Megatest

;;     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
;;     GNU General Public License for more details.
;; 
;;     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)

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

(declare (unit server))

(declare (uses common))
(declare (uses db))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.
;; (declare (uses synchash))
(declare (uses http-transport))
;;(declare (uses rpc-transport))
(declare (uses launch))
;; (declare (uses daemon))

(include "common_records.scm")
(include "db_records.scm")
;; (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)
;; 
;; (use spiffy uri-common intarweb http-client spiffy-request-vars)
;; 
;; (declare (unit server))
;; 
;; (declare (uses common))
;; (declare (uses db))
;; (declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.
;; ;; (declare (uses synchash))
;; (declare (uses http-transport))
;; ;;(declare (uses rpc-transport))
;; (declare (uses launch))
;; ;; (declare (uses daemon))
;; 
;; (include "common_records.scm")
;; (include "db_records.scm")

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

(define  *server-loop-heart-beat* (current-seconds))

;;======================================================================
;; P K T S   S T U F F 
;;======================================================================

;; ???

	     (not (string-match (conc "("curr-host "|" curr-host"\\..*)") target-host))
	     (not (equal? curr-ip target-host)))
	(begin
	  (debug:print-info 0 *default-log-port* "Starting server on " target-host ", logfile is " logfile)
	  (setenv "TARGETHOST" target-host)))
      
    (setenv "TARGETHOST_LOGF" logfile)
    (thread-sleep! (/ (random 5000) 1000)) ;; add about a random (up to 5 seconds) initial delay. It seems pretty common that many running tests request a server at the same time
    (thread-sleep! (/ (pseudo-random-integer 5000) 1000)) ;; add about a random (up to 5 seconds) initial delay. It seems pretty common that many running tests request a server at the same time
    ;; (common:wait-for-normalized-load load-limit " delaying server start due to load" target-host) ;; do not try starting servers on an already overloaded machine, just wait forever
    #;(common:wait-for-homehost-load load-limit (conc " delaying server start due to load on homehost. limit is " load-limit))
    (system (conc "nbfake " cmdln))
    (unsetenv "TARGETHOST_LOGF")
    (if (get-environment-variable "TARGETHOST")(unsetenv "TARGETHOST"))
    (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 ((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 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))
                   )
		    (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$"))
	(day-seconds (* 24 60 60)))
    ;; if the directory exists continue to get the list
    ;; otherwise attempt to create the logs dir and then
    ;; continue
    (if (if (directory-exists? (conc areapath "/logs"))
	    '()
	    (if (file-write-access? areapath)
		(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")))
		'()))

        ;; 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 (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
				   (file-modification-time hed))) ;; default to *very* old so log gets ignored if deleted
		       (down-time (- (current-seconds) mod-time))
		       (serv-dat  (if (or (< num-serv-logs 10)
				  	  (< down-time 900)) ;; day-seconds))
				      (server:logf-get-start-info hed)
				      '())) ;; don't waste time processing server files not touched in the 15 minutes if there are more than ten servers to look at
		       (serv-rec (cons mod-time serv-dat))
		       (fmatch   (string-match fname-rx hed))
		       (pid      (if fmatch (string->number (list-ref fmatch 2)) #f))
		       (new-res  (if (null? serv-dat)
				     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 (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
         (begin 
          (debug:print-info 0 *default-log-port*  "Unable to get server start-time and/or mod-time from " server ", exn=" exn))
       (match-let (((mod-time host port start-time server-id pid)
		    server))
	 (let* ((uptime  (- (current-seconds) mod-time))
		(runtime (if start-time
			     (- mod-time start-time)
			     0)))
	   (if (< uptime 5)(set! num-alive (+ num-alive 1)))))))
     srvlst)
    num-alive))

;; given a list of servers get a list of valid servers, i.e. at least
;; 10 seconds old, has started and is less than 1 hour old and is
;; active (i.e. mod-time < 10 seconds
;;
;; mod-time host port start-time pid
;;
;; sort by start-time descending. I.e. get the oldest first. Young servers will thus drop off
;; and servers should stick around for about two hours or so.
;;
(define (server:get-best srvlst)
  (let* ((nums (server:get-num-servers))
	 (now  (current-seconds))
	 (slst (sort
		(filter (lambda (rec)
			  (if (and (list? rec)
				   (> (length rec) 2))
			      (let ((start-time (list-ref rec 3))
				    (mod-time   (list-ref rec 0)))
				;; (print "start-time: " start-time " mod-time: " mod-time)
				(and start-time mod-time
				     (> (- now start-time) 0)    ;; been running at least 0 seconds
				     (< (- now mod-time)   16)   ;; still alive - file touched in last 16 seconds
				     (or (not (configf:lookup *configdat* "server" "runtime")) ;; skip if not set
					 (< (- now start-time)       
					    (+ (- (string->number (configf:lookup *configdat* "server" "runtime"))
					       180)
					    (random 360)))) ;; under one hour running time +/- 180
				     ))
			      #f))
			srvlst)
		(lambda (a b)
		  (< (list-ref a 3)
		     (list-ref b 3))))))
    (if (> (length slst) nums)
	(take slst nums)
	slst)))

(define (server:get-first-best areapath)
  (let ((srvrs (server:get-best (server:get-list areapath))))
    (if (and srvrs
	     (not (null? srvrs)))
	(car srvrs)
	#f)))

(define (server:get-rand-best areapath)
  (let ((srvrs (server:get-best (server:get-list areapath))))
    (if (and (list? srvrs)
	     (not (null? srvrs)))
	(let* ((len (length srvrs))
	       (idx (random len)))
	  (list-ref srvrs idx))
	#f)))

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

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

	       (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)
		     (thread-sleep! 0.254)
		     (let ((res (with-input-from-file start-flag
				  (lambda ()
				    (read-line)))))
		       (equal? server-key res))))
	      #t ;; (system (conc "touch " start-flag)) ;; lazy but safe
	      (begin
		(debug:print-info 0 *default-log-port* "Gating server start, last start: "

	     (call-num     (car last-run-dat))
	     (when-run     (cadr last-run-dat))
	     (run-delay    (+ (case call-num
				((0)    0)
				((1)   20)
				((2)  300)
				(else 600))
			      (random 5)))   ;; add a small random number just in case a lot of jobs hit the work hosts simultaneously
			      (pseudo-random-integer 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)

		(server:kind-run areapath))
	    (thread-sleep! 5)
	    (loop (server:check-if-running areapath)
		  (+ try-num 1)))))))

(define server:try-running server:run) ;; there is no more per-run servers ;; REMOVE ME. BUG.

(define (server:get-num-servers #!key (numservers 2))
  (let ((ns (string->number
	     (or (configf:lookup *configdat* "server" "numservers") "notanumber"))))
    (or ns numservers)))

;; no longer care if multiple servers are started by accident. older servers will drop off in time.
;;
(define (server:check-if-running areapath) ;;  #!key (numservers "2"))
  (let* ((ns            (server:get-num-servers))
	 (servers       (server:get-best (server:get-list areapath))))
    (if (or (and servers
		 (null? servers))
	    (not servers)
	    (and (list? servers)
		 (< (length servers) (random ns)))) ;; somewhere between 0 and numservers
		 (< (length servers) (pseudo-random-integer ns)))) ;; somewhere between 0 and numservers
        #f
        (let loop ((hed (car servers))
                   (tal (cdr servers)))
          (let ((res (server:check-server hed)))
            (if res
                hed
                (if (null? tal)

         (duty-cycle   (configf:lookup-number *configdat* "server" "sync-duty-cycle" default: default-duty-cycle))
         (last-sync-seconds 10) ;; we will adjust this to a measurement and delay last-sync-seconds * (1 - duty-cycle)
         (calculate-off-time (lambda (work-duration duty-cycle)
                                  (* (/ (- 1 duty-cycle) duty-cycle) last-sync-seconds)))
         (off-time min-intersync-delay) ;; adjusted in closure below.
         (do-a-sync
          (lambda ()
            (BB> "Start do-a-sync with fork-to-background="fork-to-background" persist-until-sync="persist-until-sync)
            ;; (BB> "Start do-a-sync with fork-to-background="fork-to-background" persist-until-sync="persist-until-sync)
            (let* ((finalres
                    (let retry-loop ((num-tries 0))
                         (if (common:simple-file-lock lockfile)
	                     (begin
                               (cond
                                ((not (or fork-to-background persist-until-sync))
                                 (debug:print 0 *default-log-port* "INFO: syncer thread sleeping for max of (server.minimum-intersync-delay="min-intersync-delay

                                        (else
                                         (system (conc "/bin/cp "sync-log" "sync-log".fail"))
                                         (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)"] Sync failed. See log at "sync-log".fail")
                                         (if (file-exists? (conc mtdbfile ".backup"))
                                             (system (conc "/bin/cp "mtdbfile ".backup " mtdbfile)))
                                         #f))))
                                 (common:simple-file-release-lock lockfile)
                                 (BB> "released lockfile: " lockfile)
                                 (when (common:file-exists? lockfile)
                                   (BB> "DID NOT ACTUALLY RELEASE LOCKFILE"))
                                 ;; (BB> "released lockfile: " lockfile)
                                 ;; (when (common:file-exists? lockfile)
                                 ;;   (BB> "DID NOT ACTUALLY RELEASE LOCKFILE"))
                                 res2) ;; end let
                               );; end begin
                             ;; else
                             (cond
                              (persist-until-sync
                               (thread-sleep! 1)
                               (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" other SYNC in progress; we're in a fork-to-background so we need to succeed.  Let's wait a jiffy and and try again. num-tries="num-tries" (waiting for lockfile="lockfile" to disappear)")
                               (retry-loop (add1 num-tries)))
                              (else
                               (thread-sleep! (max off-time (+ last-sync-seconds min-intersync-delay)))
                               (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" other SYNC in progress; not syncing.")
                               'parallel-sync-in-progress))
                             ) ;; end if got lockfile
                         )
                    ))
              (BB> "End do-a-sync with fork-to-background="fork-to-background" persist-until-sync="persist-until-sync" and result="finalres)
              ;; (BB> "End do-a-sync with fork-to-background="fork-to-background" persist-until-sync="persist-until-sync" and result="finalres)
              finalres)
            ) ;; end lambda
          ))
    do-a-sync))

(define (server:writable-watchdog-bruteforce dbstruct)
  (thread-sleep! 1) ;; delay for startup

      (final-sync)

      (if (common:low-noise-print 30)
	  (debug:print-info 0 *default-log-port* "Exiting watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)
			    )))))

(define (server:writable-watchdog-deltasync dbstruct)
  (thread-sleep! 0.05) ;; delay for startup
  (thread-sleep! 0.054) ;; delay for startup
  (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