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 utils)

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

(declare (unit server))

(declare (uses commonmod))

(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))







(import commonmod)



(include "common_records.scm")
(include "db_records.scm")

(define (server:make-server-url hostport)
  (if (not hostport)
      #f

;;======================================================================
;; S E R V E R
;;======================================================================

;; Call this to start the actual server
;;

;; all routes though here end in exit ...
;;
;; start_server
;;
(define (server:launch run-id transport-type)
  (case transport-type
    ((http)(http-transport:launch))
    ;;((nmsg)(nmsg-transport:launch run-id))
    ;;((rpc)  (rpc-transport:launch run-id))
    (else (debug:print-error 0 *default-log-port* "unknown server type " transport-type))))

;;======================================================================
;; S E R V E R   U T I L I T I E S 
;;======================================================================

;; Get the transport
(define (server:get-transport)
  (if *transport-type*

(define (server:get-server-id)
  (if *server-id* *server-id*
      (let ((sig (server:mk-signature))) ;; clients re-use the server:mk-signature logic
        (set! *server-id* sig)
        *server-id*)))

;; When using zmq this would send the message back (two step process)
;; with spiffy or rpc this simply returns the return data to be returned
;; 
(define (server:reply return-addr query-sig success/fail result)
  (debug:print-info 11 *default-log-port* "server:reply return-addr=" return-addr ", result=" result)
  ;; (send-message pubsock target send-more: #t)
  ;; (send-message pubsock 
  (case (server:get-transport)
    ((rpc)  (db:obj->string (vector success/fail query-sig result)))
    ((http) (db:obj->string (vector success/fail query-sig result)))
    ((fs)   result)
    (else 
     (debug:print-error 0 *default-log-port* "unrecognised transport type: " *transport-type*)
     result)))

;; Given an area path,  start a server process    ### NOTE ### > file 2>&1 
;; if the target-host is set 
;; try running on that host
;;   incidental: rotate logs in logs/ dir.
;;
(define  (server:run areapath) ;; areapath is *toppath* for a given testsuite area
  (let* (;; (curr-host   (get-host-name))
         ;; (attempt-in-progress (server:start-attempted? areapath))
         ;; (dot-server-url (server:check-if-running areapath))
	 ;; (curr-ip     (server:get-best-guess-address curr-host))
	 ;; (curr-pid    (current-process-id))
	 ;; (homehost    (server:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" ))
	 ;; (target-host (car homehost))
	 (testsuite   (common:get-testsuite-name))
	 (logfile     (conc areapath "/logs/server.log")) ;; -" curr-pid "-" target-host ".log"))
	 (profile-mode (or (configf:lookup *configdat* "misc" "profilesw")
			   ""))
	 (cmdln (conc (common:get-megatest-exe)
		      " -server - ";; (or target-host "-")
		      (if (equal? (configf:lookup *configdat* "server" "daemonize") "yes")
			  " -daemonize "

(define (server:logf-get-start-info logf)
  (let ((server-rx    (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+) server-id: (\\S+) pid: (\\d+)")) ;; SERVER STARTED: host:port AT timesecs server id
        (dbprep-rx    (regexp "^SERVER: dbprep"))
        (dbprep-found 0)
	(bad-dat      (list #f #f #f #f #f)))
    (handle-exceptions
	exn
      (begin


	(debug:print-info 0 *default-log-port* "Unable to get server info from " logf ", exn=" exn)
	bad-dat) ;; 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 )
                           bad-dat))
		      (match mlst
			((_ host port start server-id pid)
			 (list host
			       (string->number port)
			       (string->number start)
			       server-id
			       (string->number pid)))
			(else
			 (debug:print 0 *current-log-port* "ERROR: did not recognise SERVER line info "mlst)
			 bad-dat))))
                (begin 
                  (if dbprep-found
                      (begin
                         (debug:print-info 2 *default-log-port* "Server is in dbprep at " (common:human-time))
                         (thread-sleep! 0.5)) ;; was 25 sec but that blocked things from starting?
                      (debug:print-info 0 *default-log-port* "Unable to get server info from " logf " at " (seconds->time-string (current-seconds))))
		  bad-dat))))))))

;; ;; get a list of servers from the log files, 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)))

		 (thread-sleep! ( + 1 idletime))
		 (server:wait-for-server-start-last-flag areapath)))))))

;; oldest server alive determines host then choose random of youngest
;; five servers on that host
;;
(define (server:get-servers-info areapath)

  (let* ((servinfodir (conc *toppath*"/.servinfo")))
    (if (not (file-exists? servinfodir))
	(create-directory servinfodir))
    (let* ((allfiles    (glob (conc servinfodir"/*")))
	   (res         (make-hash-table)))
      (for-each
       (lambda (f)
	 (let* ((hostport  (pathname-strip-directory f))
		(serverdat (server:logf-get-start-info f)))
	   (match serverdat
	     ((host port start server-id pid)
	      (if (and host port start server-id pid)
		  (hash-table-set! res hostport serverdat)
		  (debug:print-info 0 *default-log-port* "bad server info for "f": "serverdat)))
	     (else
	      (debug:print-info 0 *default-log-port* "bad server info for "f": "serverdat)))))
       allfiles)
      res)))































;; oldest server alive determines host then choose random of youngest
;; five servers on that host
;;
;; mode:
;;   best - get best server (random of newest five)
;;   home - get home host based on oldest server
;;   info - print info
(define (server:choose-server areapath #!optional (mode 'best))
  ;; age is current-starttime
  ;; find oldest alive
  ;;   1. sort by age ascending and ping until good
  ;; find alive rand from youngest
  ;;   1. sort by age descending
  ;;   2. take five
  ;;   3. check alive, discard if not and repeat














  (let* ((serversdat  (server:get-servers-info areapath))
	 (servkeys    (hash-table-keys serversdat))
	 (by-time-asc (if (not (null? servkeys))
			  (sort servkeys ;; list of "host:port"
				(lambda (a b)
				  (>= (list-ref (hash-table-ref serversdat a) 2)
				      (list-ref (hash-table-ref serversdat b) 2))))
			  '())))


    (if (not (null? by-time-asc))
	(let* ((oldest     (last by-time-asc))
	       (oldest-dat (hash-table-ref serversdat oldest))
	       (host       (list-ref oldest-dat 0))
	       (all-valid  (filter (lambda (x)
				     (equal? host (list-ref (hash-table-ref serversdat x) 0)))
				   by-time-asc))
	       (best-five  (lambda ()
			     (if (> (length all-valid) 5)


				 (take all-valid 5)
				 all-valid)))
	       (names->dats (lambda (names)
			      (map (lambda (x)
				     (hash-table-ref serversdat x))
				   names)))
	       (am-home?    (lambda ()
			      (let* ((currhost (get-host-name))
				     (bestadrs (server:get-best-guess-address currhost)))
				(or (equal? host currhost)
				    (equal? host bestadrs))))))
	  (case mode
	    ((info)
	     (print "oldest: "oldest-dat", selected host: "host", all-valid: "all-valid)
	     (print "youngest: "(hash-table-ref serversdat (car all-valid))))
	    ((home)     host)
	    ((homehost) (cons host (am-home?))) ;; shut up old code
	    ((home?)    (am-home?))
	    ((best-five)(names->dats (best-five)))
	    ((all-valid)(names->dats all-valid))
	    ((best)     (let* ((best-five (best-five))
			       (len       (length best-five)))
			  (hash-table-ref serversdat (list-ref best-five (random len)))))
	    ((count)(length all-valid))
	    (else
	     (debug:print 0 *default-log-port* "ERROR: invalid command "mode)
	     #f)))
	(begin
	  (server:run areapath)

	  (thread-sleep! 3)
	  (case mode
	    ((homehost) (cons #f #f))
	    (else	#f))))))





























;; would like to eventually get rid of this
;;
(define (common:on-homehost?)

  (server:choose-server *toppath* 'home?))


;; kind start up of server, wait before allowing another server for a given
;; area to be launched
;;
(define (server:kind-run areapath)
  ;; look for $MT_RUN_AREA_HOME/logs/server-start-last
  ;; and wait for it to be at least <server idletime> seconds old
  ;; (server:wait-for-server-start-last-flag areapath)






  (if (< (server:choose-server areapath 'count) 10)
      (server:run areapath))
  #;(if (not (server:check-if-running areapath)) ;; why try if there is already a server running?
      (let* ((lock-file    (conc areapath "/logs/server-start.lock")))
	(let* ((start-flag (conc areapath "/logs/server-start-last")))
	  (common:simple-file-lock-and-wait lock-file expire-time: 25)
	  (debug:print-info  2 *default-log-port* "server:kind-run: touching " start-flag)
	  (system (conc "touch " start-flag)) ;; lazy but safe

(define (server:start-and-wait areapath #!key (timeout 60))
  (let ((give-up-time (+ (current-seconds) timeout)))
    (let loop ((server-info (server:check-if-running areapath))
	       (try-num    0))
      (if (or server-info
	      (> (current-seconds) give-up-time)) ;; server-url will be #f if no server available.
	  (server:record->url server-info)

	  (let ((num-ok (length (server:choose-server areapath 'all-valid))))
	    (if (and (> try-num 0)  ;; first time through simply wait a little while then try again
		     (< num-ok 1))  ;; if there are no decent candidates for servers then try starting a new one
		(server:run areapath))
	    (thread-sleep! 5)
	    (loop (server:check-if-running areapath)
		  (+ try-num 1)))))))

(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)) ;; get the setting the for maximum number of servers allowed
	 (servers       (server:choose-server areapath 'best-five))) ;; (server:get-best (server:get-list areapath))))
    (if (or (and servers
		 (null? servers))
	    (not servers))
	    ;; (and (list? servers)
	    ;;	 (< (length servers) (+ 1 (random ns))))) ;; somewhere between 1 and numservers
        #f
        (let loop ((hed (car servers))

(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 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.
;;
(define (server:ping host-port-in server-id #!key (do-exit #f))
  (let ((host:port (if (not host-port-in) ;; use read-dotserver to find
		       #f ;; (server:check-if-running *toppath*)
		;; (if (number? host-port-in) ;; we were handed a server-id
		;; 	   (let ((srec (tasks:get-server-by-id (db:delay-if-busy (tasks:open-db)) host-port-in)))
		;; 	     ;; (print "srec: " srec " host-port-in: " host-port-in)
		;; 	     (if srec
		;; 		 (conc (vector-ref srec 3) ":" (vector-ref srec 4))
		;; 		 (conc "no such server-id " host-port-in)))
		       host-port-in))) ;; )
    (let* ((host-port (if host:port
			  (let ((slst (string-split   host:port ":")))
			    (if (eq? (length slst) 2)
				(list (car slst)(string->number (cadr slst)))
				#f))

			  #f)))
;;	   (toppath       (launch:setup)))
      ;; (print "host-port=" host-port)
      (if (not host-port)
	  (begin
	    (if host-port-in
		(debug:print 0 *default-log-port*  "ERROR: bad host:port"))
	    (if do-exit (exit 1))
	    #f)

	  (let* ((iface      (car host-port))
		 (port       (cadr host-port))
		 (server-dat (http-transport:client-connect iface port server-id))
		 (login-res  (rmt:login-no-auto-client-setup server-dat)))

	    (if (and (list? login-res)
		     (car login-res))
		(begin
		  ;; (print "LOGIN_OK")
		  (if do-exit (exit 0))
		  #t)
		(begin
		  ;; (print "LOGIN_FAILED")
		  (if do-exit (exit 1))






		  #f)))))))

;; run ping in separate process, safest way in some cases
;;
(define (server:ping-server ifaceport)
  (with-input-from-pipe 
   (conc (common:get-megatest-exe) " -ping " ifaceport)
   (lambda ()
     (let loop ((inl (read-line))
		(res "NOREPLY"))
       (if (eof-object? inl)
	   (case (string->symbol res)
	     ((NOREPLY)  #f)
	     ((LOGIN_OK) #t)
	     (else       #f))
	   (loop (read-line) inl))))))

;; NOT USED (well, ok, reference in rpc-transport but otherwise not used).
;;
(define (server:login toppath)
  (lambda (toppath)
    (set! *db-last-access* (current-seconds)) ;; might not be needed.
    (if (equal? *toppath* toppath)
	#t
	#f)))

;; timeout is hms string: 1h 5m 3s, default is 1 minute
;; This is currently broken. Just use the number of hours with no unit.
;; Default is 60 seconds.
;;
(define (server:expiration-timeout)
  (let ((tmo (configf:lookup *configdat* "server" "timeout")))
    (if (and (string? tmo)
	     (common:hms-string->seconds tmo)) ;; BUG: hms-string->seconds is broken, if given "10" returns 0. Also, it doesn't belong in this logic unless the string->number is changed below

        (* 3600 (string->number tmo))


	60)))

(define (server:get-best-guess-address hostname)
  (let ((res #f))
    (for-each 
     (lambda (adr)
       (if (not (eq? (u8vector-ref adr 0) 127))
	   (set! res adr)))
     ;; NOTE: This can fail when there is no mention of the host in /etc/hosts. FIXME
     (vector->list (hostinfo-addresses (hostname->hostinfo hostname))))
    (string-intersperse 
     (map number->string
	  (u8vector->list
	   (if res res (hostname->ip hostname)))) ".")))

;; (define server:sync-lock-token "SERVER_SYNC_LOCK")
;; (define (server:release-sync-lock)
;;   (db:no-sync-del! *no-sync-db* server:sync-lock-token))
;; (define (server:have-sync-lock?)
;;   (let* ((have-lock-pair (db:no-sync-get-lock *no-sync-db* server:sync-lock-token))
;;          (have-lock?     (car have-lock-pair))
;;          (lock-time      (cdr have-lock-pair))
;;          (lock-age       (- (current-seconds) lock-time)))
;;     (cond
;;      (have-lock? #t)
;;      ((>lock-age
;;        (* 3 (configf:lookup-number *configdat* "server" "minimum-intersync-delay" default: 180)))
;;       (server:release-sync-lock)
;;       (server:have-sync-lock?))
;;      (else #f))))

;; moving this here as it needs access to db and cannot be in common.
;;

(define (server:get-bruteforce-syncer dbstruct #!key (fork-to-background #f) (persist-until-sync #f))
  (debug:print "WARNING: bruteforce-syncer is called but has been disabled!")
  (lambda ()
    (debug:print "WARNING: bruteforce-syncer is called but has been disabled!"))
  #;(let* ((sqlite-exe   (or (get-environment-variable "MT_SQLITE3_EXE"))) ;; defined in cfg.sh
         (sync-log     (or (args:get-arg "-sync-log") (conc *toppath* "/logs/sync-" (current-process-id) "-" (get-host-name) ".log")))
	 (tmp-area     (common:get-db-tmp-area))
	 (tmp-db       (conc tmp-area "/megatest.db"))
	 (staging-file (conc *toppath* "/.megatest.db"))
	 (mtdbfile     (conc *toppath* "/megatest.db"))
	 (lockfile     (common:get-sync-lock-filepath))
         (sync-cmd-core     (conc sqlite-exe" " tmp-db " .dump | "sqlite-exe" " staging-file "&>"sync-log))
         (sync-cmd     (if fork-to-background 
                           (conc "/usr/bin/env NBFAKE_LOG="*toppath*"/logs/last-server-sync-"(current-process-id)".log nbfake \""sync-cmd-core" && /bin/mv -f " staging-file " " mtdbfile" \"")
                           sync-cmd-core))
         (default-min-intersync-delay 2)
	 (min-intersync-delay (configf:lookup-number *configdat* "server" "minimum-intersync-delay" default: default-min-intersync-delay))
         (default-duty-cycle 0.1)
         (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)
            (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
                                              " , off-time="off-time" seconds ]")
                                 (thread-sleep! (max off-time min-intersync-delay)))
                                (else
                                 (debug:print 0 *default-log-port* "INFO: syncer thread NOT sleeping ; maybe time-to-exit...")))

                               (if (not (configf:lookup *configdat* "server" "disable-db-snapshot"))
                                   (common:snapshot-file mtdbfile subdir: ".db-snapshot"))
		               (delete-file* staging-file)
		               (let* ((start-time (current-milliseconds))
                                      (res (system sync-cmd))
                                      (dbbackupfile (conc mtdbfile ".backup"))
                                      (res2 
                                       (cond
                                        ((eq? 0 res )
                                         (handle-exceptions
                                            exn
                                            #f
                                         (if (file-exists? dbbackupfile)
		                           (delete-file* dbbackupfile)
                                         )
                                         (if (eq? 0 (file-size sync-log))
                                             (delete-file* sync-log))
		                         (system (conc "/bin/mv " staging-file " " mtdbfile))
                                         
                                         (set! last-sync-seconds (/ (- (current-milliseconds) start-time) 1000))
                                         (set! off-time (calculate-off-time
                                                         last-sync-seconds
                                                         (cond
                                                          ((and (number? duty-cycle) (> duty-cycle 0) (< duty-cycle 1))
                                                           duty-cycle)
                                                          (else
                                                           (debug:print 0 *default-log-port* "WARNING: ["(common:human-time)"] server.sync-duty-cycle is invalid.  Should be a number between 0 and 1, but "duty-cycle" was specified.  Using default value: "default-duty-cycle)
                                                           default-duty-cycle))))
                                         
                                         (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" SYNC took "last-sync-seconds" sec")
                                         (debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] pid="(current-process-id)" SYNC took "last-sync-seconds" sec ; with duty-cycle of "duty-cycle" off time is now "off-time)
                                         'sync-completed))
                                        (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"))
                                 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)
              finalres)
            ) ;; end lambda
          ))
    do-a-sync))