Megatest

;; wait for server=start-last to be three seconds old
;;
(define (server:wait-for-server-start-last-flag areapath)
  (let* ((start-flag (conc areapath "/logs/server-start-last")))
    (if (file-exists? start-flag)
	(let* ((fmodtime (file-modification-time start-flag))
	       (reftime  (+ 3 (random 5)))
	       (delta    (- (current-seconds) fmodtime)))
	  (if (> delta reftime) ;; good enough
	       (reftime  (+ 2 (random 3)))
	       (delta    (- (current-seconds) fmodtime))
	       (all-go   (> delta reftime)))
	  (if all-go
	      #t ;; (system (conc "touch " start-flag)) ;; lazy but safe
	      (begin
		(debug:print-info 0 *default-log-port* "Ready to start server, last start: "
				  fmodtime ", delta: " delta ", reftime: " reftime)
		(debug:print-info 0 *default-log-port* "Gating server start, last start: "
				  fmodtime ", delta: " delta ", reftime: " reftime ", all-go=" all-go)
		(system (conc "touch " start-flag))) ;; lazy but safe
	      (begin
		(thread-sleep! 5)
		(server:wait-for-server-start-last-flag areapath))))
	(system (conc "touch " start-flag)))))
	#;(system (conc "touch " start-flag)))))
	      

;; kind start up of servers, wait 40 seconds before allowing another server for a given
;; run-id to be launched
;;
(define (server:kind-run areapath)
  ;; look for $MT_RUN_AREA_HOME/logs/server-start-last
  ;; and wait for it to be at least 3 seconds old
  (server:wait-for-server-start-last-flag areapath)
  (if (not (server:check-if-running areapath)) ;; why try if there is already a server running?
      (let* ((last-run-dat (hash-table-ref/default *server-kind-run* areapath '(0 0))) ;; callnum, whenrun
	     (call-num     (car last-run-dat))
	     (when-run     (cadr last-run-dat))
	     (run-delay    (+ (case call-num
				((0)    0)
				((1)   20)
				((2)  300)
				(else 600))
			      (random 5)))   ;; add a small random number just in case a lot of jobs hit the work hosts simultaneously
	     (lock-file    (conc areapath "/logs/server-start.lock")))
	(if	(> (- (current-seconds) when-run) run-delay)
		(begin
		(let* ((start-flag (conc areapath "/logs/server-start-last")))
		  (common:simple-file-lock-and-wait lock-file expire-time: 15)
		  (system (conc "touch " start-flag)) ;; lazy but safe
		  (server:run areapath)
		  (thread-sleep! 2) ;; don't release the lock for at least a few seconds
		  (common:simple-file-release-lock lock-file)))
	(hash-table-set! *server-kind-run* areapath (list (+ call-num 1)(current-seconds))))))

;; this one seems to be the general entry point
;;