Megatest

(on-exit (lambda ()
	   (let ((tdb (tasks:open-db)))
	     ;; (print "On-exit called")
	     (tasks:remove-monitor-record tdb)
	     (sqlite3:finalize! tdb))))

(define (gui-monitor db)
  (let ((keys (rdb:get-keys db))
  (let ((keys (db:get-keys db))
	(tdb  (tasks:open-db)))
    (tasks:register-monitor db tdb) ;;; let the other monitors know we are here
    (control-panel db tdb keys)
    ;(tasks:remove-monitor-record db)
    ;(sqlite3:finalize! db)
   ))
	

	(db:initialize db))
    (db:set-sync db)
    db))

;; keeping it around for debugging purposes only
(define (open-run-close-no-exception-handling  proc idb . params)
  (debug:print-info 11 "open-run-close-no-exception-handling START given a db=" (if idb "yes " "no ") ", params=" params)
  (let* ((db   (if idb idb (open-db)))
  (let* ((db   (if idb 
		   (if (procedure? idb)
		       (idb)
		       idb)
		   (open-db)))
	 (res #f))
    (set! res (apply proc db params))
    (if (not idb)(sqlite3:finalize! db))
    (debug:print-info 11 "open-run-close-no-exception-handling END" )
    res))

(define (open-run-close-exception-handling proc idb . params)
  (debug:print-info 11 "open-run-close-exception-handling START, idb=" idb ", params=" params)
  (let ((runner (lambda ()
		  (let* ((db   (if idb idb (open-db)))
			 (res #f))
		    (set! res (apply proc db params))
		    (if (not idb)(sqlite3:finalize! db))
		    (debug:print-info 11 "open-run-close-no-exception-handling END" )
		    res))))
    (handle-exceptions
     exn
     (begin
       (debug:print 0 "EXCEPTION: database probably overloaded?")
       (debug:print 0 "  " ((condition-property-accessor 'exn 'message) exn))
       (print-call-chain)
       (thread-sleep! (random 120))
       (debug:print-info 0 "trying db call one more time....")
  (handle-exceptions
   exn
   (begin
     (debug:print 0 "EXCEPTION: database probably overloaded?")
     (debug:print 0 "  " ((condition-property-accessor 'exn 'message) exn))
     (print-call-chain)
     (thread-sleep! (random 120))
     (debug:print-info 0 "trying db call one more time....")
       (runner))
     (runner))))
     (apply open-run-close-no-exception-handling proc idb params))
   (apply open-run-close-no-exception-handling proc idb params)))

(define open-run-close open-run-close-exception-handling)

(define *global-delta* 0)
(define *last-global-delta-printed* 0)

(define (open-run-close-measure  proc idb . params)

      (save-environment-as-files (args:get-arg "-env2file"))
      (set! *didsomething* #t)))

;;======================================================================
;; Start the server - can be done in conjunction with -runall or -runtests (one day...)
;;   we start the server if not running else start the client thread
;;======================================================================

(if (args:get-arg "-server")
    (server:launch)
    (server:launch))

(if (or (let ((res #f))
	  (for-each
	   (lambda (key)
	     (if (args:get-arg key)(set! res #t)))
	   (list "-h" "-version" "-gen-megatest-area" "-gen-megatest-test"))
	  res)
	(eq? (length (hash-table-keys args:arg-hash)) 0))
    (debug:print-info 1 "No server needed")
    (server:client-launch))

;;======================================================================
;; Remove old run(s)
;;======================================================================

;; since several actions can be specified on the command line the removal

(import (prefix sqlite3 sqlite3:))

(declare (unit server))

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

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

(define (server:make-server-url hostport)
  (if (not hostport)
      #f
      (conc "tcp://" (car hostport) ":" (cadr hostport))))

(define (server:run hostn)
  (debug:print 0 "Attempting to start the server ...")
  (if (not *toppath*)(setup-for-run))
  (let ((host:port      (open-run-close db:get-var #f "SERVER"))) ;; do whe already have a server running?
  (let* ((hostport      (open-run-close tasks:get-best-server tasks:open-db)) ;; do whe already have a server running?
	 (host:port (server:make-server-url hostport)))
    (if host:port 
	(begin
	  (debug:print 0 "NOTE: server already running.")
	  (if (server:client-setup)
	      (begin 
		(debug:print-info 0 "Server is alive, not starting another")
		(debug:print-info 0 "Server is alive, not starting another"))
		;;(exit)
		)
	      (begin
		(debug:print-info 0 "Server is dead, removing flag and trying again")
		(open-run-close db:del-var #f "SERVER")
		(server:run hostn))))
		(debug:print-info 0 "Server is dead, deregistering it, please try again")
		(open-run-close tasks:server-deregister tasks:open-db (car hostport) port: (cadr port))
		;; (server:run hostn)
		(debug:print 0 "WOULD NORMALLY START ANOTHER SERVER HERE")
		)
	      )
	  )
	(let* ((zmq-socket     #f)
	       (hostname       (if (string=? "-" hostn)
				   (get-host-name) 
				   hostn))
	       (ipaddrstr      (let ((ipstr (if (string=? "-" hostn)
						(string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
						#f)))
				 (if ipstr ipstr hostname))))
	  (set! zmq-socket (server:find-free-port-and-open ipaddrstr zmq-socket 5555))
	  (set! zmq-socket (server:find-free-port-and-open ipaddrstr zmq-socket 5555 0))
	  (set! *cache-on* #t)
	  
	  ;; what to do when we quit
	  ;;
	  (on-exit (lambda ()
		     (open-run-close db:del-var #f "SERVER")
		     (open-run-close tasks:server-deregister-self tasks:open-db)
		     (let loop () 
		       (let ((queue-len 0))
			 (thread-sleep! (random 5))
			 (mutex-lock! *incoming-mutex*)
			 (set! queue-len (length *incoming-data*))
			 (mutex-unlock! *incoming-mutex*)
			 (if (> queue-len 0)

		(open-run-close db:del-var #f "SERVER")
		(thread-sleep! 10)
		(debug:print-info 0 "Max cached queries was " *max-cache-size*)
		(debug:print-info 0 "Server shutdown complete. Exiting")
		;; (exit)))
		))))))

(define (server:find-free-port-and-open host s port)
(define (server:find-free-port-and-open host s port trynum)
  (let ((s (if s s (make-socket 'rep)))
	(p (if (number? port) port 5555)))
    (handle-exceptions
     exn
     (begin
       (debug:print 0 "Failed to bind to port " p ", trying next port")
       (debug:print 0 "   EXCEPTION: " ((condition-property-accessor 'exn 'message) exn))
       (if (< trynum 100)
       (server:find-free-port-and-open host s (+ p 1)))
	   (server:find-free-port-and-open host s (+ p 1) (+ trynum 1))
	   (debug:print-info 0 "Tried ports from " (- p trynum) " to " p 
			     " but all were in use. Please try a different port range by starting the server with parameter \" -port N\" where N is the starting port number to use")))
     (let ((zmq-url (conc "tcp://" host ":" p)))
       (print "Trying to start server on " zmq-url)
       (bind-socket s zmq-url)
       (set! *runremote* #f)
       (debug:print 0 "Server started on " zmq-url)
       (open-run-close db:set-var #f "SERVER" zmq-url)
       (open-run-close tasks:server-register tasks:open-db (current-process-id) host p 0 'live)
       s))))

(define (server:client-setup)
  (if (not *toppath*)(setup-for-run))
  (let* ((hostinfo   (open-run-close db:get-var #f "SERVER"))
  (let* ((hostinfo   (open-run-close tasks:get-best-server tasks:open-db))
	 (zmq-socket (make-socket 'req)))
    (if hostinfo
	(begin
	  (debug:print-info 2 "Setting up to connect to " hostinfo)
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print 0 "ERROR: Failed to open a connection to the server at: " hostinfo)
	     (debug:print 0 "   EXCEPTION: " ((condition-property-accessor 'exn 'message) exn))
	     (debug:print 0 "   perhaps jobs killed with -9? Removing server records")
	     (open-run-close db:del-var #f "SERVER")
	     (exit)
	     (open-run-close tasks:server-deregister tasks:open-db (car hostinfo) port: (cadr hostinfo))
	     ;; (exit) ;; why forced exit?
	     #f)
	   (let ((connect-ok #f))
	     (connect-socket zmq-socket hostinfo)
	   (let ((connect-ok #f)
		 (conurl     (server:make-server-url hostinfo)))
	     (connect-socket zmq-socket conurl)
	     (set! connect-ok (cdb:client-call zmq-socket 'login #t *toppath*))
	     (if connect-ok
		 (begin
		   (debug:print-info 2 "Logged in and connected to " hostinfo)
		   (debug:print-info 2 "Logged in and connected to " conurl)
		   (set! *runremote* zmq-socket)
		   #t)
		 (begin
		   (debug:print-info 2 "Failed to login or connect to " hostinfo)
		   (debug:print-info 2 "Failed to login or connect to " conurl)
		   (set! *runremote* #f)
		   #f)))))
	(begin
	  (debug:print-info 0 "NO SERVER RUNNING! PLEASE START ONE! E.g. \"megatest -server - &\"")
	  (debug:print-info 2 "No server available, attempting to start one...")
	  (system (conc "megatest -server - " (if (args:get-arg "-debug")
						  (conc "-debug " (args:get-arg "-debug"))
						  "")
			" &"))
	  (sleep 5)
	  (server:client-setup)))))
	;;   (debug:print-info 2 "No server available, attempting to start one...")
	;;   (system (conc (car (argv)) " -server - " (if (args:get-arg "-debug")
	;; 					  (conc "-debug " (args:get-arg "-debug"))
	;; 					  "")
	;; 		" &"))
	  ;; (sleep 5)
	  ;; (server:client-setup)
	  ))))

(define (server:launch)
  (let* ((toppath (setup-for-run)))
    (debug:print-info 0 "Starting the standalone server")
    (if *toppath* 
	(let* ((th2 (make-thread (lambda ()
				   (server:run (args:get-arg "-server")))))

;;======================================================================
;; Tasks db
;;======================================================================

(define (tasks:open-db)
  (let* ((dbpath  (conc *toppath* "/monitor.db"))
	 (exists  (file-exists? dbpath))
	 (tdb     (sqlite3:open-database dbpath)) ;; (never-give-up-open-db dbpath))
	 (mdb     (sqlite3:open-database dbpath)) ;; (never-give-up-open-db dbpath))
	 (handler (make-busy-timeout 36000)))
    (sqlite3:set-busy-handler! tdb handler)
    (sqlite3:set-busy-handler! mdb handler)
    (if (not exists)
	(begin
	  (sqlite3:execute tdb "CREATE TABLE IF NOT EXISTS tasks_queue (id INTEGER PRIMARY KEY,
	  (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 '',
                                name TEXT DEFAULT '',
                                test TEXT DEFAULT '',
                                item TEXT DEFAULT '',
                                keylock TEXT,
                                params TEXT,
                                creation_time TIMESTAMP,
                                execution_time TIMESTAMP);")
	  (sqlite3:execute tdb "CREATE TABLE IF NOT EXISTS monitors (id INTEGER PRIMARY KEY,
	  (sqlite3:execute mdb "CREATE TABLE IF NOT EXISTS monitors (id INTEGER PRIMARY KEY,
                                pid INTEGER,
                                start_time TIMESTAMP,
                                last_update TIMESTAMP,
                                hostname TEXT,
                                username TEXT,
                               CONSTRAINT monitors_constraint UNIQUE (pid,hostname));")))
    tdb))
                               CONSTRAINT monitors_constraint UNIQUE (pid,hostname));")
	  (sqlite3:execute mdb "CREATE TABLE IF NOT EXISTS servers (id INTEGER PRIMARY KEY,
                                  pid INTEGER,
                                  hostname TEXT,
                                  port INTEGER,
                                  start_time TIMESTAMP,
                                  priority INTEGER,
                                  state TEXT,
                               CONSTRAINT servers_constraint UNIQUE (pid,hostname));")
	  (sqlite3:execute mdb "CREATE TABLE IF NOT EXISTS clients (id INTEGER PRIMARY KEY,
                                  server_id INTEGER,
                                  pid INTEGER,
                                  hostname TEXT,
                                  cmdline TEXT,
                                  login_time TIMESTAMP,
                                  logout_time TIMESTAMP DEFAULT -1,
                                CONSTRAINT clients_constraint UNIQUE (pid,hostname));")
                                  
	  ))
    mdb))
    
;;======================================================================
;; Server and client management
;;======================================================================

;; state: 'live, 'shutting-down, 'dead
(define (tasks:server-register mdb pid hostname port priority state)
  (sqlite3:execute 
   mdb 
   "INSERT OR REPLACE INTO servers (pid,hostname,port,start_time,priority,state) VALUES(?,?,?,strftime('%s','now'),?,?);"
   pid hostname port priority (conc state)))

(define (tasks:server-deregister mdb hostname #!key (port #f)(pid #f))
  (if pid
      (sqlite3:execute mdb "DELETE FROM servers WHERE  hostname=? AND pid=?;" hostname pid)
      (if port
	  (sqlite3:execute mdb "DELETE FROM servers WHERE  hostname=? AND port=?;" hostname port)
	  (debug:print 0 "ERROR: tasks:server-deregister called with neither pid nor port specified"))))

(define (tasks:server-deregister-self mdb)
  (tasks:server-deregister mdb pid: (current-process-id) (get-host-name)))

(define (tasks:server-get-server-id mdb)
  ;; dunno yet
  0)

(define (tasks:client-register mdb pid hostname cmdline)
  (sqlite3:execute
   mdb
   "INSERT OR REPLACE INTO clients (server_id,pid,hostname,cmdline,login_time) VALUES(?,?,?,?,strftime('%s','now'));")
  (tasks:server-get-server-id mdb)
  pid hostname cmdline)

(define (tasks:client-logout mdb pid hostname cmdline)
  (sqlite3:execute
   mdb
   "UPDATE clients SET logout_time=strftime('%s','now') WHERE pid=? AND hostname=? AND cmdline=?;"
   pid hostname cmdline))

(define (tasks:get-logged-in-clients mdb server-id)
  (let ((res '()))
    (sqlite3:for-each-row 
     (lambda (id server-id pid hostname cmdline login-time logout-time)
       (set! res (cons (vector id server-id pid hostname cmdline login-time lougout-time) res)))
     mdb
     "SELECT id,server_id,pid,hostname,cmdline,login_time,logout_time FROM clients WHERE server_id=?;"
     server-id)))

(define (tasks:have-clients? mdb server-id)
  (null? (tasks:get-logged-in-clients mdb server-id)))

(define (tasks:get-best-server mdb)
  (let ((res #f))
    (sqlite3:for-each-row
     (lambda (id hostname port)
       (set! res (list hostname port)))
     mdb
     "SELECT id,hostname,port FROM servers WHERE state='live' ORDER BY start_time DESC LIMIT 1;")
    res))


;;======================================================================
;; Tasks and Task monitors
;;======================================================================


;;======================================================================
;; Tasks
;;======================================================================



;;======================================================================
;; Task Monitors
;;======================================================================

(define (tasks:register-monitor db tdb)
(define (tasks:register-monitor db mdb)
  (let* ((pid (current-process-id))
	 (hostname (get-host-name))
	 (userinfo (user-information (current-user-id)))
	 (username (car userinfo)))
    (print "Register monitor, pid: " pid ", hostname: " hostname ", username: " username)
    (sqlite3:execute tdb "INSERT INTO monitors (pid,start_time,last_update,hostname,username) VALUES (?,strftime('%s','now'),strftime('%s','now'),?,?);"
    (sqlite3:execute mdb "INSERT INTO monitors (pid,start_time,last_update,hostname,username) VALUES (?,strftime('%s','now'),strftime('%s','now'),?,?);"
		     pid hostname username)))

(define (tasks:get-num-alive-monitors tdb)
(define (tasks:get-num-alive-monitors mdb)
  (let ((res 0))
    (sqlite3:for-each-row 
     (lambda (count)
       (set! res count))
     tdb
     mdb
     "SELECT count(id) FROM monitors WHERE last_update < (strftime('%s','now') - 300) AND username=?;"
     (car (user-information (current-user-id))))
    res))

;; register a task
(define (tasks:add tdb action owner target runname test item params)
  (sqlite3:execute tdb "INSERT INTO tasks_queue (action,owner,state,target,name,test,item,params,creation_time,execution_time)
(define (tasks:add mdb action owner target runname test item params)
  (sqlite3:execute mdb "INSERT INTO tasks_queue (action,owner,state,target,name,test,item,params,creation_time,execution_time)
                       VALUES (?,?,'new',?,?,?,?,?,strftime('%s','now'),0);" 
		   action
		   owner
		   target
		   runname
		   test
		   item
		   (if params params "")))

(define (keys:key-vals-hash->target keys key-params)
  (let ((tmp (hash-table-ref/default key-params (vector-ref (car keys) 0) "")))
    (if (> (length keys) 1)
	(for-each (lambda (key)
		    (set! tmp (conc tmp "/" (hash-table-ref/default key-params (vector-ref key 0) ""))))
		  (cdr keys)))
    tmp))
								
;; for use from the gui
(define (tasks:add-from-params tdb action keys key-params var-params)
(define (tasks:add-from-params mdb action keys key-params var-params)
  (let ((target    (keys:key-vals-hash->target keys key-params))
	(owner     (car (user-information (current-user-id))))
	(runname   (hash-table-ref/default var-params "runname" #f))
	(testpatts (hash-table-ref/default var-params "testpatts" "%"))
	(itempatts (hash-table-ref/default var-params "itempatts" "%"))
	(params    (hash-table-ref/default var-params "params"    "")))
    (tasks:add tdb action owner target runname testpatts itempatts params)))
    (tasks:add mdb action owner target runname testpatts itempatts params)))

;; return one task from those who are 'new' OR 'waiting' AND more than 10sec old
;;
(define (tasks:snag-a-task tdb)
(define (tasks:snag-a-task mdb)
  (let ((res    #f)
	(keytxt (conc (current-process-id) "-" (get-host-name) "-" (car (user-information (current-user-id))))))

    ;; first randomly set a new to pid-hostname-hostname
    (sqlite3:execute
     tdb 
     mdb 
     "UPDATE tasks_queue SET keylock=? WHERE id IN
        (SELECT id FROM tasks_queue 
           WHERE state='new' OR 
                 (state='waiting' AND (strftime('%s','now')-execution_time) > 10) OR
                 state='reset'
           ORDER BY RANDOM() LIMIT 1);" keytxt)

    (sqlite3:for-each-row
     (lambda (id . rem)
       (set! res (apply vector id rem)))
     tdb
     mdb
     "SELECT id,action,owner,state,target,name,test,item,params,creation_time,execution_time FROM tasks_queue WHERE keylock=? ORDER BY execution_time ASC LIMIT 1;" keytxt)
    (if res ;; yep, have work to be done
	(begin
	  (sqlite3:execute tdb "UPDATE tasks_queue SET state='inprogress',execution_time=strftime('%s','now') WHERE id=?;"
	  (sqlite3:execute mdb "UPDATE tasks_queue SET state='inprogress',execution_time=strftime('%s','now') WHERE id=?;"
			   (tasks:task-get-id res))
	  res)
	#f)))

(define (tasks:reset-stuck-tasks tdb)
(define (tasks:reset-stuck-tasks mdb)
  (let ((res '()))
    (sqlite3:for-each-row
     (lambda (id delta)
       (set! res (cons id res)))
     tdb
     mdb
     "SELECT id,strftime('%s','now')-execution_time AS delta FROM tasks_queue WHERE state='inprogress' AND delta>700 ORDER BY delta DESC LIMIT 2;")
    (sqlite3:execute 
     tdb 
     mdb 
     (conc "UPDATE tasks_queue SET state='reset' WHERE id IN ('" (string-intersperse (map conc res) "','") "');"))))

;; return all tasks in the tasks_queue table
;;
(define (tasks:get-tasks tdb types states)
(define (tasks:get-tasks mdb types states)
  (let ((res '()))
    (sqlite3:for-each-row
     (lambda (id . rem)
       (set! res (cons (apply vector id rem) res)))
     tdb
     mdb
     (conc "SELECT id,action,owner,state,target,name,test,item,params,creation_time,execution_time 
               FROM tasks_queue "
               ;; WHERE  
               ;;   state IN " statesstr " AND 
	       ;;   action IN " actionsstr 
	   " ORDER BY creation_time DESC;"))
    res))

;; remove tasks given by a string of numbers comma separated
(define (tasks:remove-queue-entries tdb task-ids)
  (sqlite3:execute tdb (conc "DELETE FROM tasks_queue WHERE id IN (" task-ids ");")))
(define (tasks:remove-queue-entries mdb task-ids)
  (sqlite3:execute mdb (conc "DELETE FROM tasks_queue WHERE id IN (" task-ids ");")))

;; 
(define (tasks:start-monitor db tdb)
  (if (> (tasks:get-num-alive-monitors tdb) 2) ;; have two running, no need for more
(define (tasks:start-monitor db mdb)
  (if (> (tasks:get-num-alive-monitors mdb) 2) ;; have two running, no need for more
      (debug:print-info 1 "Not starting monitor, already have more than two running")
      (let* ((megatestdb     (conc *toppath* "/megatest.db"))
	     (monitordbf     (conc *toppath* "/monitor.db"))
	     (last-db-update 0)) ;; (file-modification-time megatestdb)))
	(task:register-monitor tdb)
	(task:register-monitor mdb)
	(let loop ((count      0)
		   (next-touch 0)) ;; next-touch is the time where we need to update last_update
	  ;; if the db has been modified we'd best look at the task queue
	  (let ((modtime (file-modification-time megatestdbpath )))
	    (if (> modtime last-db-update)
		(tasks:process-queue db tdb last-db-update megatestdb next-touch))
		(tasks:process-queue db mdb last-db-update megatestdb next-touch))
	    ;; WARNING: Possible race conditon here!!
	    ;; should this update be immediately after the task-get-action call above?
	    (if (> (current-seconds) next-touch)
		(begin
		  (tasks:monitors-update tdb)
		  (tasks:monitors-update mdb)
		  (loop (+ count 1)(+ (current-seconds) 240)))
		(loop (+ count 1) next-touch)))))))
      
(define (tasks:process-queue db tdb)
  (let* ((task   (tasks:snag-a-task tdb))
(define (tasks:process-queue db mdb)
  (let* ((task   (tasks:snag-a-task mdb))
	 (action (if task (tasks:task-get-action task) #f)))
    (if action (print "tasks:process-queue task: " task))
    (if action
	(case (string->symbol action)
	  ((run)       (tasks:start-run   db tdb task))
	  ((remove)    (tasks:remove-runs db tdb task))
	  ((lock)      (tasks:lock-runs   db tdb task))
	  ((run)       (tasks:start-run   db mdb task))
	  ((remove)    (tasks:remove-runs db mdb task))
	  ((lock)      (tasks:lock-runs   db mdb task))
	  ;; ((monitor)   (tasks:start-monitor db task))
	  ((rollup)    (tasks:rollup-runs db tdb task))
	  ((updatemeta)(tasks:update-meta db tdb task))
	  ((kill)      (tasks:kill-monitors db tdb task))))))
	  ((rollup)    (tasks:rollup-runs db mdb task))
	  ((updatemeta)(tasks:update-meta db mdb task))
	  ((kill)      (tasks:kill-monitors db mdb task))))))

(define (tasks:get-monitors tdb)
(define (tasks:get-monitors mdb)
  (let ((res '()))
    (sqlite3:for-each-row
     (lambda (a . rem)
       (set! res (cons (apply vector a rem) res)))
     tdb
     mdb
     "SELECT id,pid,strftime('%m/%d/%Y %H:%M',datetime(start_time,'unixepoch'),'localtime'),strftime('%m/%d/%Y %H:%M:%S',datetime(last_update,'unixepoch'),'localtime'),hostname,username FROM monitors ORDER BY last_update ASC;")
    (reverse res)
    ))

(define (tasks:tasks->text tasks)
  (let ((fmtstr "~10a~10a~10a~12a~20a~12a~12a~12a~10a"))
    (conc (format #f fmtstr "id" "action" "owner" "state" "target" "runname" "testpatts" "itempatts" "params") "\n"

			  (tasks:monitor-get-hostname    monitor)
			  (tasks:monitor-get-username    monitor)))
		monitors)
	   "\n"))))
   
;; update the last_update field with the current time and
;; if any monitors appear dead, remove them
(define (tasks:monitors-update tdb)
  (sqlite3:execute tdb "UPDATE monitors SET last_update=strftime('%s','now') WHERE pid=? AND hostname=?;"
(define (tasks:monitors-update mdb)
  (sqlite3:execute mdb "UPDATE monitors SET last_update=strftime('%s','now') WHERE pid=? AND hostname=?;"
			  (current-process-id)
			  (get-host-name))
  (let ((deadlist '()))
    (sqlite3:for-each-row
     (lambda (id pid host last-update delta)
       (print "Going to delete stale record for monitor with pid " pid " on host " host " last updated " delta " seconds ago")
       (set! deadlist (cons id deadlist)))
     tdb 
     mdb 
     "SELECT id,pid,hostname,last_update,strftime('%s','now')-last_update AS delta FROM monitors WHERE delta > 700;")
    (sqlite3:execute tdb (conc "DELETE FROM monitors WHERE id IN ('" (string-intersperse (map conc deadlist) "','") "');")))
    (sqlite3:execute mdb (conc "DELETE FROM monitors WHERE id IN ('" (string-intersperse (map conc deadlist) "','") "');")))
  )

(define (tasks:remove-monitor-record tdb)
  (sqlite3:execute tdb "DELETE FROM monitors WHERE pid=? AND hostname=?;"
(define (tasks:remove-monitor-record mdb)
  (sqlite3:execute mdb "DELETE FROM monitors WHERE pid=? AND hostname=?;"
		   (current-process-id)
		   (get-host-name)))

(define (tasks:set-state tdb task-id state)
  (sqlite3:execute tdb "UPDATE tasks_queue SET state=? WHERE id=?;" 
(define (tasks:set-state mdb task-id state)
  (sqlite3:execute mdb "UPDATE tasks_queue SET state=? WHERE id=?;" 
		   state 
		   task-id))

;;======================================================================
;; The routines to process tasks
;;======================================================================

;; NOTE: It might be good to add one more layer of checking to ensure
;;       that no task gets run in parallel.

(define (tasks:start-run db tdb task)
(define (tasks:start-run db mdb task)
  (let ((flags (make-hash-table)))
    (hash-table-set! flags "-rerun" "NOT_STARTED")
    (if (not (string=? (tasks:task-get-params task) ""))
	(hash-table-set! flags "-setvars" (tasks:task-get-params task)))
    (print "Starting run " task)
    ;; sillyness, just call the damn routine with the task vector and be done with it. FIXME SOMEDAY
    (runs:run-tests db
		    (tasks:task-get-target task)
		    (tasks:task-get-name   task)
		    (tasks:task-get-test   task)
		    (tasks:task-get-item   task)
		    (tasks:task-get-owner  task)
		    flags)
    (tasks:set-state tdb (tasks:task-get-id task) "waiting")))
    (tasks:set-state mdb (tasks:task-get-id task) "waiting")))

(define (tasks:rollup-runs db tdb task)
(define (tasks:rollup-runs db mdb task)
  (let* ((flags (make-hash-table)) 
	 (keys  (db:get-keys db))
	 (keyvallst (keys:target->keyval keys (tasks:task-get-target task))))
    ;; (hash-table-set! flags "-rerun" "NOT_STARTED")
    (print "Starting rollup " task)
    ;; sillyness, just call the damn routine with the task vector and be done with it. FIXME SOMEDAY
    (runs:rollup-run db
		     keys 
		     keyvallst
		     (tasks:task-get-name  task)
		     (tasks:task-get-owner  task))
    (tasks:set-state tdb (tasks:task-get-id task) "waiting")))
    (tasks:set-state mdb (tasks:task-get-id task) "waiting")))

# The NEWTARGET causes some tests to fail. Do not use until this is fixed.
NEWTARGET  = "-target $(OS)/$(FS)/$(VER)"
TARGET     = "-target ubuntu/nfs/none"

all : test1 test2 test3 test4 test5

test0 : cleanprep
	cd simplerun ; $(MEGATEST) -server - -debug $(DEBUG)&

test1 : cleanprep
	rm -f simplerun/megatest.db
	rm -rf simplelinks/ simpleruns/
	mkdir -p simplelinks simpleruns
	cd simplerun;cp ../../*_records.scm .;perl -pi.bak -e 's/define-inline/define/' *_records.scm
	cd simplerun;echo '(load "../tests.scm")' | $(MEGATEST) -repl -debug $(DEBUG)

# NOTE: Only one instance can be a server
test5 : fullprep
	cd fullrun;$(MEGATEST) $(SERVER) $(LOGGING) &
	cd fullrun;sleep 10;$(MEGATEST) -runall $(TARGET) :runname $(RUNNAME)_aa -debug $(DEBUG) $(LOGGING) > aa.log 2> aa.log &
	cd fullrun;sleep 10;$(MEGATEST) -runall $(TARGET) :runname $(RUNNAME)_ab -debug $(DEBUG) $(LOGGING) > ab.log 2> ab.log &
	cd fullrun;sleep 10;$(MEGATEST) -runall $(TARGET) :runname $(RUNNAME)_ac -debug $(DEBUG) $(LOGGING) > ac.log 2> ac.log &
	cd fullrun;sleep 10;$(MEGATEST) -runall $(TARGET) :runname $(RUNNAME)_ad -debug $(DEBUG) $(LOGGING) > ad.log 2> ad.log &	
	# cd fullrun;sleep 10;$(MEGATEST) -runall $(TARGET) :runname $(RUNNAME)_ae -debug $(DEBUG) $(LOGGING) > ae.log 2> ae.log &	
	# cd fullrun;sleep 10;$(MEGATEST) -runall $(TARGET) :runname $(RUNNAME)_af -debug $(DEBUG) $(LOGGING) > af.log 2> af.log &	
	cd fullrun;sleep 10;$(MEGATEST) -runall $(TARGET) :runname $(RUNNAME)_ae -debug $(DEBUG) $(LOGGING) > ae.log 2> ae.log &	
	cd fullrun;sleep 10;$(MEGATEST) -runall $(TARGET) :runname $(RUNNAME)_af -debug $(DEBUG) $(LOGGING) > af.log 2> af.log &	

test6: fullprep
	cd fullrun;$(MEGATEST) -runtests runfirst -testpatt %/1 -reqtarg ubuntu/nfs/none :runname $(RUNNAME)_itempatt -v
	cd fullrun;$(MEGATEST) -runtests runfirst -testpatt %blahha% -reqtarg ubuntu/nfs/none :runname $(RUNNAME)_itempatt -debug 10
	cd fullrun;$(MEGATEST) -rollup :runname newrun -target ubuntu/nfs/none -debug 10


cleanprep : ../*.scm Makefile */*.config
	# if [ -e fullrun/megatest.db ]; then sqlite3 fullrun/megatest.db "delete from metadat where var='SERVER';";fi
	mkdir -p /tmp/mt_runs /tmp/mt_links
	cd ..;make install
	rm -f fullrun/logging.db
	rm -f */logging.db */monitor.db
	touch cleanprep

fullprep : cleanprep
	cd fullrun;$(MEGATEST) -server - -debug $(DEBUG) &
	sleep 5;cd fullrun;$(MEGATEST) -remove-runs :runname $(RUNNAME)% -target %/%/% -testpatt %/%
	cd fullrun;$(BINPATH)/dboard -rows 15 &

dashboard : cleanprep
	cd fullrun && $(BINPATH)/dashboard -rows 25 &

remove :
	cd fullrun;$(MEGATEST) -remove-runs :runname $(RUN)  -testpatt % -itempatt % :sysname % :fsname % :datapath %

clean  : 
	rm cleanprep

kill :
	killall -v mtest main.sh dboard || true
	rm -f fullrun/megatest.db fullrun/logging.db || true
	killall -v -9 mtest dboard || true
	killall -v mtest dboard || true

hardkill : kill
	sleep 5;killall -v mtest main.sh dboard -9

runforever :
	while(ls); do runname=`date +%F-%R:%S`;(cd fullrun;$(MEGATEST) -runall -target ubuntu/nfs/none :runname $$runname;/home/matt/data/megatest/megatest -runall -target ubuntu/nfs/none :runname $$runname;/home/matt/data/megatest/megatest -runall -target ubuntu/nfs/none :runname $$runname);done

;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

;;  strftime('%m/%d/%Y %H:%M:%S','now','localtime')

(require-extension test)
(require-extension regex)
(require-extension srfi-18)
(import srfi-18)
(require-extension zmq)
(import zmq)

(define test-work-dir (current-directory))

;; read in all the _record files
(let ((files (glob "*_records.scm")))
  (for-each
   (lambda (file)
     (print "Loading " file)
     (load file))
   files))

(define *runremote* #f)

;;======================================================================
;; P R O C E S S E S
;;======================================================================

(test "cmd-run-with-stderr->list" '("No such file or directory")
      (let ((reslst (cmd-run-with-stderr->list "ls" "/tmp/ihadbetternotexist")))

(test #f "item_path GLOB ''" (db:patt->like "item_path" ""))

;; test:match->sqlqry
(test #f "(testname GLOB 'a' AND item_path GLOB 'b') OR (testname LIKE 'a%' AND item_path LIKE '%') OR (testname GLOB '' AND item_path LIKE 'b%')"
      (tests:match->sqlqry "a/b,a%,/b%"))
(test #f "(testname GLOB 'a' AND item_path GLOB 'b') OR (testname LIKE 'a%' AND item_path LIKE '%') OR (testname LIKE '%' AND item_path LIKE 'b%')"
      (tests:match->sqlqry "a/b,a%,%/b%"))

;;======================================================================
;; S E R V E R
;;======================================================================

(test "setup for run" #t (begin (setup-for-run)
				(string? (getenv "MT_RUN_AREA_HOME"))))

(test "server-register, get-best-server" '("bob" 1234) (let ((res #f))
							 (open-run-close tasks:server-register tasks:open-db 1 "bob" 1234 100 'live)
							 (set! res (open-run-close tasks:get-best-server tasks:open-db))
							 res))
(test "de-register server" #f (let ((res #f))
				(open-run-close tasks:server-deregister tasks:open-db "bob" port: 1234)
				(open-run-close tasks:get-best-server tasks:open-db)))


;; (exit)

(set! *verbosity* 10)
(define server-pid (process-run "../../bin/megatest" (list "-server" "-" "-debug" (conc *verbosity*))))
(sleep 3)

(define th1 (make-thread (lambda ()(server:client-setup))))
(thread-start! th1)

(test #f #t (socket? *runremote*))

;;======================================================================
;; C O N F I G   F I L E S 
;;======================================================================

(define conffile #f)
(test "Read a config" #t (hash-table? (read-config "test.config" #f #f)))
(test "Read a config that doesn't exist" #t (hash-table? (read-config "nada.config" #f #f)))

;; db
(define row    (vector "a" "b" "c" "blah"))
(define header (list "col1" "col2" "col3" "col4"))
(test "Get row by header" "blah" (db:get-value-by-header row header "col4"))

;; (define *toppath* "tests")
(define *db* #f)
(test "setup for run" #t (begin (setup-for-run)
				(string? (getenv "MT_RUN_AREA_HOME"))))
(test "open-db" #t (begin
		     (set! *db* (open-db))
		     (if *db* #t #f)))

;; quit wasting time, I'm changing *db* to db
(define db *db*)

                                      (and (file-exists? "nada.sh")
    			                 (file-exists? "nada.csh"))))

(test "get all legal tests" (list "test1" "test2") (sort (get-all-legal-tests) string<=?))

(test "register-test, test info" "NOT_STARTED"
      (begin
	(cdb:tests-register-test *remoterun* 1 "nada" "")
	(cdb:tests-register-test *runremote* 1 "nada" "")
	;; (rdb:flush-queue)
	(vector-ref (db:get-test-info *db* 1 "nada" "") 3)))

(test #f "NOT_STARTED"    
      (begin
	(rdb:tests-register-test #f 1 "nada" "")
	;; (rdb:flush-queue)

;; (exit)

;;======================================================================
;; R E M O T E   C A L L S 
;;======================================================================

;; start a server process
(set! *verbosity* 10)
;; (define server-pid (process-run "../../bin/megatest" (list "-server" "-" "-debug" (conc *verbosity*))))
;; (sleep 2)

(define th1 (make-thread server:launch))
(thread-start! th1)

(define start-wait (current-seconds))
(server:client-setup)
(print "Starting intensive cache and rpc test")
(for-each (lambda (params)
	    ;;; (rdb:tests-register-test #f 1 (conc "test" (random 20)) "")
	    (apply cdb:test-set-status-state *remoterun* test-id params)
	    (apply cdb:test-set-status-state *runremote* test-id params)
	    (rdb:pass-fail-counts test-id (random 100) (random 100))
	    (rdb:test-rollup-test_data-pass-fail test-id)
	    (thread-sleep! 0.01)) ;; cache ordering granularity is at the second level. Should really be at the ms level
	  '(("COMPLETED"    "PASS" #f)
	    ("NOT_STARTED"  "FAIL" "Just testing")
	    ("NOT_STARTED"  "FAIL" "Just testing")
	    ("NOT_STARTED"  "FAIL" "Just testing")

(test "Rollup the run(s)" #t (begin
			       (runs:rollup-run keys (keys->alist keys "na") "rollup" "matt")
			       #t))

(hash-table-set! args:arg-hash ":runname" "%")

(test "Remove the rollup run" #t (begin (operate-on 'remove-runs)))

(thread-join! th1 th2 th3)

;; ADD ME!!!! (db:get-prereqs-not-met *db* 1 '("runfirst") "" mode: 'normal)
;; ADD ME!!!! (rdb:get-tests-for-run *db* 1 "runfirst" #f '() '())