Megatest

;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.
;;

(require-extension (srfi 18) extras tcp s11n)

(use srfi-1 posix regex regex-case srfi-69 hostinfo md5 message-digest
     directory-utils posix-extras matchable)
     directory-utils posix-extras matchable utils)

(use spiffy uri-common intarweb http-client spiffy-request-vars)

(declare (unit server))

(declare (uses common))
(declare (uses db))

    (thread-join! log-rotate)
    (pop-directory)))

;; given a path to a server log return: host port startseconds
;; any changes to number of elements returned by this fuction will dirctly affect server:record->url,server:record->id,server:kill,server:get-num-alive which uses match let 

(define (server:logf-get-start-info logf)
  (let ((rx (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+) server-id: (\\S+)"))) ;; SERVER STARTED: host:port AT timesecs server id
  (let ((server-rx (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+) server-id: (\\S+)")) ;; SERVER STARTED: host:port AT timesecs server id
        (dbprep-rx (regexp "^SERVER: dbprep"))
        (dbprep-found 0)) 
    (handle-exceptions
	exn
      (begin
	(debug:print-info 0 *default-log-port* "Unable to get server info from " logf ", exn=" exn)
	(list #f #f #f #f)) ;; no idea what went wrong, call it a bad server
      (with-input-from-file
	  logf
	(lambda ()
	  (let loop ((inl  (read-line))
		     (lnum 0))
	    (if (not (eof-object? inl))
		(let ((mlst (string-match rx inl)))
		(let ((mlst (string-match server-rx inl))
                      (dbprep (string-match dbprep-rx inl))
                      )
                  (if dbprep
                    (set! dbprep-found 1)
                  )
		  (if (not mlst)
		      (if (< lnum 500) ;; give up if more than 500 lines of server log read
			  (loop (read-line)(+ lnum 1))
			  (begin 
                           (debug:print-info 0 *default-log-port* "Unable to get server info from first 500 lines of " logf )
                           (list #f #f #f #f)))
		      (let ((dat  (cdr mlst)))
			(list (car dat) ;; host
			      (string->number (cadr dat)) ;; port
			      (string->number (caddr dat))
                              (cadr (cddr dat))))))
                (begin 
                   (if dbprep-found
                      (begin
                         (debug:print-info 0 *default-log-port* "Server is in dbprep at " (current-seconds))
                         (thread-sleep! 25)
                      )
                   (debug:print-info 0 *default-log-port* "Unable to get server info from " logf " at " (current-seconds))
                      (debug:print-info 0 *default-log-port* "Unable to get server info from " logf " at " (current-seconds))
                   )
		    (list #f #f #f #f)))))))))

;; get a list of servers with all relevant data
;; ( mod-time host port start-time pid )
;;
(define (server:get-list areapath #!key (limit #f))
  (let ((fname-rx    (regexp "^(|.*/)server-(\\d+)-(\\S+).log$"))

		(begin
		  (condition-case
		   (create-directory (conc areapath "/logs") #t)
		   (exn (i/o file)(debug:print 0 *default-log-port* "ERROR: Cannot create directory at " (conc areapath "/logs")))
		   (exn ()(debug:print 0 *default-log-port* "ERROR: Unknown error attemtping to get server list. exn=" exn)))
		  (directory-exists? (conc areapath "/logs")))
		'()))
	(let* ((server-logs   (glob (conc areapath "/logs/server-*.log")))

        ;; Get the list of server logs that do not contain "exiting". Ignore logs for servers that have exited.
	(let* ((server-logs-cmd  (conc "grep -iL exiting " areapath "/logs/server-*-*.log"))
               (server-logs   (string-split (string-chomp (call-with-input-pipe server-logs-cmd read-all))))
	       (num-serv-logs (length server-logs)))
	  (if (null? server-logs)
	      '()
	      (let loop ((hed  (car server-logs))
	  (if (or (null? server-logs) (= num-serv-logs 0))
              (let ()
                 (debug:print 1  *default-log-port* "There are no servers running")
	         '()
              )
	      (let loop ((hed  (string-chomp (car server-logs)))
			 (tal  (cdr server-logs))
			 (res '()))
		(let* ((mod-time  (handle-exceptions
				   exn
				   (begin
				     (debug:print 0 *default-log-port* "failed to get modification time on " hed ", exn=" exn)
				     (current-seconds)) ;; 0

				     res
				     (cons (append serv-rec (list pid)) res)))) ;; any changes to number of elements in new-res will dirctly affect server:record->url,server:record->id,server:kill,server:get-num-alive which uses match let 
		  (if (null? tal)
		      (if (and limit
			       (> (length new-res) limit))
			  new-res ;; (take new-res limit)  <= need intelligent sorting before this will work
			  new-res)
		      (loop (car tal)(cdr tal) new-res)))))))))
		      (loop (string-chomp (car tal)) (cdr tal) new-res)))))))))

(define (server:get-num-alive srvlst)
  (let ((num-alive 0))
    (for-each
     (lambda (server)
        (handle-exceptions
          exn

	 (server-key (conc (get-host-name) "-" (current-process-id))))
    (if (file-exists? start-flag)
	(let* ((fmodtime (file-modification-time start-flag))
	       (delta    (- (current-seconds) fmodtime))
	       (all-go   (> delta reftime)))
	  (if (and all-go
		   (begin
                     (debug:print-info 0 *default-log-port* "Writing " start-flag)
		     (with-output-to-file start-flag
		       (lambda ()
			 (print server-key)))
		     (thread-sleep! 0.25)
		     (let ((res (with-input-from-file start-flag
				  (lambda ()
				    (read-line)))))

				((2)  300)
				(else 600))
			      (random 5)))   ;; add a small random number just in case a lot of jobs hit the work hosts simultaneously
	     (lock-file    (conc areapath "/logs/server-start.lock")))
	(if	(> (- (current-seconds) when-run) run-delay)
		(let* ((start-flag (conc areapath "/logs/server-start-last")))
		  (common:simple-file-lock-and-wait lock-file expire-time: 15)
                  (debug:print-info  0 *default-log-port* "server:kind-run: touching " start-flag)
		  (system (conc "touch " start-flag)) ;; lazy but safe
		  (server:run areapath)
		  (thread-sleep! 2) ;; don't release the lock for at least a few seconds
		  (common:simple-file-release-lock lock-file)))
	(hash-table-set! *server-kind-run* areapath (list (+ call-num 1)(current-seconds))))))

;; this one seems to be the general entry point

(define (server:kill servr)
  (handle-exceptions
    exn
    (begin 
      (debug:print-info 0 *default-log-port*  "Unable to get host and/or port from " servr ", exn=" exn)     
    #f)
  (match-let (((mod-time hostname port start-time pid)
  (match-let (((mod-time hostname port start-time server-id pid)
	       servr))
    (tasks:kill-server hostname pid))))

;; called in megatest.scm, host-port is string hostname:port
;;
;; NOTE: This is NOT called directly from clients as not all transports support a client running
;;       in the same process as the server.

  (let ((legacy-sync  (common:run-sync?))
        (sync-stale-seconds (configf:lookup-number *configdat* "server" "sync-stale-seconds" default: 300))
	(debug-mode   (debug:debug-mode 1))
	(last-time    (current-seconds))
	(no-sync-db   (db:open-no-sync-db))
	(stmt-cache   (dbr:dbstruct-stmt-cache dbstruct))
        (sync-duration 0) ;; run time of the sync in milliseconds
        ;;(this-wd-num  (begin (mutex-lock! *wdnum*mutex) (let ((x *wdnum*)) (set! *wdnum* (add1 *wdnum*)) (mutex-unlock! *wdnum*mutex) x)))
        )
    (set! *no-sync-db* no-sync-db) ;; make the no sync db available to api calls
    (debug:print-info 2 *default-log-port* "Periodic sync thread started.")
    (debug:print-info 3 *default-log-port* "watchdog starting. legacy-sync is " legacy-sync" pid="(current-process-id)  );;  " this-wd-num="this-wd-num)
    (if (and legacy-sync (not *time-to-exit*))
	(let* (;;(dbstruct (db:setup))
	       (mtdb       (dbr:dbstruct-mtdb dbstruct))