vector-ref vec 6)) (define (tasks:need-server run-id) (equal? (configf:lookup *configdat* "server" "required") "yes")) ;; no elegance here ... ;; (define (tasks:kill-server hostname pid #!key (kill-switch "")) (debug:print-info 0 *default-log-port* "Attempting to kill server process " pid " on host " hostname) (setenv "TARGETHOST" hostname) (setenv "TARGETHOST_LOGF" "server-kills.log") (system (conc "nbfake kill "kill-switch" "pid)) (unsetenv "TARGETHOST_LOGF") (unsetenv "TARGETHOST")) ;;====================================================================== ;; M O N I T O R S ;;====================================================================== (define (tasks:remove-monitor-record mdb) (sqlite3:execute mdb "DELETE FROM monitors WHERE pid=? AND hostname=?;" (current-process-id) |
| ︙ | ︙ |