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 rmtmod))
(declare (uses commonmod))
(declare (uses configfmod))
(declare (uses apimod))
(declare (uses itemsmod))
(declare (uses debugprint))
(declare (uses mtver))
(declare (uses tasksmod))
(declare (uses pgdb))
(declare (uses mtargs))

	chicken.base
	chicken.condition
	chicken.sort
	chicken.time
	chicken.base
	chicken.file
	chicken.format
	chicken.process
	chicken.file.posix
	chicken.process-context.posix
	chicken.process-context
	
	(prefix sqlite3 sqlite3:)
	typed-records
	srfi-1
	srfi-13
	srfi-18
	srfi-69
	commonmod
	apimod
	itemsmod
	debugprint
	mtver
	tasksmod
	pgdb
	(prefix mtargs args:)
	dbmod
	http-transportmod
	servermod
	clientmod
	configfmod
       
	)

(defstruct alldat
  (areapath #f)
  (ulexdat  #f)
  )

	(let ((result (vector-ref res 1)))
	  (debug:print 4 *default-log-port* "Using lazy value res: " result)
	  result)
	(let ((newres (rmt:get-prereqs-not-met run-id waitons ref-item-path mode: mode itemmaps: itemmaps)))
	  (hash-table-set! *pre-reqs-met-cache* key (vector (current-seconds) newres))
	  newres))))

;;======================================================================
;; from metadat lookup MEGATEST_VERSION
;;
(define (common:get-last-run-version) ;; RADT => How does this work in send-receive function??; assume it is the value saved in some DB
  (rmt:get-var "MEGATEST_VERSION"))

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

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

;;======================================================================
;; faux-lock is deprecated. Please use simple-lock below
;;
(define (common:faux-lock keyname #!key (wait-time 8)(allow-lock-steal #t))
  (if (rmt:no-sync-get/default keyname #f) ;; do not be tempted to compare to pid. locking is a one-shot action, if already locked for this pid it doesn't actually count
      (if (> wait-time 0)
	  (begin
	    (thread-sleep! 1)
	    (if (eq? wait-time 1) ;; only one second left, steal the lock
		(begin
		  (debug:print-info 0 *default-log-port* "stealing lock for " keyname)
		  (common:faux-unlock keyname force: #t)))
	    (common:faux-lock keyname wait-time: (- wait-time 1)))
	  #f)
      (begin
        (rmt:no-sync-set keyname (conc (current-process-id)))
        (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))))

(define (common:faux-unlock keyname #!key (force #f))
  (if (or force (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))
      (begin
        (if (rmt:no-sync-get/default keyname #f) (rmt:no-sync-del! keyname))
        #t)
      #f))

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

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

;;======================================================================
;; 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 (common:api-changed?)
      (common:set-last-run-version)))

;; This login does no retries under the hood - it acts a bit like a ping.
;; Deprecated for nmsg-transport.
;;
(define (rmt:login-no-auto-client-setup connection-info)
  (rmt:send-receive-no-auto-client-setup connection-info 'login 0 (list *toppath* megatest-version *my-client-signature*)))

(define (rmt:send-receive-no-auto-client-setup connection-info cmd run-id params)
  (let* ((run-id   (if run-id run-id 0))
         (res      (handle-exceptions
                    exn
                    #f
                    (http-transport:client-api-send-receive run-id connection-info cmd params))))
    (if (and res (vector-ref res 0))
        (vector-ref res 1) ;;; YES!! THIS IS CORRECT!! CHANGE IT HERE, THEN CHANGE rmt:send-receive ALSO!!!
        #f)))

(define (std-exit-procedure)
  ;;(common:telemetry-log-close)
  (on-exit (lambda () 0))
  ;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
  (let ((no-hurry  (if (bdat-time-to-exit *bdat*) ;; hurry up
		       #f
		       (begin
			 (bdat-time-to-exit-set! *bdat* #t)
			 #t))))
    (debug:print-info 4 *default-log-port* "starting exit process, finalizing databases.")
    (if (and no-hurry (debug:debug-mode 18))
	(rmt:print-db-stats))
    (let ((th1 (make-thread (lambda () ;; thread for cleaning up, give it five seconds
                              (if *dbstruct-db* (db:close-all *dbstruct-db*)) ;; one second allocated
			      (if (bdat-task-db *bdat*)    
				  (let ((db (cdr (bdat-task-db *bdat*))))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! (bdat-task-db *bdat*) 0 #f)
					  (bdat-task-db-set! *bdat* #f)))))
                              (http-client#close-idle-connections!)
                              ;; (if (and *runremote*
                              ;;          (remote-conndat *runremote*))
                              ;;     (begin
                              ;;       (http-client#close-all-connections!))) ;; for http-client
                              (if (not (eq? *default-log-port* (current-error-port)))
                                  (close-output-port *default-log-port*))
			      (set! *default-log-port* (current-error-port))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 *default-log-port* "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
                                  (begin
                                    (thread-sleep! 5)) ;; give the clean up few seconds to do it's stuff
                                  (begin
      				  (thread-sleep! 2)))
      			      (debug:print 4 *default-log-port* " ... done")
      			      )
			    "clean exit")))
      (thread-start! th1)
      (thread-start! th2)
      (thread-join! th1)
      )
    )

  0)

;;======================================================================
;; 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?)
      (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.db"))
                (read-only (not (file-writable? dbfile)))
                (dbstruct (db:setup #t)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (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))))

(define (common:run-sync?)
    (and (common:on-homehost?)
	 (args:get-arg "-server")))




;; this one seems to be the general entry point
;;
(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:get-best (server:get-list areapath)))))
	    (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:kind-run areapath))
	    (thread-sleep! 5)
	    (loop (server:check-if-running areapath)
		  (+ try-num 1)))))))

(define (make-and-init-remote)
  (make-remote hh-dat:      (common:get-homehost)
	       server-info:  (if *toppath* (server:check-if-running *toppath*) #f)
	       server-timeout: (server:expiration-timeout)))




;; 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)))))))

;; ping the given server
;;
(define (server:check-server server-record)
  (let* ((server-url (server:record->url server-record))
         (server-id (server:record->id server-record)) 
         (res        (case *transport-type*
                       ((http)(server:ping server-url server-id))
                       ;; ((nmsg)(nmsg-transport:ping (tasks:hostinfo-get-interface server)
                       )))
    (if res
        server-url
	#f)))

;; 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) (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)
                    #f
                    (loop (car tal)(cdr tal)))))))))

;; kind start up of servers, wait 40 seconds before allowing another server for a given
;; run-id 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 3 seconds old
  (server:wait-for-server-start-last-flag areapath)
  (if (not (server:check-if-running areapath)) ;; why try if there is already a server running?
      (let* ((last-run-dat (hash-table-ref/default *server-kind-run* areapath '(0 0))) ;; callnum, whenrun
	     (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))
			      (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)
		  (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))))))

#;(define (client:connect iface port)
  (http-transport:client-connect iface port)
  #;(case (server:get-transport)
    ((rpc)  (rpc:client-connect  iface port))
    ((http) (http:client-connect iface port))
    ((zmq)  (zmq:client-connect  iface port))
    (else   (rpc:client-connect  iface port))))

(define (client:setup areapath #!key (remaining-tries 100) (failed-connects 0))
  (client:setup-http areapath remaining-tries: remaining-tries failed-connects: failed-connects)
  #;(case (server:get-transport)
    ((rpc) (rpc-transport:client-setup remaining-tries: remaining-tries failed-connects: failed-connects)) ;;(client:setup-rpc run-id))
    ((http)(client:setup-http areapath remaining-tries: remaining-tries failed-connects: failed-connects))
    (else  (rpc-transport:client-setup remaining-tries: remaining-tries failed-connects: failed-connects)))) ;; (client:setup-rpc run-id))))

;; Do all the connection work, look up the transport type and set up the
;; connection if required.
;;
;; There are two scenarios. 
;;   1. We are a test manager and we received *transport-type* and *runremote* via cmdline
;;   2. We are a run tests, list runs or other interactive process and we must figure out
;;      *transport-type* and *runremote* from the monitor.db
;;
;; client:setup
;;
;; lookup_server, need to remove *runremote* stuff
;;

(define (client:setup-http areapath #!key (remaining-tries 100) (failed-connects 0)(area-dat #f))
  (debug:print-info 2 *default-log-port* "client:setup remaining-tries=" remaining-tries)
  (server:start-and-wait areapath)
  (if (<= remaining-tries 0)
      (begin
	(debug:print-error 0 *default-log-port* "failed to start or connect to server")
	(exit 1))
      ;;
      ;; Alternatively here, we can get the list of candidate servers and work our way
      ;; through them searching for a good one.
      ;;
      (let* ((server-dat (server:get-rand-best areapath)) ;; (server:get-first-best areapath))
	     (runremote  (or area-dat *runremote*)))
	(if (not server-dat) ;; no server found
	    (client:setup-http areapath remaining-tries: (- remaining-tries 1))
	    (let ((host  (cadr  server-dat))
		  (port  (caddr server-dat))
                  (server-id (caddr (cddr server-dat))))
	      (debug:print-info 4 *default-log-port* "client:setup server-dat=" server-dat ", remaining-tries=" remaining-tries)
	      (if (and (not area-dat)
		       (not *runremote*))
                  (begin       
		    (set! *runremote* (make-and-init-remote))
                    (let* ((server-info (remote-server-info *runremote*))) 
                      (if server-info
                        (begin
                          (remote-server-url-set! *runremote* (server:record->url server-info))
                          (remote-server-id-set! *runremote* (server:record->id server-info)))))))
	      (if (and host port server-id)
		  (let* ((start-res (case *transport-type*
				      ((http)(http-transport:client-connect host port server-id))))
			 (ping-res  (case *transport-type* 
				      ((http)(rmt:login-no-auto-client-setup start-res)))))
		    (if (and start-res
			     ping-res)
			(let ((runremote (or area-dat *runremote*))) ;; it might have been generated only a few statements ago
			  (remote-conndat-set! runremote start-res) ;; (hash-table-set! runremote run-id start-res)
			  (debug:print-info 2 *default-log-port* "connected to " (http-transport:server-dat-make-url start-res))
			  start-res)
			(begin    ;; login failed but have a server record, clean out the record and try again
			  (debug:print-info 0 *default-log-port* "client:setup, login unsuccessful, will attempt to start server ... start-res=" start-res ", server-dat=" server-dat) ;; had runid.  Fixes part of Randy;s ticket 1405717332
			  (case *transport-type* 
			    ((http)(http-transport:close-connections)))
			  (remote-conndat-set! runremote #f)  ;; (hash-table-delete! runremote run-id)
			  (thread-sleep! 1)
			  (client:setup-http areapath remaining-tries: (- remaining-tries 1))
			  )))
		  (begin    ;; no server registered
		    ;; (server:kind-run areapath)
		    (server:start-and-wait areapath)
		    (debug:print-info 0 *default-log-port* "client:setup, no server registered, remaining-tries=" remaining-tries)
		    (thread-sleep! 1) ;; (+ 5 (pseudo-random-integer (- 20 remaining-tries))))  ;; give server a little time to start up, randomize a little to avoid start storms.
		    (client:setup-http areapath remaining-tries: (- remaining-tries 1)))))))))

)