Megatest

;; 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 0 *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 (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 (file-exists? fullpath)
			     (- count 1)))
		     (begin
		       (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 (configf:lookup *configdat* "setup" "linktree") "/.db"))))
    (handle-exceptions
     exn
     (begin
       (debug:print-error 0 *default-log-port* "Couldn't create path to " dbdir)
       (exit 1))
     (if (not (directory? dbdir))(create-directory dbdir #t)))
    dbdir))

;; If file exists AND
;;    file readable
;;         ==> open it

  (if *task-db*
      *task-db*
      (handle-exceptions
       exn
       (if (> numretries 0)
	   (begin
	     (print-call-chain (current-error-port))
	     (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	     (debug:print 0 *default-log-port* " exn=" (condition->list exn))
	     (thread-sleep! 1)
	     (tasks:open-db numretries (- numretries 1)))
	   (begin
	     (print-call-chain (current-error-port))
	     (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	     (debug:print 0 *default-log-port* " exn=" (condition->list exn))))
       (let* ((dbpath       (tasks:get-task-db-path))
	      (dbfile       (conc dbpath "/monitor.db"))
	      (avail        (tasks:wait-on-journal dbpath 10)) ;; wait up to about 10 seconds for the journal to go away
	      (exists       (file-exists? dbpath))
	      (write-access (file-write-access? dbpath))
	      (mdb          (cond ;; what the hek is *toppath* doing here?
			     ((and (string? *toppath*)(file-write-access? *toppath*))

		(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)
		     #f;; (begin (debug:print-error 0 *default-log-port* "no servers listed, should be at least one by now.") 
		       ;;      (sqlite3:finalize! mdb)
		       ;;      (exit 1))
		     (car (db:get-rows all)))))
    (if first
	(let* ((header   (db:get-header all))
	       (id       (db:get-value-by-header first header "id"))
	       (hostname (db:get-value-by-header first header "hostname"))
	       (pid      (db:get-value-by-header first header "pid"))
	       (priority (db:get-value-by-header first header "priority")))
	  ;; (debug:print 0 *default-log-port* "INFO: am-i-the-server got record " first)
	  ;; for now a basic check. add tiebreaking by priority later
	  (if (and (equal? hostname (get-host-name))
		   (equal? pid      (current-process-id)))
	      id
	      #f))
	#f)))
	     

(define (tasks:get-server mdb run-id #!key (retries 10))
  (let ((res  #f)
	(best #f))
    (handle-exceptions
     exn
     (begin
       (print-call-chain (current-error-port))
       (debug:print 0 *default-log-port* "WARNING: tasks:get-server db access error.")
       (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
       (debug:print 0 *default-log-port* " for run " run-id)
       (print-call-chain (current-error-port))
       (if (> retries 0)
	   (begin
	     (debug:print 0 *default-log-port* " trying call to tasks:get-server again in 10 seconds")
	     (thread-sleep! 10)
	     (tasks:get-server mdb run-id retries: (- retries 0)))
	   (debug:print 0 *default-log-port* "10 tries of tasks:get-server all crashed and burned. Giving up and returning \"no server found\"")))
     (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

  (configf:lookup *configdat* "server" "required"))

;; 	(maxqry (cdr (rmt:get-max-query-average run-id)))
;; 	(threshold   (string->number (or (configf:lookup *configdat* "server" "server-query-threshold") "10"))))
;;     (cond
;;      (forced 
;;       (if (common:low-noise-print 60 run-id "server required is set")
;; 	  (debug:print-info 0 *default-log-port* "Server required is set, starting server for run-id " run-id "."))
;;       #t)
;;      ((> maxqry threshold)
;;       (if (common:low-noise-print 60 run-id "Max query time execeeded")
;; 	  (debug:print-info 0 *default-log-port* "Max avg query time of " maxqry "ms exceeds limit of " threshold "ms, server needed for run-id " run-id "."))
;;       #t)
;;      (else
;;       #f))))

;; try to start a server and wait for it to be available
;;
(define (tasks:start-and-wait-for-server tdbdat run-id delay-max-tries)
  ;; ensure a server is running for this run
  (let loop ((server-dat (tasks:get-server (db:delay-if-busy tdbdat) run-id))
	     (delay-time 0))
      (if (and (not server-dat)
	       (< delay-time delay-max-tries))
	  (begin
	    (if (common:low-noise-print 60 "tasks:start-and-wait-for-server" run-id)
		(debug:print 0 *default-log-port* "Try starting server for run-id " run-id))
	    (thread-sleep! (/ (random 2000) 1000))
	    (server:kind-run run-id)
	    (thread-sleep! (min delay-time 1))
	    (loop (tasks:get-server (db:delay-if-busy tdbdat) run-id)(+ delay-time 1))))))

(define (tasks:get-all-servers mdb)
  (let ((res '()))

        FROM servers WHERE run_id=? AND state NOT LIKE 'defunct%' ORDER BY start_time DESC;"
     run-id)
    (reverse res)))

;; no elegance here ...
;;
(define (tasks:kill-server hostname pid)
  (debug:print-info 0 *default-log-port* "Attempting to kill server process " pid " on host " hostname)
  (setenv "TARGETHOST" hostname)
  (setenv "TARGETHOST_LOGF" "server-kills.log")
  (system (conc "nbfake kill " pid))
  (unsetenv "TARGETHOST_LOGF")
  (unsetenv "TARGETHOST"))
 
;; look up a server by run-id and send it a kill, also delete the record for that server
;;
(define (tasks:kill-server-run-id run-id #!key (tag "default"))
  (let* ((tdbdat  (tasks:open-db))
	 (sdat    (tasks:get-server (db:delay-if-busy tdbdat) run-id)))
    (if sdat
	(let ((hostname (vector-ref sdat 6))
	      (pid      (vector-ref sdat 5))
	      (server-id (vector-ref sdat 0)))
	  (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "killed")
	  (debug:print-info 0 *default-log-port* "Killing server " server-id " for run-id " run-id " on host " hostname " with pid " pid)
	  (tasks:kill-server hostname pid)
	  (tasks:server-delete-record (db:delay-if-busy tdbdat) server-id tag) )
	(debug:print-info 0 *default-log-port* "No server found for run-id " run-id ", nothing to kill"))
    ;; (sqlite3:finalize! tdb)
    ))
    
;;======================================================================
;; M O N I T O R S
;;======================================================================

     "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

;;   NOTE:: These operate on task_queue which is in main.db
;;
;;======================================================================

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

;; id INTEGER PRIMARY KEY,
;; action TEXT DEFAULT '',
;; owner TEXT,
;; state TEXT DEFAULT 'new',
;; target TEXT DEFAULT '',
;; name TEXT DEFAULT '',
;; testpatt TEXT DEFAULT '',
;; keylock TEXT,
;; params TEXT,
;; creation_time TIMESTAMP DEFAULT (strftime('%s','now')),
;; execution_time TIMESTAMP);


;; register a task
(define (tasks:add dbstruct action owner target runname testpatt params)
  (db:with-db 
   dbstruct #f #t
   (lambda (db)

	(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))))

(define (tasks:get-last dbstruct target runname)
  (let ((res #f))
    (db:with-db
     dbstruct #f #f
     (lambda (db)
       (sqlite3:for-each-row
	(lambda (id . rem)
	  (set! res (apply vector id rem)))
	db
	(conc "SELECT id,action,owner,state,target,name,testpatt,keylock,params,creation_time,execution_time 
                  FROM tasks_queue 
 	       WHERE  
	        target = ? AND name =?
	       ORDER BY creation_time DESC LIMIT 1;")
	target runname)
       res))))

;; 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 ");")))))

;; 
;; 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"))
	(hostpid-rx (regexp "\\s+(\\w+)\\s+(\\d+)$"))) ;; host pid is at end of param string
    (if (null? records)
	(debug:print 0 *default-log-port* "No run launching processes found for " target " / " run-name " with testpatt " (or testpatt "* no testpatt specified! *"))
	(debug:print 0 *default-log-port* "Found " (length records) " run(s) to kill."))
    (for-each 
     (lambda (record)
       (let* ((param-key (list-ref record 8))
	      (match-dat (string-search hostpid-rx param-key)))
	 (if match-dat
	     (let ((hostname  (cadr match-dat))
		   (pid       (string->number (caddr match-dat))))
	       (debug:print 0 *default-log-port* "Sending SIGINT to process " pid " on host " hostname)
	       (if (equal? (get-host-name) hostname)
		   (if (process:alive? pid)
		       (begin
			 (handle-exceptions
			  exn
			  (begin
			    (debug:print 0 *default-log-port* "Kill of process " pid " on host " hostname " failed.")
			    (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
			    #t)
			  (process-signal pid signal/int)
			  (thread-sleep! 5)
			  (if (process:alive? pid)
			      (process-signal pid signal/kill)))))
		   ;;  (call-with-environment-variables
		   (let ((old-targethost (getenv "TARGETHOST")))
		     (setenv "TARGETHOST" hostname)
		     (setenv "TARGETHOST_LOGF" "server-kills.log")
		     (system (conc "nbfake kill " pid))
		     (if old-targethost (setenv "TARGETHOST" old-targethost))
		     (unsetenv "TARGETHOST")
		     (unsetenv "TARGETHOST_LOGF"))))
	     (debug:print-error 0 *default-log-port* "no record or improper record for " target "/" run-name " in tasks_queue in main.db"))))
     records)))

;; (define (tasks:start-run dbstruct 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)))