Megatest

(declare (unit tasks))
(declare (uses db))
(declare (uses rmt))
(declare (uses common))
(declare (uses pgdb))

;; (import pgdb) ;; pgdb is a module










(include "task_records.scm")
(include "db_records.scm")

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

;; wait up to aprox n seconds for a journal to go away
;;
(define (tasks:wait-on-journal path n #!key (remove #f)(waiting-msg #f))
  (if (not (string? path))
      (debug:print-error 0 *default-log-port* "Called tasks:wait-on-journal with path=" path " (not a string)")
      (let ((fullpath (conc path "-journal")))
	(handle-exceptions
	 exn
	 (begin
	   (print-call-chain (current-error-port))
	   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	   (debug:print 5 *default-log-port* " exn=" (condition->list exn))
	   (debug:print 0 *default-log-port* "tasks:wait-on-journal failed. Continuing on, you can ignore this call-chain")
	   #t) ;; if stuff goes wrong just allow it to move on
	 (let loop ((journal-exists (common:file-exists? fullpath))
		    (count          n)) ;; wait ten times ...
	   (if journal-exists
	       (begin
		 (if (and waiting-msg
			  (eq? (modulo n 30) 0))
		     (debug:print 0 *default-log-port* waiting-msg))
		 (if (> count 0)
		     (begin
		       (thread-sleep! 1)
		       (loop (common:file-exists? fullpath)
			     (- count 1)))
		     (begin
		       (debug:print 0 *default-log-port* "ERROR: removing the journal file " fullpath ", this is not good. Look for disk full, write access and other issues.")
		       (if remove (system (conc "rm -rf " fullpath)))
		       #f)))
	       #t))))))

(define (tasks:get-task-db-path)
  (let ((dbdir  (or (configf:lookup *configdat* "setup" "monitordir")
		    (configf:lookup *configdat* "setup" "dbdir")
		    (conc (common:get-linktree) "/.db"))))
    (handle-exceptions
     exn
     (begin

(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:need-server run-id)
  (equal? (configf:lookup *configdat* "server" "required") "yes"))

;; no elegance here ...
;;
(define (tasks:kill-server hostname pid #!key (kill-switch ""))
  (debug:print-info 0 *default-log-port* "Attempting to kill server process " pid " on host " hostname)
  (setenv "TARGETHOST" hostname)
  (let* ((logdir (if (directory-exists? "logs")
                    "logs/"
                    ""))
         (logfile (if logdir (conc "logs/server-"pid"-"hostname".log") #f))
         (gzfile  (if logfile (conc logfile ".gz"))))
    (setenv "TARGETHOST_LOGF" (conc logdir "server-kills.log"))

    (system (conc "nbfake kill "kill-switch" "pid))

    (when logfile
      (thread-sleep! 0.5)
      (if (common:file-exists? gzfile) (delete-file gzfile))
      (system (conc "gzip " logfile))
      
      (unsetenv "TARGETHOST_LOGF")
      (unsetenv "TARGETHOST"))))
    
 
;;======================================================================
;; M O N I T O R S
;;======================================================================

(define (tasks:remove-monitor-record mdb)
  (sqlite3:execute mdb "DELETE FROM monitors WHERE pid=? AND hostname=?;"

       (set! res count))
     mdb
     "SELECT count(id) FROM monitors WHERE last_update < (strftime('%s','now') - 300) AND username=?;"
     (car (user-information (current-user-id))))
    res))

;; 
#;(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 *default-log-port* "Not starting monitor, already have more than two running")
      (let* ((megatestdb     (conc *toppath* "/megatest.db"))
	     (monitordbf     (conc (db:dbfile-path #f) "/monitor.db"))
	     (last-db-update 0)) ;; (file-modification-time megatestdb)))
	(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)) ;; BROKEN. 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 mdb)
		  (loop (+ count 1)(+ (current-seconds) 240)))
		(loop (+ count 1) next-touch)))))))
      
;;======================================================================
;; T A S K S   Q U E U E
;;
;;   NOTE:: These operate on task_queue which is in main.db
;;
;;======================================================================

;; remove tasks given by a string of numbers comma separated
(define (tasks:remove-queue-entries dbstruct task-ids)
  (db:with-db
   dbstruct #f #t
   (lambda (db)
     (sqlite3:execute db (conc "DELETE FROM tasks_queue WHERE id IN (" task-ids ");")))))

#;(define (tasks:process-queue dbstruct)
  (let* ((task   (tasks:snag-a-task dbstruct))
	 (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     dbstruct task))
	  ((remove)    (tasks:remove-runs   dbstruct task))
	  ((lock)      (tasks:lock-runs     dbstruct task))
	  ;; ((monitor)   (tasks:start-monitor db task))
	  #;((rollup)    (tasks:rollup-runs   dbstruct task))
	  ((updatemeta)(tasks:update-meta   dbstruct task))
	  #;((kill)      (tasks:kill-monitors dbstruct task))))))

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

			 param-key state-patt action-patt test-patt)))))

(define (tasks:find-task-queue-records dbstruct target run-name test-patt state-patt action-patt)
  ;; (handle-exceptions
  ;;  exn
  ;;  '()
  ;;  (sqlite3:first-row
  (let ((db (db:delay-if-busy (db:get-db dbstruct)))
	(res '()))



    (sqlite3:for-each-row 
     (lambda (a . b)
       (set! res (cons (cons a b) res)))

     db "SELECT id,action,owner,state,target,name,testpatt,keylock,params FROM tasks_queue 
           WHERE
              target = ? AND name = ? AND state LIKE ? AND action LIKE ? AND testpatt LIKE ?;"
     target run-name state-patt action-patt test-patt)
    res)) ;; )

;; kill any runner processes (i.e. processes handling -runtests) that match target/runname
;; 
;; do a remote call to get the task queue info but do the killing as self here.
;;
(define (tasks:kill-runner target run-name testpatt)
  (let ((records    (rmt:tasks-find-task-queue-records target run-name testpatt "running" "run-tests"))