Megatest

Diff
Login

Differences From Artifact [6da0e18c74]:

To Artifact [c243a4652b]:


89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
(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! 0.2) ;; Try removing this. It may not be needed.
	(tasks:server-am-i-the-server? mdb run-id))
      #f))
	







|







89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
(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 " tasks:server-lock-slot")
  (if (< (tasks:num-in-available-state mdb run-id) 4)
      (begin 
	(tasks:server-set-available mdb run-id)
	(thread-sleep! 0.2) ;; Try removing this. It may not be needed.
	(tasks:server-am-i-the-server? mdb run-id))
      #f))
	
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

(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) > 100 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))







|



|
|
>

|
|
>

|
|
|

|
|
>

|
>
>
>
>
|
|
>
|
|


|

>
>

>
|
>
>
>
|
|
|
|
|
|
|
|
>


|

|



|
|
>
>
>
>
|
>
>
|







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

(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=? AND state = 'available';"
     run-id)
    res))

(define (tasks:server-clean-out-old-records-for-run-id mdb run-id tag)
  (sqlite3:execute mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE state in ('available','shutting-down') AND (strftime('%s','now') - start_time) > 50 AND run_id=?;"
		   (conc "defunct" tag) run-id))

(define (tasks:server-force-clean-running-records-for-run-id mdb run-id tag)
  (sqlite3:execute mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE state = 'running' AND run_id=?;"
		   (conc "defunct" tag) run-id))

(define (tasks:server-force-clean-run-record mdb run-id iface port tag)
  (sqlite3:execute mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE state = 'running' AND run_id=? AND interface=? AND port=?;"
		   (conc "defunct" tag) run-id iface port))

(define (tasks:server-delete-records-for-this-pid mdb tag)
  (sqlite3:execute mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE hostname=? AND pid=?;"
		   (conc "defunct" tag) (get-host-name) (current-process-id)))

(define (tasks:server-delete-record mdb server-id tag) 
  (sqlite3:execute mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE id=?;"
		   (conc "defunct" tag) server-id)
  ;; use this opportuntity to clean out records over one month old or over 10 minutes old with port = -1 (i.e. a never used placeholder)
  (sqlite3:execute mdb "DELETE FROM servers WHERE state not in ('running','shutting-down') AND (strftime('%s','now') - start_time) > 2628000;")
  (sqlite3:execute mdb "DELETE FROM servers WHERE state like 'defunct%' AND port=-1 AND (strftime('%s','now') - start_time) > 600;")
  )

(define (tasks:server-set-state! mdb server-id state)
  (sqlite3:execute mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE id=?;" state server-id))

(define (tasks:server-set-interface-port mdb server-id interface port)
  (sqlite3:execute mdb "UPDATE servers SET interface=?,port=?,heartbeat=strftime('%s','now') WHERE id=?;" interface port server-id))

;; Get random port not used in long time
;;
(define (tasks:server-get-next-port mdb)
  (let* ((lownum        30000)
	(highnum        64000)
	(used-ports     '())
	(get-rand-port  (lambda ()
			  (+ lownum (random (- highnum lownum)))))
	(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! used-ports (cons port used-ports)))
     mdb
     "SELECT 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)
     (else
      (let loop ((port     (get-rand-port))
		 (remtries 100))
	(if (member port used-ports)
	    (if (> remtries 0)
		(loop (get-rand-port)(- remtries 1))
		(get-rand-port))
	    port))))))

(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))
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
   (let* ((header (list "id" "hostname" "pid" "interface" "port" "pubport" "state" "run_id" "priority" "start_time"))
	  (selstr (string-intersperse header ","))
	  (res    '()))
    (sqlite3:for-each-row
     (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))







|




















|
>
|

|







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
263
   (let* ((header (list "id" "hostname" "pid" "interface" "port" "pubport" "state" "run_id" "priority" "start_time"))
	  (selstr (string-intersperse header ","))
	  (res    '()))
    (sqlite3:for-each-row
     (lambda (a . b)
       (set! res (cons (apply vector a b) res)))
     mdb
     (conc "SELECT " selstr " FROM servers WHERE run_id=? AND state in ('available','running') 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 run-id)
       ;;                       0   1        2         3    4       5          6        7     8          9          10        11     12
       (set! res (cons (vector id pid hostname interface port pubport start-time priority state mt-version last-update transport run-id) res)))
     mdb
     "SELECT id,pid,hostname,interface,port,pubport,start_time,priority,state,mt_version,strftime('%s','now')-heartbeat AS last_update,transport,run_id FROM servers WHERE state NOT LIKE 'defunct%' 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))
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
	     (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







|







273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
	     (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)))
		(hash-table-set! *runremote* run-id (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