Megatest

(define (common:version-db-delta)
  (- megatest-version (common:get-last-run-version-number)))

(define (common:version-changed?)
  (not (equal? (common:get-last-run-version)
               (common:version-signature))))


;; From 1.70 to 1.80, db's are compatible.

(define (common:api-changed?)
  (let* (
    (megatest-major-version (substring (->string megatest-version) 0 4))
    (run-major-version (substring (conc (common:get-last-run-version)) 0 4))
   )
   (and (not (equal? megatest-major-version "1.80"))
     (not (equal? megatest-major-version megatest-run-version)))
  )
)

;;======================================================================
;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma
;;
(define (common:cleanup-db dbstruct #!key (full #f))
  (apply db:multi-db-sync 

	    (debug:print-info 0 *default-log-port* "Deleted " (length files) " files from logs, keeping " max-allowed " files."))))))

;;======================================================================
;; Force a megatest cleanup-db if version is changed and skip-version-check not specified
;; Do NOT check if not on homehost!
;;
(define (common:exit-on-version-changed)
  (if (and *toppath*              ;; do nothing if *toppath* not yet provided
	   (common:on-homehost?))
      (if (common:api-changed?)
	  (let* ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config"))
                 (dbfile  (conc (get-environment-variable "MT_RUN_AREA_HOME") ".megatest/main.db"))
                 (read-only (not (file-write-access? dbfile)))
                 (dbstruct (db:setup #t))) ;; (db:setup-db *dbstruct-dbs* *toppath* #f))) ;;  #t)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (common:get-last-run-version))
            (cond

                 (print-call-chain (current-error-port))
                 (exit 1))
               (common:cleanup-db dbstruct)))
             ((not (common:file-exists? mtconf))
              (debug:print 0 *default-log-port* "   megatest.config does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (common:file-exists? dbfile))
              (debug:print 0 *default-log-port* "   .megatest/main.db does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (eq? (current-user-id)(file-owner mtconf)))
              (debug:print 0 *default-log-port* "   You do not own .megatest/main.db in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (read-only
              (debug:print 0 *default-log-port* "   You have read-only access to this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (else
              (debug:print 0 *default-log-port* " to switch versions you can run: \"megatest -cleanup-db\"")
              (exit 1)))))))

  (message-digest-string (md5-primitive) str))

;;======================================================================
;; E X I T   H A N D L I N G
;;======================================================================

(define (common:run-sync?)
  (and *toppath*               ;; gate if called before *toppath* is set
       (common:on-homehost?)
       (args:get-arg "-server")))

(define (common:human-time)
  (time->string (seconds->local-time (current-seconds)) "%Y-%m-%d %H:%M:%S"))


(define (std-signal-handler signum)
  ;; (signal-mask! signum)

		  (begin ;; found a host, return it
		    (debug:print 0 *default-log-port* "INFO: Found host: " new-best " load: " load " last-used: " delta " seconds ago, with job-rate: " job-rate)
		    (host-last-used-set! rec curr-time)
		    new-best)
		  (if (null? tal) #f (loop (car tal)(cdr tal) best-host)))))))))

(define (common:wait-for-homehost-load maxnormload msg)
  (let loop ((start-time (current-seconds))) ;; we saw some instances of this being called before *toppath* was set. This might be an early setup race. This delay should help but it is impossible to test...
    (if (not *toppath*)
	(begin
	  (debug:print 0 *default-log-port* "ERROR: common:wait-for-homehost-load called before *toppath* set.")
	  (thread-sleep! 30)
	  (if (< (- (current-seconds) start-time) 300)
	      (loop start-time)))))
  (let* ((hh-dat (if (common:on-homehost?) ;; if we are on the homehost then pass in #f so the calls are local.
                     #f
                     (server:choose-server *toppath* 'homehost)))
         (hh     (if hh-dat (car hh-dat) #f)))
    (common:wait-for-normalized-load maxnormload msg hh)))

(define (common:get-num-cpus remote-host)

(define (dbfile:print-err . params)
  (with-output-to-port
      (current-error-port)
    (lambda ()
      (apply print params))))
    
(define (dbfile:cautious-open-database fname init-proc sync-mode journal-mode #!optional (tries-left 500))
  (let* ((busy-file  (conc fname "-journal"))
	 (delay-time (* (- 51 tries-left) 1.1))
      	 (write-access (file-write-access? fname))
         (dir-access (file-write-access? (pathname-directory fname)))
         (retry      (lambda ()
		       (thread-sleep! delay-time)
		       (if (> tries-left 0)
			   (dbfile:cautious-open-database fname init-proc

	  (dbfile:print-err "INFO: db:lock-and-delta-sync copying db "runid" at "(current-seconds))
	  (set! *db-sync-in-progress* #t)
	  (db:sync-touched dbstruct runid keys dbinit)
	  (set! *db-sync-in-progress* #f)
	  (delete-file* lock-file)
	  #t)
        (begin
	  (if (common:low-noise-print 120 (conc "no lock "from-db-file))
	      (dbfile:print-err "INFO: could not get lock for " from-db-file ", sync likely in progress."))
	  #f
	  ))))

;; ;; Get a lock from the no-sync-db for the from-db, then delta sync the from-db to the to-db, otherwise return #f
;; ;;
;; (define (db:lock-and-delta-sync-orig no-sync-db dbstruct from-db-file runid keys dbinit)
;;   (assert (not *db-sync-in-progress*) "FATAL: db:lock-and-sync called while a sync is in progress.")

(define dbfile:db-init-proc (make-parameter #f))

;; (db:with-db dbstruct run-id sqlite3:exec "select blah fgrom blaz;")
;; r/w is a flag to indicate if the db is modified by this query #t = yes, #f = no
;;
(define (db:with-db dbstruct run-id r/w proc . params)
  (assert dbstruct "FATAL: db:with-db called with dbstruct "#f)
  (let* ((use-mutex (> *api-process-request-count* 25)) ;; risk of db corruption
	 (have-struct (dbr:dbstruct? dbstruct))
         (dbdat     (if have-struct                ;; this stuff just allows us to call with a db handle directly
			(db:open-db dbstruct run-id (dbfile:db-init-proc)) ;; (dbfile:get-subdb dbstruct run-id)
			#f))
	 (db        (if have-struct                ;; this stuff just allows us to call with a db handle directly
			(dbr:dbdat-dbh dbdat)
			dbstruct))
	 (fname     (if dbdat
			(dbr:dbdat-dbfile dbdat)
			"nofilenameavailable"))
	 (jfile     (conc fname"-journal"))
	 #;(subdb     (if have-struct
			(dbfile:get-subdb dbstruct run-id)
			#f))
	 ) ;; was 25
    (assert (sqlite3:database? db) "FATAL: db:with-db, db is not a database, db="db", fname="fname)
    (if (file-exists? jfile)
	(begin
	  (dbfile:print-err "INFO: "jfile" exists, delaying to reduce database load")
	  (thread-sleep! 0.2)))
    (if (and use-mutex
	     (common:low-noise-print 120 "over-50-parallel-api-requests"))
	(dbfile:print-err *api-process-request-count* " parallel api requests being processed in process "

	    (set! server-going #t)
	    (debug:print 0 *default-log-port* "SERVER: running, megatest version: " (common:get-full-version))) ;; NOTE: the server is NOT yet marked as running in the log. We do that in the keep-running routine.
	  (if (and no-sync-db
		   (common:low-noise-print 10 "sync-all")) ;; cheesy way to reduce frequency of running sync :)
              (begin
		(if (common:low-noise-print 120 "sync-all-print")
                    (debug:print 0 *default-log-port* "keep-running calling db:all-db-sync at " (time->string (seconds->local-time) "%H:%M:%S")))
		(db:all-db-sync *dbstruct-dbs*)
		(if (common:low-noise-print 1800 "sync-back") ;; every half hour do a reverse sync
		    (let* ((lockfile (conc *toppath*"/.megatest/syncback.lock"))
			   (locked   (common:simple-file-lock lockfile)))
		      (if locked
			  (begin
			    (db:multi-db-sync
			     *dbstruct-dbs*
			     'dejunk
			     'adj-testids
			     'old2new)
			    (common:simple-file-release-lock lockfile)))))
		)))
      
      ;; when things go wrong we don't want to be doing the various queries too often
      ;; so we strive to run this stuff only every four seconds or so.
      (let* ((sync-time (- (current-milliseconds) start-time))
	    (rem-time  (quotient (- 4000 sync-time) 1000)))
	(if (and (<= rem-time 4)
		 (>  rem-time 0))

		  (current-seconds)))
          (if (common:low-noise-print 120 "server continuing")
              (debug:print-info 0 *default-log-port* "Server continuing, seconds since last db access: " (- (current-seconds) last-access))
	      (let ((curr-time (current-seconds)))
		(handle-exceptions
		    exn
		    (debug:print 0 *default-log-port* "ERROR: Failed to change timestamp on info file " servinfofile ". Are you out of space on that disk? exn=" exn)
		    (if (and ;; (< (- (current-seconds) server-start-time) 600) ;; run for ten minutes for experiment, 3600 thereafter
			     (not *server-overloaded*)
			     (file-exists? servinfofile))
			(change-file-times servinfofile curr-time curr-time)))
		(if (or (common:low-noise-print 120 "start new server")
			(> *api-process-request-count* 50)) ;; if this server is kind of busy start up another
		    (begin
		      (debug:print-info 0 *default-log-port* "Server is busy, parallel-api-count "*api-process-request-count*", start another if possible...")

    (exit)))

;; all routes though here end in exit ...
;;
;; start_server? 
;;
(define (http-transport:launch)
  ;; check the .servinfo directory, are there other servers running on this
  ;; or another host?
  (let* ((server-start-is-ok  (server:minimal-check *toppath*)))















    (if (not server-start-is-ok)
	(begin






	  (debug:print 0 *default-log-port* "ERROR: server start not ok, exiting now.")
	  (exit 1))))
    
  ;; check that a server start is in progress, pause or exit if so


  (let* ((th2 (make-thread (lambda ()
                             (debug:print-info 0 *default-log-port* "Server run thread started")
                             (http-transport:run 
                              (if (args:get-arg "-server")
                                  (args:get-arg "-server")
                                  "-")
                              )) "Server run"))
         (th3 (make-thread (lambda ()
                             (debug:print-info 0 *default-log-port* "Server monitor thread started")
                             (http-transport:keep-running)
                             "Keep running"))))
    (thread-start! th2)
    (thread-sleep! 0.25) ;; give the server time to settle before starting the keep-running monitor.
    (thread-start! th3)
    (set! *didsomething* #t)
    (thread-join! th2)
    (exit)))

;; (define (http-transport:server-signal-handler signum)
;;   (signal-mask! signum)
;;   (handle-exceptions
;;    exn
;;    (debug:print 0 *default-log-port* " ... exiting ...")
;;    (let ((th1 (make-thread (lambda ()

;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;; Always use two or four digit decimal
;; 1.01, 1.02...1.10,1.11,1.1101 ... 1.99,2.00..

;; (declare (unit megatest-version))

(define megatest-version 1.8002)

;; Start the server - can be done in conjunction with -runall or -runtests (one day...)
;;   we start the server if not running else start the client thread
;;======================================================================

;; Server? Start up here.
;;
(if (args:get-arg "-server")
    (let ((tl        (launch:setup)))

      ;; (server:launch 0 'http)
      (http-transport:launch)
      (set! *didsomething* #t)))

;; The adjutant is a bit different, it does NOT run (launch:setup) as it is not necessarily tied to
;; a specific Megatest area. Detail are being hashed out and this may change.
;;
(if (args:get-arg "-adjutant")
    (begin

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

;; Call this to start the actual server
;;












;;======================================================================
;; 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*

;; 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* ((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
       ;; WARNING: this is potentially dangerous to blanket ignore the errors
       (if (file-exists? logf)
	   (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)
  (assert *toppath* "FATAL: server:get-servers-info called before *toppath* has been set.")
  (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)))

;; check the .servinfo directory, are there other servers running on this
;; or another host?
;;
;; returns #t => ok to start another server
;;         #f => not ok to start another server
;;
(define (server:minimal-check areapath)
  (server:clean-up-old areapath)
  (let* ((srvdir      (conc areapath"/.servinfo"))
	 (servrs      (glob (conc srvdir"/*")))
	 (thishostip  (server:get-best-guess-address (get-host-name)))
	 (thisservrs  (glob (conc srvdir"/"thishostip":*")))
	 (homehostinf (server:choose-server areapath 'homehost))
	 (havehome    (car homehostinf))
	 (wearehome   (cdr homehostinf)))
    (debug:print-info 0 *default-log-port* thishostip", have homehost: "havehome", we are homehost: "wearehome
		      ", numservers: "(length thisservrs))
    (cond
     ((not havehome) #t) ;; no homehost yet, go for it
     ((and havehome wearehome (< (length thisservrs) 20)) #t) ;; we are home and less than 20 servers, ok to start another
     ((and havehome (not wearehome)) #f)     ;; we are not the home host
     ((and havehome wearehome (>= (length thisservrs) 20)) #f) ;; have enough running
     (else
      (debug:print 0 *default-log-port* "WARNING: Unrecognised scenario, servrs="servrs", thishostip="thishostip", thisservrs="thisservrs)
      #t))))
	 
;; 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
  ;; first we clean up old server files
  (server:clean-up-old areapath)
  (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-ten  (lambda ()
			     (if (> (length all-valid) 10)
				 (take all-valid 10)
				 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-ten)(names->dats (best-ten)))
	    ((all-valid)(names->dats all-valid))
	    ((best)     (let* ((best-ten (best-ten))
			       (len       (length best-ten)))
			  (hash-table-ref serversdat (list-ref best-ten (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))))))

(define (server:clean-up-old areapath)
  ;; any server file that has not been touched in ten minutes is effectively dead
  (let* ((sfiles (glob (conc areapath"/.servinfo/*"))))
    (for-each
     (lambda (sfile)
       (let* ((modtime (handle-exceptions
			   exn
			 (begin
			   (debug:print 0 *default-log-port* "WARNING: failed to get modification file for "sfile)
			   (current-seconds))
			 (file-modification-time sfile))))
	 (if (and (number? modtime)
		  (> (- (current-seconds) modtime)
		     600))
	     (begin
	       (debug:print 0 *default-log-port* "WARNING: found old server info file "sfile", removing it.")
	       (handle-exceptions
		   exn
		 (debug:print 0 *default-log-port* "WARNING: failed to delete old server info file "sfile)
		 (delete-file sfile))))))
     sfiles)))

;; 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) 20)
      (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

	     (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-ten))) ;; (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))