Megatest

;; 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')

(use sqlite3 srfi-1 posix regex regex-case srfi-69 dot-locking format)
(import (prefix sqlite3 sqlite3:))
(include "/nfs/site/disks/icf_fdk_cw_gwa002/srehman/fossil/dbi/dbi.scm")
(import (prefix dbi dbi:))


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

(include "task_records.scm")

	     (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))
        (dbdat '())
	      (mdb          (cond ;; what the hek is *toppath* doing here?
			     ((and (string? *toppath*)(file-write-access? *toppath*))
            (set! dbdat (cons (cons 'dbname dbfile) dbdat))
			      (dbi:open 'sqlite3 dbdat))
			     ((file-read-access? dbpath)    (dbi:open 'sqlite3 dbdat))
			     (else (dbi:open 'sqlite3 '((dbname . ":memory:"))))))) ;; (never-give-up-open-db dbpath))
	      ;;(handler      (make-busy-timeout 36000)))

	 (if (and exists
		  (not write-access))
	     (set! *db-write-access* write-access)) ;; only unset so other db's also can use this control
	 ;;(sqlite3:set-busy-handler! mdb handler)
	 (db:set-sync mdb) ;; (dbi:exec mdb (conc "PRAGMA synchronous = 0;"))
	 ;;  (if (or (and (not exists)
	 ;; 	      (file-write-access? *toppath*))
	 ;; 	 (not (file-read-access? dbpath)))
	 ;;      (begin
	 ;; 
	 ;; TASKS QUEUE MOVED TO main.db
	 ;;
	 ;; (dbi:exec 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 '',
         ;;                        testpatt TEXT DEFAULT '',
         ;;                        keylock TEXT,
         ;;                        params TEXT,
         ;;                        creation_time TIMESTAMP,
         ;;                        execution_time TIMESTAMP);")
	 (dbi:exec 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));")
	 (dbi:exec mdb "CREATE TABLE IF NOT EXISTS servers (id INTEGER PRIMARY KEY,
                                  pid INTEGER,
                                  interface TEXT,
                                  hostname TEXT,
                                  port INTEGER,
                                  pubport INTEGER,
                                  start_time TIMESTAMP,
                                  priority INTEGER,
                                  state TEXT,
                                  mt_version TEXT,
                                  heartbeat TIMESTAMP,
                                  transport TEXT,
                                  run_id INTEGER);")
	 ;;                               CONSTRAINT servers_constraint UNIQUE (pid,hostname,port));")
	 (dbi:exec 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));")

	(tasks:server-set-available mdb run-id)
	(thread-sleep! (/ (random 1500) 1000)) ;; (thread-sleep! 2) ;; Try removing this. It may not be needed.
	(tasks:server-am-i-the-server? mdb run-id))
      #f))
	
;; register that this server may come online (first to register goes though with the process)
(define (tasks:server-set-available mdb run-id)
  (dbi:exec 
   mdb 
   "INSERT INTO servers (pid,hostname,port,pubport,start_time,      priority,state,mt_version,heartbeat,   interface,transport,run_id)
                   VALUES(?, ?,       ?,   ?, strftime('%s','now'), ?,       ?,    ?,-1,?,        ?,        ?);"
   (current-process-id)          ;; pid
   (get-host-name)               ;; hostname
   -1                            ;; port
   -1                            ;; pubport
   (random 1000)                 ;; priority (used a tiebreaker on get-available)
   "available"                   ;; state
   (common:version-signature)    ;; mt_version
   -1                            ;; interface
   ;; (conc (server:get-transport)) ;; transport
   (conc *transport-type*)    ;; transport
   run-id
   ))

(define (tasks:num-in-available-state mdb run-id)
  (let ((res 0))
    (dbi:for-each-row
      (lambda (output)
     (lambda (num-in-queue)
       (set! res num-in-queue)))
     mdb
     "SELECT count(id) FROM servers WHERE run_id=? AND state = 'available' AND (strftime('%s','now') - start_time) < 30 ;"
     run-id)
    res))

(define (tasks:num-servers-non-zero-running mdb)
  (let ((res 0))
    (dbi:for-each-row
      (lambda (output)
     (lambda (num-running)
       (set! res num-running)))
     mdb
     "SELECT count(id) FROM servers WHERE run_id != 0 AND state = 'running';")
    res))

(define (tasks:server-clean-out-old-records-for-run-id mdb run-id tag)
  (dbi:exec mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE state in ('available','dbprep','shutting-down') AND (strftime('%s','now') - start_time) > 50 AND run_id=?;"
		   (conc "defunct" tag) run-id))

(define (tasks:server-force-clean-running-records-for-run-id mdb run-id tag)
  (dbi:exec mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE state = 'running' AND run_id=?;"
		   (conc "defunct" tag) run-id))

(define (tasks:server-force-clean-run-record mdb run-id iface port tag)
  (dbi:exec mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE state = 'running' AND run_id=? AND interface=? AND port=?;"
		   (conc "defunct" tag) run-id iface port))


;; BB> adding missing func for --list-servers
(define (tasks:server-deregister mdb hostname #!key (pullport #f) (pid #f) (action #f)) ;;pullport pid: pid action: 'delete))
  (if (eq? action 'delete)
      (dbi:exec mdb "DELETE FROM servers WHERE pid=? AND port=? AND hostname=?;" pid pullport hostname)
      (dbi:exec mdb "UPDATE servers SET state='defunct', heartbeat=strftime('%s','now') WHERE hostname=? AND pid=?;"
                       hostname pid)))

(define (tasks:server-delete-records-for-this-pid mdb tag)
  (dbi:exec mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE hostname=? AND pid=?;"
		   (conc "defunct" tag) (get-host-name) (current-process-id)))

(define (tasks:server-delete-record mdb server-id tag) 
  (dbi:exec mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE id=?;"
		   (conc "defunct" tag) server-id)
  ;; use this opportuntity to clean out records over one month old or over 10 minutes old with port = -1 (i.e. a never used placeholder)
  (dbi:exec mdb "DELETE FROM servers WHERE state not in ('running','shutting-down','dbprep') AND (strftime('%s','now') - start_time) > 2628000;")
  (dbi:exec mdb "DELETE FROM servers WHERE state like 'defunct%' AND port=-1 AND (strftime('%s','now') - start_time) > 600;")
  )

(define (tasks:server-set-state! mdb server-id state)
  (dbi:exec mdb "UPDATE servers SET state=?,heartbeat=strftime('%s','now') WHERE id=?;" state server-id))

(define (tasks:server-set-interface-port mdb server-id interface port)
  (dbi:exec mdb "UPDATE servers SET interface=?,port=?,heartbeat=strftime('%s','now') WHERE id=?;" interface port server-id))

;; Get random port not used in long time
;;
(define (tasks:server-get-next-port mdb)
  (let* ((lownum        30000)
	(highnum        64000)
	(used-ports     '())
	(get-rand-port  (lambda ()
			  (+ lownum (random (- highnum lownum)))))
	(port-param     (if (and (args:get-arg "-port")
				 (string->number (args:get-arg "-port")))
			    (string->number (args:get-arg "-port"))
			    #f))
	;; (config-port    (if (and (config-lookup  *configdat* "server" "port")
	;; 			 (string->number (config-lookup  *configdat* "server" "port")))
	;; 		    (string->number (config-lookup  *configdat* "server" "port"))
	;; 		    #f))
	)
    (dbi:for-each-row
      (lambda (output)
     (lambda (port)
       (set! used-ports (cons port used-ports))))
     mdb
     "SELECT port FROM servers;")
    (cond
     ((and port-param res)   (if (> res port-param) res port-param))
     (port-param             port-param)
     ;; ((and config-port res)  (if (> res config-port) res config-port))
     ;; (config-port            config-port)

;; Use: (db:get-value-by-header (car (db:get-rows dat)) (db:get-header dat) "fieldname")
;;  to extract info from the structure returned
;;
(define (tasks:server-get-servers-vying-for-run-id mdb run-id)
   (let* ((header (list "id" "hostname" "pid" "interface" "port" "pubport" "state" "run_id" "priority" "start_time"))
	  (selstr (string-intersperse header ","))
	  (res    '()))
    (dbi:for-each-row
      (lambda (output)
     (lambda (a . b)
       (set! res (cons (apply vector a b) res))))
     mdb
     (conc "SELECT " selstr " FROM servers WHERE run_id=? AND state in ('available','running','dbprep') ORDER BY start_time DESC;")
     run-id)
    (vector header res)))

(define (tasks:get-server mdb run-id #!key (retries 10))
  (let ((res  #f)

       (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\"")))
     (dbi:for-each-row
      (lambda (output)
      (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
          WHERE run_id=? AND state='running'
          ORDER BY start_time DESC LIMIT 1;" run-id) ;; (common:version-signature) run-id)
     res)))

(define (tasks:server-running-or-starting? mdb run-id)
  (let ((res #f))
    (dbi:for-each-row
      (lambda (output)
     (lambda (id)
       (set! res id)))
     mdb ;; NEEDS dbprep ADDED
     "SELECT id FROM servers WHERE run_id=? AND (state = 'running' OR (state = 'dbprep' AND  (strftime('%s','now') - start_time) < 60));" run-id)
    res))

(define (tasks:server-running? mdb run-id)
  (let ((res #f))
    (dbi:for-each-row
      (lambda (output)
     (lambda (id)
       (set! res id)))
     mdb ;; NEEDS dbprep ADDED
     "SELECT id FROM servers WHERE run_id=? AND state = 'running';" run-id)
    res))

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

                         (server:read-dotserver *toppath*))) ;; no point in trying
                (loop (tasks:get-server (db:delay-if-busy tdbdat) run-id)(+ delay-time 1))
                #f))
          #f)))

(define (tasks:get-all-servers mdb)
  (let ((res '()))
    (dbi:for-each-row
      (lambda (output)
     (lambda (id pid hostname interface port pubport start-time priority state mt-version last-update transport run-id)
       ;;                       0  1     2         3      4     5          6        7     8          9          10        11     12
       (set! res (cons (vector id pid hostname interface port pubport start-time priority state mt-version last-update transport run-id) res))))
     mdb
     "SELECT id,pid,hostname,interface,port,pubport,start_time,priority,state,mt_version,strftime('%s','now')-heartbeat AS last_update,transport,run_id 
        FROM servers WHERE state NOT LIKE 'defunct%' ORDER BY start_time DESC;")
    res))

(define (tasks:get-server-by-id mdb id)
  (let ((res #f))
    (dbi:for-each-row
      (lambda (output)

     (lambda (id pid hostname interface port pubport start-time priority state mt-version last-update transport run-id)
       ;;                       0  1     2         3      4     5          6        7     8          9          10        11     12
       (set! res (vector id pid hostname interface port pubport start-time priority state mt-version last-update transport run-id))))
     mdb
     "SELECT id,pid,hostname,interface,port,pubport,start_time,priority,state,mt_version,strftime('%s','now')-heartbeat AS last_update,transport,run_id 
        FROM servers WHERE id=?;"
     id)
    res))

(define (tasks:get-server-records mdb run-id)
  (let ((res '()))
    (dbi:for-each-row
      (lambda (output)

     (lambda (id pid hostname interface port pubport start-time priority state mt-version last-update transport run-id)
       ;;                       0  1     2         3      4     5          6        7     8          9          10        11     12
       (set! res (cons (vector id pid hostname interface port pubport start-time priority state mt-version last-update transport run-id) res))))
     mdb
     "SELECT id,pid,hostname,interface,port,pubport,start_time,priority,state,mt_version,strftime('%s','now')-heartbeat AS last_update,transport,run_id 
        FROM servers WHERE run_id=? AND state NOT LIKE 'defunct%' ORDER BY start_time DESC;"
     run-id)
    (reverse res)))

;; no elegance here ...

    ))
    
;;======================================================================
;; M O N I T O R S
;;======================================================================

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

(define (tasks:get-monitors mdb)
  (let ((res '()))
    (dbi:for-each-row
      (lambda (output)
     (lambda (a . rem)
       (set! res (cons (apply vector a rem) res))))
     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:monitors->text-table monitors)
  (let ((fmtstr "~4a~8a~20a~20a~10a~10a"))

			  (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 mdb)
  (dbi:exec mdb "UPDATE monitors SET last_update=strftime('%s','now') WHERE pid=? AND hostname=?;"
			  (current-process-id)
			  (get-host-name))
  (let ((deadlist '()))
    (dbi:for-each-row
      (lambda (output)
     (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))))
     mdb 
     "SELECT id,pid,hostname,last_update,strftime('%s','now')-last_update AS delta FROM monitors WHERE delta > 700;")
    (dbi:exec mdb (conc "DELETE FROM monitors WHERE id IN ('" (string-intersperse (map conc deadlist) "','") "');")))
  )
(define (tasks:register-monitor db port)
  (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 ", port: " port ", username: " username)
    (dbi:exec db "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 mdb)
  (let ((res 0))
    (dbi:for-each-row 
      (lambda (output)
     (lambda (count)
       (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)

;; register a task
(define (tasks:add dbstruct action owner target runname testpatt params)
  (db:with-db 
   dbstruct #f #t
   (lambda (db)
     (dbi:exec db "INSERT INTO tasks_queue (action,owner,state,target,name,testpatt,params,creation_time,execution_time)
                             VALUES (?,?,'new',?,?,?,?,strftime('%s','now'),0);" 
		      action
		      owner
		      target
		      runname
		      testpatt
		      (if params params "")))))

(define (tasks:snag-a-task dbstruct)
  (let ((res    #f)
	(keytxt (conc (current-process-id) "-" (get-host-name) "-" (car (user-information (current-user-id))))))
    (db:with-db
     dbstruct #f #t
     (lambda (db)
       ;; first randomly set a new to pid-hostname-hostname
       (dbi:exec
	db 
	"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)

       (dbi:for-each-row
        (lambda (output)
	(lambda (id . rem)
	  (set! res (apply vector id rem))))
	db
	"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
	     (dbi:exec db "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 dbstruct)
  (let ((res '()))
    (db:with-db
     dbstruct #f #t
     (lambda (db)
       (dbi:for-each-row
      (lambda (output)
	(lambda (id delta)
	  (set! res (cons id res))))
	db
	"SELECT id,strftime('%s','now')-execution_time AS delta FROM tasks_queue WHERE state='inprogress' AND delta>700 ORDER BY delta DESC LIMIT 2;")
       (dbi:exec 
	db 
	(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 dbstruct types states)
  (let ((res '()))
    (db:with-db
     dbstruct #f #f
     (lambda (db)
       (dbi:for-each-row
        (lambda (output)
	(lambda (id . rem)
	  (set! res (cons (apply vector id rem) res))))
	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)
       (dbi:for-each-row
	(lambda (output)
    (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)
     (dbi:exec 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)

			  (tasks:task-get-params task)))
		tasks) "\n"))))
   
(define (tasks:set-state dbstruct task-id state)
  (db:with-db 
   dbstruct #f #t
   (lambda (db)
     (dbi:exec db "UPDATE tasks_queue SET state=? WHERE id=?;" 
		      state 
		      task-id))))

;;======================================================================
;; Access using task key (stored in params; (hash-table->alist flags) hostname pid
;;======================================================================

(define (tasks:param-key->id dbstruct task-params)
  (db:with-db
   dbstruct #f #f
   (lambda (db)
     (handle-exceptions
      exn
      #f
      (dbi:get-one db "SELECT id FROM tasks_queue WHERE params LIKE ?;"
			    task-params)))))

(define (tasks:set-state-given-param-key dbstruct param-key new-state)
  (db:with-db
   dbstruct #f #t
   (lambda (db)
     (dbi:exec db "UPDATE tasks_queue SET state=? WHERE params LIKE ?;" new-state param-key))))

(define (tasks:get-records-given-param-key dbstruct param-key state-patt action-patt test-patt)
  (db:with-db
   dbstruct #f #f
   (lambda (db)
     (handle-exceptions
      exn
      '()
      (dbi:get-one-row db "SELECT id,action,owner,state,target,name,testpatt,keylock,params WHERE
                               params LIKE ? AND state LIKE ? AND action LIKE ? AND testpatt LIKE ?;"
			 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 #f)))
	(res '()))
    (dbi:for-each-row
    (lambda (output) 
     (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