Megatest

(declare (unit http-transport))

(declare (uses common))
(declare (uses db))
(declare (uses tests))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.
(declare (uses server))
(declare (uses daemon))

(include "common_records.scm")
(include "db_records.scm")

(define (http-transport:make-server-url hostport)
  (if (not hostport)
      #f

(define (http-transport:try-start-server ipaddrstr portnum)
  (handle-exceptions
   exn
   (begin
     (print-error-message exn)
     (if (< portnum 9000)
	 (begin 
	   (debug:print 0 "WARNING: failed to start on portnum: " portnum ", trying next port")
	   (thread-sleep! 0.1)
	   ;; (open-run-close tasks:remove-server-records tasks:open-db)
	   (open-run-close tasks:server-delete tasks:open-db ipaddrstr portnum)
	   (http-transport:try-start-server ipaddrstr (+ portnum 1)))
	 (print "ERROR: Tried and tried but could not start the server")))
   ;; any error in following steps will result in a retry
   (set! *runremote* (list ipaddrstr portnum))

	(begin
	  (debug:print-info 2 "Logged in and connected to " iface ":" port)
	  (set! *runremote* serverdat)
	  serverdat)
	(begin
	  (debug:print-info 2 "Failed to login or connect to " iface ":" port)
	  (set! *runremote* #f)
	  (set! *transport-type* 'fs)
	  #f))))


;; 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)

                              (begin
                                (sleep 4)
                                (loop))))))
         (iface       (car server-info))
         (port        (cadr server-info))
         (last-access 0)
	 (tdb         (tasks:open-db))
	 (spid        ;;(open-run-close tasks:server-get-server-id tasks:open-db #f iface port #f))
	   (tasks:server-get-server-id tdb #f iface port #f))
	 (server-timeout (let ((tmo (config-lookup  *configdat* "server" "timeout")))
			   (if (and (string? tmo)
				    (string->number tmo))
			       (* 60 60 (string->number tmo))
			       ;; default to three days
			       (* 3 24 60 60)))))
    (debug:print-info 2 "server-timeout: " server-timeout ", server pid: " spid " on " iface ":" port)
    (let loop ((count 0))
      (thread-sleep! 4) ;; no need to do this very often
      ;; NB// sync currently does NOT return queue-length
      (let () ;; (queue-len (cdb:client-call server-info 'sync #t 1)))
      ;; (print "Server running, count is " count)
        (if (< count 1) ;; 3x3 = 9 secs aprox
            (loop (+ count 1)))
        
	;; Check that iface and port have not changed (can happen if server port collides)
	(mutex-lock! *heartbeat-mutex*)
	(set! sdat *runremote*)
	(mutex-unlock! *heartbeat-mutex*)

	(if (or (not (equal? sdat (list iface port)))
		(not spid))
	    (begin 
	      (debug:print-info 0 "interface changed, refreshing iface and port info")
	      (set! iface (car sdat))
	      (set! port  (cadr sdat))
	      (set! spid  (tasks:server-get-server-id tdb #f iface port #f))))

        ;; NOTE: Get rid of this mechanism! It really is not needed...
        ;; (open-run-close tasks:server-update-heartbeat tasks:open-db spid)
        (tasks:server-update-heartbeat tdb spid)
      
        ;; (if ;; (or (> numrunning 0) ;; stay alive for two days after last access
        (mutex-lock! *heartbeat-mutex*)
        (set! last-access *last-db-access*)
        (mutex-unlock! *heartbeat-mutex*)
	;; (debug:print 11 "last-access=" last-access ", server-timeout=" server-timeout)
        (if (> (+ last-access server-timeout)
               (current-seconds))
            (begin
              (debug:print-info 2 "Server continuing, seconds since last db access: " (- (current-seconds) last-access))
              (loop 0))
            (begin
              (debug:print-info 0 "Starting to shutdown the server.")
              ;; need to delete only *my* server entry (future use)
              (set! *time-to-exit* #t)
              (open-run-close tasks:server-deregister-self tasks:open-db (get-host-name))
              (thread-sleep! 1)
              (debug:print-info 0 "Max cached queries was    " *max-cache-size*)
	      (debug:print-info 0 "Number of cached writes   " *number-of-writes*)
	      (debug:print-info 0 "Average cached write time "
				(if (eq? *number-of-writes* 0)
				    "n/a (no writes)"
				    (/ *writes-total-delay*

	      (thread-start! th3)
	      (thread-start! th1)
	      (set! *didsomething* #t)
	      (thread-join! th2))
	    (debug:print 0 "ERROR: Failed to setup for megatest")))
    (exit)))

;; (use trace)
;; (trace http-transport:keep-running 
;;        tasks:server-update-heartbeat
;;        tasks:server-get-server-id)
;;        tasks:get-best-server
;;        http-transport:run
;;        http-transport:launch
;;        http-transport:try-start-server
;;        http-transport:client-send-receive
;;        http-transport:make-server-url
;;        tasks:server-register
;;        tasks:server-delete
;;        start-server
;;        hostname->ip
;;        with-input-from-request
;;        tasks:server-deregister-self)

(define (http-transport:server-signal-handler signum)
  (handle-exceptions
   exn
   (debug:print " ... exiting ...")
   (let ((th1 (make-thread (lambda ()
			     (thread-sleep! 1))
			     ;; (if (not *received-response*)

;; Always use two digit decimal
;; 1.01, 1.02...1.10,1.11 ... 1.99,2.00..

(declare (unit megatest-version))

(define megatest-version 1.5416)

(define (tasks:open-db)
  (let* ((dbpath  (conc *toppath* "/monitor.db"))
	 (exists  (file-exists? dbpath))
	 (mdb     (sqlite3:open-database dbpath)) ;; (never-give-up-open-db dbpath))
	 (handler (make-busy-timeout 36000)))
    (sqlite3:set-busy-handler! mdb handler)
    (sqlite3:execute mdb (conc "PRAGMA synchronous = 1;"))
    (if (not exists)
	(begin
	  (sqlite3:execute mdb "CREATE TABLE IF NOT EXISTS tasks_queue (id INTEGER PRIMARY KEY,
                                action TEXT DEFAULT '',
                                owner TEXT,
                                state TEXT DEFAULT 'new',
                                target TEXT DEFAULT '',

       (begin
	 (debug:print 0 "ERROR: tasks:server-get-server-id needs (hostname and pid) OR (iface and port) OR (hostname and port)")
	 "SELECT id FROM servers WHERE pid=-999;")))
     (if hostname hostname iface)(if pid pid port))
    res))

(define (tasks:server-update-heartbeat mdb server-id)
  (debug:print-info 0 "Heart beat update of server id=" 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)))

#!/bin/bash

prefix=$1
cmd=$2

echo "#!/bin/bash"
if [ "$LD_LIBRARY_PATH" != "" ];then
  echo "INFO: Using LD_LIBRARY_PATH=$LD_LIBRARY_PATH" >&2
  echo "export LD_LIBRARY_PATH=$LD_LIBRARY_PATH"
else
  echo "INFO: LD_LIBRARY_PATH not set" >&2
fi

fullcmd="$prefix/bin/$cmd"

echo "$fullcmd \"\$@\""