Megatest

Diff
Login

Differences From Artifact [db6ec670d8]:

To Artifact [a457dae400]:


89
90
91
92
93
94
95
96
97
98


99
100
101

102
103
104
105
106
107
108
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
(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:server-lock-slot mdb run-id)
  (let ((res '())
	(best #f))
    (tasks:server-clean-out-old-records-for-run-id mdb run-id)


    (tasks:server-set-available mdb run-id)
    (thread-sleep! 2) ;; Try removing this. It may not be needed.
    (tasks:server-am-i-the-server? mdb run-id)))

	
;; register that this server may come online (first to register goes though with the process)
(define (tasks:server-set-available mdb run-id)
  (sqlite3:execute 
   mdb 
   "INSERT INTO servers (pid,hostname,port,pubport,start_time,      priority,state,mt_version,heartbeat,   interface,transport,run_id)
                   VALUES(?, ?,       ?,   ?, strftime('%s','now'), ?,       ?,    ?, strftime('%s','now'),?,        ?,        ?);"
   (current-process-id)       ;; pid
   (get-host-name)            ;; hostname
   -1                         ;; port
   -1                         ;; pubport
   (random 1000)              ;; priority (used a tiebreaker on get-available)
   "available"                ;; state
   (common:version-signature) ;; mt_version
   -1                         ;; interface
   "http"                     ;; transport
   run-id
   ))











(define (tasks:server-clean-out-old-records-for-run-id mdb run-id)
  (sqlite3:execute mdb "DELETE FROM servers WHERE state in ('available','shutting-down') AND (strftime('%s','now') - start_time) > 30 AND run_id=?;" run-id)


  (sqlite3:execute mdb "DELETE FROM servers WHERE state='running' AND (strftime('%s','now') - heartbeat)  > 10 AND run_id=?;" run-id)
  )
  



(define (tasks:server-set-state! mdb server-id state)
  (sqlite3:execute mdb "UPDATE servers SET state=? WHERE id=?;" state server-id))

(define (tasks:server-delete-record! mdb server-id)
  (sqlite3:execute mdb "DELETE FROM servers WHERE id=?;" server-id))

(define (tasks:server-delete-records-for-this-pid mdb)
  (sqlite3:execute mdb "DELETE FROM servers WHERE hostname=? AND pid=?;" (get-host-name) (current-process-id)))

(define (tasks:server-set-interface-port mdb server-id interface port)
  (sqlite3:execute mdb "UPDATE servers SET interface=?,port=? WHERE id=?;" interface port server-id))
























(define (tasks:server-am-i-the-server? mdb run-id)
  (let* ((all    (tasks:server-get-servers-vying-for-run-id mdb run-id))
	 (first  (if (null? all)
		     (begin (debug:print 0 "ERROR: no servers listed, should be at least one by now.") 
			    (sqlite3:finalize! mdb)
			    (exit 1))
		     (car (db:get-rows all))))







<
<
|
>
>
|
|
|
>






|












>
>
>
>
>
>
>
>
>
>

|
>
>
|
|
|
>
>













>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>







89
90
91
92
93
94
95


96
97
98
99
100
101
102
103
104
105
106
107
108
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
(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:server-lock-slot mdb run-id)


  (tasks:server-clean-out-old-records-for-run-id mdb run-id)
  (if (< (tasks:num-in-available-state mdb run-id) 4)
      (begin 
	(tasks:server-set-available mdb run-id)
	(thread-sleep! 2) ;; Try removing this. It may not be needed.
	(tasks:server-am-i-the-server? mdb run-id))
      #f))      
	
;; register that this server may come online (first to register goes though with the process)
(define (tasks:server-set-available mdb run-id)
  (sqlite3:execute 
   mdb 
   "INSERT INTO servers (pid,hostname,port,pubport,start_time,      priority,state,mt_version,heartbeat,   interface,transport,run_id)
                   VALUES(?, ?,       ?,   ?, strftime('%s','now'), ?,       ?,    ?,-1,?,        ?,        ?);"
   (current-process-id)       ;; pid
   (get-host-name)            ;; hostname
   -1                         ;; port
   -1                         ;; pubport
   (random 1000)              ;; priority (used a tiebreaker on get-available)
   "available"                ;; state
   (common:version-signature) ;; mt_version
   -1                         ;; interface
   "http"                     ;; transport
   run-id
   ))

(define (tasks:num-in-available-state mdb run-id)
  (let ((res 0))
    (sqlite3:for-each-row
     (lambda (num-in-queue)
       (set! res num-in-queue))
     mdb
     "SELECT count(id) FROM servers WHERE run_id=?;"
     run-id)
    res))

(define (tasks:server-clean-out-old-records-for-run-id mdb run-id)
  (sqlite3:execute mdb "DELETE FROM servers WHERE state in ('available','shutting-down') AND (strftime('%s','now') - start_time) > 10 AND run_id=?;" run-id))

(define (tasks:server-force-clean-running-records-for-run-id mdb run-id)
  (sqlite3:execute mdb "DELETE FROM servers WHERE state = 'running' AND run_id=?;" run-id))

(define (tasks:server-force-clean-run-record mdb run-id iface port)
  (sqlite3:execute mdb "DELETE FROM servers WHERE state = 'running' AND run_id=? AND interface=? AND port=?;"
		   run-id iface port))

(define (tasks:server-set-state! mdb server-id state)
  (sqlite3:execute mdb "UPDATE servers SET state=? WHERE id=?;" state server-id))

(define (tasks:server-delete-record! mdb server-id)
  (sqlite3:execute mdb "DELETE FROM servers WHERE id=?;" server-id))

(define (tasks:server-delete-records-for-this-pid mdb)
  (sqlite3:execute mdb "DELETE FROM servers WHERE hostname=? AND pid=?;" (get-host-name) (current-process-id)))

(define (tasks:server-set-interface-port mdb server-id interface port)
  (sqlite3:execute mdb "UPDATE servers SET interface=?,port=? WHERE id=?;" interface port server-id))

(define (tasks:server-get-next-port mdb)
  (let ((res         #f)
	(port-param  (if (and (args:get-arg "-port")
			      (string->number (args:get-arg "-port")))
			 (string->number (args:get-arg "-port"))
			 #f))
	(config-port (if (and (config-lookup  *configdat* "server" "port")
			      (string->number (config-lookup  *configdat* "server" "port")))
			 (string->number (config-lookup  *configdat* "server" "port"))
			 #f)))
    (sqlite3:for-each-row
     (lambda (port)
       (set! res (+ port 1))) ;; set to next
     mdb
     "SELECT max(port) FROM servers;")
    (cond
     ((and port-param res)   (if (> res port-param) res port-param))
     (port-param             port-param)
     ((and config-port res)  (if (> res config-port) res config-port))
     (config-port            config-port)
     ((and res (> res 8080)) res)
     (else                   (+ 5000 (random 1001))))))

(define (tasks:server-am-i-the-server? mdb run-id)
  (let* ((all    (tasks:server-get-servers-vying-for-run-id mdb run-id))
	 (first  (if (null? all)
		     (begin (debug:print 0 "ERROR: no servers listed, should be at least one by now.") 
			    (sqlite3:finalize! mdb)
			    (exit 1))
		     (car (db:get-rows all))))
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
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
     (lambda (a . b)
       (set! res (cons (apply vector a b) res)))
     mdb
     (conc "SELECT " selstr " FROM servers WHERE run_id=? ORDER BY start_time DESC;")
     run-id)
    (vector header res)))

(define (tasks:server-update-heartbeat mdb server-id)
  (debug:print-info 1 "Heart beat update of server id=" server-id)
  (handle-exceptions
   exn
   (begin
     (debug:print 0 "WARNING: probable timeout on monitor.db access")
     (thread-sleep! 1)
     (tasks:server-update-heartbeat mdb server-id))
   (sqlite3:execute mdb "UPDATE servers SET heartbeat=strftime('%s','now') WHERE id=?;" server-id)))

;; alive servers keep the heartbeat field upto date with seconds every 6 or so seconds
(define (tasks:server-alive? mdb server-id #!key (iface #f)(hostname #f)(port #f)(pid #f))
  (let* ((server-id  (if server-id 
			 server-id
			 (tasks:server-get-server-id mdb hostname iface port pid)))
	 (heartbeat-delta 99e9))
    (sqlite3:for-each-row
     (lambda (delta)
       (set! heartbeat-delta delta))
     mdb "SELECT strftime('%s','now')-heartbeat FROM servers WHERE id=?;" server-id)
    (< heartbeat-delta 10)))

(define (tasks:get-server mdb run-id)
  (let ((res  #f)
	(best #f))
    (sqlite3:for-each-row
     (lambda (id interface port pubport transport pid hostname)
       (set! res (vector id interface port pubport transport pid hostname)))
     mdb


     "SELECT id,interface,port,pubport,transport,pid,hostname FROM servers
          WHERE strftime('%s','now')-heartbeat < 10 
          AND mt_version=? AND run_id=? AND state='running'
          ORDER BY start_time DESC LIMIT 1;" (common:version-signature) run-id)
    res))

;; (define (tasks:get-all-servers mdb)
;;   (let ((res  '()))
;;     (sqlite3:for-each-row
;;      (lambda (id interface port pubport transport pid hostname)
;;        (set! res (cons (vector id interface port pubport transport pid hostname) res)))
;;      mdb
;;      "SELECT id,interface,port,pubport,transport,pid,hostname FROM servers
;;           WHERE strftime('%s','now')-heartbeat < 10 
;;           AND mt_version=? 
;;           ORDER BY start_time DESC;" (common:version-signature))
;;     res))

(define (tasks:get-all-servers mdb)
  (let ((res '()))
    (sqlite3:for-each-row
     (lambda (id pid hostname interface port pubport start-time priority state mt-version last-update transport)
       (set! res (cons (vector id pid hostname interface port pubport start-time priority state mt-version last-update transport) res)))
     mdb
     "SELECT id,pid,hostname,interface,port,pubport,start_time,priority,state,mt_version,strftime('%s','now')-heartbeat AS last_update,transport FROM servers ORDER BY start_time DESC;")
    res))

(define (tasks:kill-server status hostname port pid transport)
  (debug:print-info 1 "Removing defunct server record for " hostname ":" port)
  (if port
      (open-run-close tasks:server-deregister tasks:open-db hostname port: port)
      (open-run-close tasks:server-deregister tasks:open-db hostname pid:  pid))
  (if status ;; #t means alive
      (begin
	(if (equal? hostname (get-host-name))
	    (handle-exceptions
	     exn
	     (debug:print-info 0 "server may or may not be dead, check for megatest -server running as pid " pid "\n"
			       "  EXCEPTION: " ((condition-property-accessor 'exn 'message) exn))
	     (debug:print 1 "Sending signal/term to " pid " on " hostname)
	     (process-signal pid signal/term)
	     (thread-sleep! 5) ;; give it five seconds to die peacefully then do a brutal kill
	     ;;(process-signal pid signal/kill)
	     ) ;; local machine, send sig term
	    (begin
	      ;;(debug:print-info 1 "Stopping remote servers not yet supported."))))
	      (debug:print-info 1 "Telling alive server on " hostname ":" port " to commit servercide")
	      (let ((serverdat (list hostname port)))
	      	(case (if (string? transport) (string->symbol transport) transport)
	      	  ((http)(http-transport:client-connect hostname port))
	      	  (else  (debug:print "ERROR: remote stopping servers of type " transport " not supported yet")))
	      	(cdb:kill-server serverdat pid)))))    ;; remote machine, try telling server to commit suicide
      (begin
	(if status 
	    (if (equal? hostname (get-host-name))
		(begin
		  (debug:print-info 1 "Sending signal/term to " pid " on " hostname)
		  (process-signal pid signal/term)  ;; local machine, send sig term







<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<







>
>

<
|
|


<
<
<
<
<
<
<
<
<
<
<
<









|




















<
|
<







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

256
257
258
259
260
261
262
     (lambda (a . b)
       (set! res (cons (apply vector a b) res)))
     mdb
     (conc "SELECT " selstr " FROM servers WHERE run_id=? ORDER BY start_time DESC;")
     run-id)
    (vector header res)))























(define (tasks:get-server mdb run-id)
  (let ((res  #f)
	(best #f))
    (sqlite3:for-each-row
     (lambda (id interface port pubport transport pid hostname)
       (set! res (vector id interface port pubport transport pid hostname)))
     mdb
     ;; removed:
     ;; strftime('%s','now')-heartbeat < 10 AND mt_version = ?
     "SELECT id,interface,port,pubport,transport,pid,hostname FROM servers

          WHERE run_id=? AND state='running'
          ORDER BY start_time DESC LIMIT 1;" run-id) ;; (common:version-signature) run-id)
    res))













(define (tasks:get-all-servers mdb)
  (let ((res '()))
    (sqlite3:for-each-row
     (lambda (id pid hostname interface port pubport start-time priority state mt-version last-update transport)
       (set! res (cons (vector id pid hostname interface port pubport start-time priority state mt-version last-update transport) res)))
     mdb
     "SELECT id,pid,hostname,interface,port,pubport,start_time,priority,state,mt_version,strftime('%s','now')-heartbeat AS last_update,transport FROM servers ORDER BY start_time DESC;")
    res))

(define (tasks:kill-server status hostname port pid)
  (debug:print-info 1 "Removing defunct server record for " hostname ":" port)
  (if port
      (open-run-close tasks:server-deregister tasks:open-db hostname port: port)
      (open-run-close tasks:server-deregister tasks:open-db hostname pid:  pid))
  (if status ;; #t means alive
      (begin
	(if (equal? hostname (get-host-name))
	    (handle-exceptions
	     exn
	     (debug:print-info 0 "server may or may not be dead, check for megatest -server running as pid " pid "\n"
			       "  EXCEPTION: " ((condition-property-accessor 'exn 'message) exn))
	     (debug:print 1 "Sending signal/term to " pid " on " hostname)
	     (process-signal pid signal/term)
	     (thread-sleep! 5) ;; give it five seconds to die peacefully then do a brutal kill
	     ;;(process-signal pid signal/kill)
	     ) ;; local machine, send sig term
	    (begin
	      ;;(debug:print-info 1 "Stopping remote servers not yet supported."))))
	      (debug:print-info 1 "Telling alive server on " hostname ":" port " to commit servercide")
	      (let ((serverdat (list hostname port)))

		(http-transport:client-connect hostname port)

	      	(cdb:kill-server serverdat pid)))))    ;; remote machine, try telling server to commit suicide
      (begin
	(if status 
	    (if (equal? hostname (get-host-name))
		(begin
		  (debug:print-info 1 "Sending signal/term to " pid " on " hostname)
		  (process-signal pid signal/term)  ;; local machine, send sig term