Megatest

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

;;======================================================================

(declare (unit dbmod))
(declare (uses commonmod))
(declare (uses debugprint))
(declare (uses ods))
(declare (uses configfmod))
(declare (uses margsmod))

(module dbmod
	*
	
(import commonmod)
(import debugprint)
(import ods)
(import configfmod)
(import margsmod)

(import scheme chicken data-structures extras ports)
(import
  (prefix base64 base64:)

                 (else
                  (debug:print-error 0 *default-log-port* "Unknown server/sync-method specified ("syncer") - valid values are brute-force-sync and delta-sync.")
                  (exit 1)))
                ;;(debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] Syncer started (method="syncer")")
                )))
	    (debug:print-info 13 *default-log-port* "watchdog done."))
	  (debug:print-info 13 *default-log-port* "no need for watchdog on non-homehost"))))

;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(define *watchdog* (make-thread
		    (lambda ()
		      (handle-exceptions
			  exn
			  (begin
			    (print-call-chain)
			    (print " message: " ((condition-property-accessor 'exn 'message) exn) ", exn=" exn))
			(common:watchdog)))
		    "Watchdog thread"))


;;======================================================================
;; 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 
   dbstruct
   'schema
   ;; 'new2old
   'killservers
   'adj-target
   ;; 'old2new
   'new2old
   ;; (if full
       '(dejunk)
       ;; '())
       )
  (if (db:api-changed? dbstruct)
      (db:set-last-run-version dbstruct)))

;;======================================================================
;; 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 (common:on-homehost?)
      (let ((dbstruct (db:setup #t)))
	(if (db:api-changed? dbstruct)
	    (let* ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config"))
		   (dbfile  (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.db"))
		   (read-only (not (file-write-access? dbfile))))

	      (debug:print 0 *default-log-port*
			   "WARNING: Version mismatch!\n"
			   "   expected: " (common:version-signature) "\n"
			   "   got:      " (db:get-var dbstruct "MEGATEST_VERSION")) ;; (common:get-last-run-version))
	      (cond
	       ((get-environment-variable "MT_SKIP_DB_MIGRATE") #t)
	       ((and (common:file-exists? mtconf) (common:file-exists? dbfile) (not read-only)
		     (eq? (current-user-id)(file-owner mtconf))) ;; safe to run -cleanup-db
		(debug:print 0 *default-log-port* "   I see you are the owner of megatest.config, attempting to cleanup and reset to new version")
		(handle-exceptions
		 exn
		 (begin
		   (debug:print 0 *default-log-port* "Failed to switch versions. exn=" exn)
		   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
		   (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.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.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))))))))

;;======================================================================
;;      (begin
;;	(debug:print 0 *default-log-port* "ERROR: cannot migrate version unless on homehost. Exiting.")
;;	(exit 1))))

;;======================================================================
;; currently the primary job of the watchdog is to run the sync back to megatest.db from the db in /tmp

  (hash-table-ref/default 
   (or configf (read-config "megatest.config" #f #t))
   "disks" '("none" "")))

;;======================================================================
;; postive number if megatest version > db version
;; negative number if megatest version < db version

(define (db:get-last-run-version dbstruct)
  (db:get-var dbstruct "MEGATEST_VERSION"))

(define (db:set-last-run-version dbstruct)
  (db:set-var dbstruct "MEGATEST_VERSION" (common:version-signature)))

(define (db:get-last-run-version-number dbstruct)
  (string->number 
   (substring (db:get-last-run-version dbstruct) 0 6)))

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

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

(define (db:api-changed? dbstruct)
  (not (equal? (substring (->string megatest-version) 0 4)
               (substring (conc (db:get-last-run-version dbstruct)) 0 4))))

(define (server:writable-watchdog-bruteforce dbstruct)
  (thread-sleep! 1) ;; delay for startup
  (let* ((do-a-sync  (server:get-bruteforce-syncer dbstruct))
         (final-sync (server:get-bruteforce-syncer dbstruct fork-to-background: #t persist-until-sync: #t)))
    (when (and (not (args:get-arg "-sync-to-megatest.db")) ;; conditions under which we do not run the sync
	       (args:get-arg "-server"))

			(delay-loop (+ count 1))))
		  (if (not *time-to-exit*) (loop))))
	    ;; time to exit, close the no-sync db here
	    (db:no-sync-close-db no-sync-db stmt-cache)
	    (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) ))))))) ;;" this-wd-num="this-wd-num)))))))

;; 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))
  (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))

)