Megatest

	
	directory-utils
	;; http-client
	;; intarweb
	matchable
	md5
	message-digest
	nanomsg
	(prefix base64 base64:)
	(prefix sqlite3 sqlite3:)
	regex
	s11n
	;; spiffy
	;; spiffy-directory-listing
	;; spiffy-request-vars
	srfi-1
	srfi-13
	srfi-18
	srfi-69
	stack
	system-information
	tcp6
	;; tcp6
	typed-records
	uri-common
	z3
       
	apimod
	commonmod
	configfmod

;; info about me as a server
;;
(defstruct servdat
  (host #f)
  (port #f)
  (uuid #f)
  (rep  #f)
  (dbfile #f)
  (api-url #f)
  (api-uri #f)
  (api-req #f)
  (status 'starting)
  (trynum 0) ;; count the number of ports we've tried
  ) 

		#t)
	      (start-main-srv)))
	(start-main-srv))))

;; NB// remote is a rmt:remote struct
;;
(define (rmt:general-open-connection remote apath dbname #!key (num-tries 5))
  (let ((mdbname (db:run-id->dbname #f)))
  (let* ((mdbname (db:run-id->dbname #f))
	 (mconn   (rmt:get-conn remote apath mdbname)))
    (cond
     ((not (rmt:get-conn remote apath mdbname)) ;; no channel open to main? 
     ((or (not mconn) ;; no channel open to main?
	  (< (rmt:conn-expires mconn)(+ (current-seconds) 2))) ;; restablish connection if less than 2 seconds on the lease
      (rmt:open-main-connection remote apath)
      (thread-sleep! 2)
      (rmt:general-open-connection remote apath mdbname))
     ((not (rmt:get-conn remote apath dbname))                 ;; no channel open to dbname?     
      (let* ((res (rmt:send-receive-real remote apath mdbname 'get-server `(,apath ,dbname))))
	(case res
	  ((server-started)
	   (if (> num-tries 0)
	       (begin
		 (thread-sleep! 2)
		 (rmt:general-open-connection remote apath dbname num-tries: (- num-tries 1)))
	       (begin
		 (debug:print-error 0 *default-log-port* "Failed to start servers needed or open channel to "apath", "dbname)
		 (exit 1))))
	  (else
	   (if (list? res) ;; server has been registered and the info was returned. pass it on.
	       (begin ;;  ("192.168.0.9" 53817
		      ;;  "5e34239f48e8973b3813221e54701a01" "24310"
		      ;;  "192.168.0.9"
		      ;;  "/home/matt/data/megatest/tests/simplerun"
		 ;;  ".db/1.db")
		 (match
	       res
		  res
		  ((host port servkey pid ipaddr apath dbname)
		   (debug:print-info 0 *default-log-port* "got "res)
		   (hash-table-set! (rmt:remote-conns remote)
				    dbname
				    (make-rmt:conn
				     apath: apath
				     dbname: dbname
				     hostport: (conc host":"port)
				     ipaddr: ipaddr
				     port: port
				     srvkey: servkey
				     lastmsg: (current-seconds)
				     expires: (+ (current-seconds) 60))))
		  (else
		   (debug:print-info 0 *default-log-port* "return data from starting server did not match host port servkey pid ipaddr apath dbname " res)))
		 res)
	       (begin
		 (debug:print-info 0 *default-log-port* "Unexpected result: " res)
		 res)))))))))
		 res))))))
     

     )))

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

;; FOR DEBUGGING SET TO #t
(define *localmode* #t)
(define *dbstruct* (make-dbr:dbstruct))

;; Defaults to current area
;;
(define (rmt:send-receive cmd rid params #!key (attemptnum 1)(area-dat #f))
  (if (not *rmt:remote*)(set! *rmt:remote* (make-rmt:remote)))
  (let* ((apath *toppath*)
	 (conns *rmt:remote*)
	 (dbname (db:run-id->dbname rid)))
    (rmt:general-open-connection conns apath dbname)
    (rmt:send-receive-real conns apath dbname cmd params)))
  (let* ((apath      *toppath*)
	 (conns      *rmt:remote*)
	 (dbname     (db:run-id->dbname rid)))
    (if *localmode*
	(let* ((dbdat    (dbr:dbstruct-get-dbdat *dbstruct* dbname))
	       (indat    `((cmd . ,cmd)(params . ,params))))
	  (api:process-request *dbstruct* indat))
	(begin
	  (rmt:general-open-connection conns apath dbname)
	  (rmt:send-receive-real conns apath dbname cmd params)))))

#;(define (rmt:send-receive-setup conn)
  (if (not (rmt:conn-inport conn))
      (let-values (((i o) (tcp-connect (rmt:conn-ipaddr conn)
				       (rmt:conn-port conn))))
	(rmt:conn-inport-set! conn i)
	(rmt:conn-outport-set! conn o))))
  
;; db is at apath/.db/dbname, rid is an intermediary solution and will be removed
;; sometime in the future
;;
(define (rmt:send-receive-real remote apath dbname cmd params)
  (let* ((conn (rmt:get-conn remote apath dbname)))
    (assert conn "FATAL: rmt:send-receive-real called without the needed channels opened")
    (pp (rmt:conn->alist conn))
    (let* ((key     #f)
    ;; (rmt:send-receive-setup conn)
    (let-values (((i o) (tcp-connect (rmt:conn-ipaddr conn)
				     (rmt:conn-port conn))))
	   (host    (rmt:conn-ipaddr conn))
	   (port    (rmt:conn-port   conn))
      (let* ((key     #f)
	     (payload `((cmd    . ,cmd)
			(key    . ,(rmt:conn-srvkey conn))
			(params . ,params)))
	     (res      (begin
	   (payload `((cmd    . ,cmd)
		      (key    . ,(rmt:conn-srvkey conn))
		      (params . ,params)))
	   (res      (open-send-receive-nn (conc host":"port)
			 (write payload o) ;; (rmt:conn-outport conn))
			 (with-input-from-port
			     i ;; (rmt:conn-inport conn)
			   read))))
					   (sexpr->string payload))))
	(close-input-port i)
	(close-output-port o)
	res))))
;;  (if (string? res)
;; 	  (string->sexpr res)
      (string->sexpr res))))
;; 	  res))))



;; db is at apath/.db/dbname, rid is an intermediary solution and will be removed
;; sometime in the future.
;;
;; Purpose - call the main.db server and request a server be started
;; for the given area path and dbname
;;
;; (define (rmt:send-receive-server-start remote apath dbname)
;;   (let* ((conn (rmt:get-conn remote apath dbname)))
;;     (assert conn "FATAL: Unable to connect to db "apath"/"dbname)
;;     #;(let* ((res      (with-input-from-request
;; 		      (rmt:conn->uri conn "api") 
;; 		      `((params . (,apath ,dbname)))
;; 		      read-string)))
;;       (string->sexpr res))))

(define (rmt:print-db-stats)
  (let ((fmtstr "~40a~7-d~9-d~20,2-f")) ;; "~20,2-f"
    (debug:print 18 *default-log-port* "DB Stats\n========")
    (debug:print 18 *default-log-port* "DB Stats, "(seconds->year-week/day-time (current-seconds))"\n=====================")
    (debug:print 18 *default-log-port* (format #f "~40a~8a~10a~10a" "Cmd" "Count" "TotTime" "Avg"))
    (for-each (lambda (cmd)
		(let ((cmd-dat (hash-table-ref *db-stats* cmd)))
		  (debug:print 18 *default-log-port* (format #f fmtstr cmd (vector-ref cmd-dat 0) (vector-ref cmd-dat 1) (/ (vector-ref cmd-dat 1)(vector-ref cmd-dat 0))))))
	      (sort (hash-table-keys *db-stats*)
		    (lambda (a b)
		      (> (vector-ref (hash-table-ref *db-stats* a) 0)

;; Use the special run-id == #f scenario here since there is no run yet
(define (rmt:register-run keyvals runname state status user contour)
  ;; first register in main.db (thus the #f)
  (let* ((run-id (rmt:send-receive 'register-run #f (list keyvals runname state status user contour))))
    ;; now register in the run db itself

    ;; NEED A RECORD INSERT INCLUDING SETTING id
    (rmt:send-receive 'register-run run-id (list keyvals runname state status user contour))
    (rmt:send-receive 'insert-run run-id (list run-id keyvals runname state status user contour))
    
    run-id))
  
(define (rmt:get-run-name-from-id run-id)
  (rmt:send-receive 'get-run-name-from-id run-id (list run-id)))

(define (rmt:delete-run run-id)
  (rmt:send-receive 'delete-run run-id (list run-id)))

  ;; (if (rmt:send-receive 'have-incompletes? run-id (list run-id ovr-deadtime))
  (rmt:send-receive 'mark-incomplete run-id (list run-id ovr-deadtime))
  ) ;; )

(define (rmt:get-main-run-stats run-id)
  (rmt:send-receive 'get-main-run-stats #f (list run-id)))

(define (rmt:get-var varname)
  (rmt:send-receive 'get-var #f (list varname)))
(define (rmt:get-var run-id varname)
  (rmt:send-receive 'get-var run-id (list run-id varname)))

(define (rmt:del-var varname)
  (rmt:send-receive 'del-var #f (list varname)))
(define (rmt:del-var run-id varname)
  (rmt:send-receive 'del-var run-id (list run-id varname)))

(define (rmt:set-var varname value)
  (rmt:send-receive 'set-var #f (list varname value)))
(define (rmt:set-var run-id varname value)
  (rmt:send-receive 'set-var run-id (list run-id varname value)))

(define (rmt:inc-var varname)
  (rmt:send-receive 'inc-var #f (list varname)))
(define (rmt:inc-var run-id varname)
  (rmt:send-receive 'inc-var #f (list run-id varname)))

(define (rmt:dec-var varname)
  (rmt:send-receive 'dec-var #f (list varname)))
(define (rmt:dec-var run-id varname)
  (rmt:send-receive 'dec-var run-id (list run-id varname)))

(define (rmt:add-var varname value)
  (rmt:send-receive 'add-var #f (list varname value)))
(define (rmt:add-var run-id varname value)
  (rmt:send-receive 'add-var run-id (list run-id varname value)))

;;======================================================================
;; M U L T I R U N   Q U E R I E S
;;======================================================================

;; Need to move this to multi-run section and make associated changes
(define (rmt:find-and-mark-incomplete-all-runs #!key (ovr-deadtime #f))

;;  1. Do a remote call to get the test path
;;  2. Continue as above
;; 
;;(define (rmt:get-steps-for-test run-id test-id)
;;  (rmt:send-receive 'get-steps-data run-id (list test-id)))

(define (rmt:teststep-set-status! run-id test-id teststep-name state-in status-in comment logfile)
  (let* ((valid-values (configf:lookup *configdat* "validvalues" "state"))
  (let* ((state     (items:check-valid-items "state" state-in))
	 (status    (items:check-valid-items "status" status-in)))
	 (state        (items:check-valid-items valid-values "state" state-in))
	 (status       (items:check-valid-items valid-values "status" status-in)))
    (if (or (not state)(not status))
	(debug:print 3 *default-log-port* "WARNING: Invalid " (if status "status" "state")
		     " value \"" (if status state-in status-in) "\", update your validvalues section in megatest.config"))
    (rmt:send-receive 'teststep-set-status! run-id (list run-id test-id teststep-name state-in status-in comment logfile))))


(define (rmt:delete-steps-for-test! run-id test-id)

	  (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"))
  (rmt:get-var #f "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)))
  (rmt:set-var #f "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)

(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))))

(define (rmt:server-shutdown)
  (let ((dbfile   (servdat-dbfile *server-info*)))
    (debug:print-info 0 *default-log-port* "dbfile is "dbfile)
    (if dbfile
	(let* ((am-server  (args:get-arg "-server"))
	       (dbfile     (args:get-arg "-db"))
	       (apath      *toppath*)
	       (dbdat      (db:get-dbdat *dbstruct-db* apath dbfile))
	       (db         (dbr:dbdat-db dbdat))
	       (inmem      (dbr:dbdat-db dbdat))
	       )
	  ;; do a final sync here
	  (debug:print-info 0 *default-log-port* "Doing final sync for "apath" "dbfile" at "(current-seconds))
	  (db:sync-inmem->disk *dbstruct-db* apath dbfile force-sync: #t)
	  ;; let's finalize here
	  (debug:print-info 0 *default-log-port* "Finalizing db and inmem")
	  (sqlite3:finalize! db)
	  (sqlite3:finalize! inmem)
	  (debug:print-info 0 *default-log-port* "Finalizing db and inmem complete")
	  (if am-server
	      (if (string-match ".*/main.db$" dbfile)
		  (let ((pkt-file (conc (get-pkts-dir *toppath*)
					"/" (servdat-uuid *server-info*)
					".pkt")))
		    (debug:print-info 0 *default-log-port* "removing pkt "pkt-file)
		    (delete-file* pkt-file)
		    (debug:print-info 0 *default-log-port* "Releasing lock for "dbfile)
		    (db:with-lock-db (servdat-dbfile *server-info*)
				     (lambda (dbh dbfile)
				       (db:release-lock dbh dbfile))))
		  (let* ((sdat *server-info*) ;; we have a run-id server
			 (host (servdat-host sdat))
			 (port (servdat-port sdat))
			 (uuid (servdat-uuid sdat)))
		    (if (not (string-match ".db/main.db" (args:get-arg "-db")))
			(let* ((res (rmt:deregister-server *rmt:remote* ;; TODO/BUG: why is this requiring *rmt:remote*?
							   *toppath*
							   (servdat-host *server-info*)   ;; iface
							   (servdat-port *server-info*)
							   (servdat-uuid *server-info*)
							   (current-process-id)
							   )))
			  (debug:print-info 0 *default-log-port* "deregistered-server, res="res)))
		    
		    (debug:print-info 0 *default-log-port* "deregistering server "host":"port" with uuid "uuid)
		    )))))))

(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
    (let ((th1 (make-thread (lambda () ;; thread for cleaning up, give it five seconds
                              (if *server-info*
		(lambda () ;; thread for cleaning up, give it five seconds
		  (let* ((start-time (current-seconds)))
		    (if (and *server-info*
				  (let ((pkt-file (conc (get-pkts-dir *toppath*)
							"/" (servdat-uuid *server-info*)
							".pkt"))
					(dbfile   (servdat-dbfile *server-info*)))
				    (if dbfile
					(begin

			     *unclean-shutdown*)
					  ;; do a final sync here
					  
					(if (string-match ".*/main.db$" dbfile)
					    (begin
					      (debug:print-info 0 *default-log-port* "removing pkt "pkt-file)
			(begin
			  (debug:print-info 0 *default-log-port* "Unclean server exit, calling server-shtudown")
					      (delete-file* pkt-file)
					      (debug:print-info 0 *default-log-port* "Releasing lock for "dbfile)
			  (rmt:server-shutdown)))
		    (debug:print-info 0 *default-log-port* "Shutdown activities completed in "(- (current-seconds) start-time)" seconds"))
					      (db:with-lock-db (servdat-dbfile *server-info*)
							       (lambda (dbh dbfile)
								 (db:release-lock dbh dbfile))))
					    (let* ((sdat *server-info*)) ;; we have a run-id server
					      (rmt:send-receive-real *rmt:remote* *toppath*
								     (db:run-id->dbname #f)
								     'deregister-server
								     `(,(servdat-uuid sdat)
								       ,(current-process-id)
								       ,(servdat-host sdat)   ;; iface
								       ,(servdat-port sdat)))))))))
			      (if *dbstruct-db* (db:close-all *dbstruct-db*)) ;; one second allocated
			      (if (bdat-task-db *bdat*)    ;; TODO: Check that this is correct for task db
				  (let ((db (cdr (bdat-task-db *bdat*))))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  (bdat-task-db-set! *bdat* #f)))))
                              #;(http-client#close-idle-connections!)
                              (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"))
		  ;; (if *dbstruct-db* (db:close-all *dbstruct-db*)) ;; one second allocated
		  #;(if (bdat-task-db *bdat*)    ;; TODO: Check that this is correct for task db
		  (let ((db (cdr (bdat-task-db *bdat*))))
		  (if (sqlite3:database? db)
		  (begin
		  (debug:print-info 0 *default-log-port* "Closing down task db "db)
		  (sqlite3:interrupt! db)
		  (sqlite3:finalize! db #t)
		  (bdat-task-db-set! *bdat* #f)))))
		  #;(http-client#close-idle-connections!)
		  (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...")
			      (debug:print 4 *default-log-port* "Attempting clean exit. Mode="(if no-hurry "no-hurry" "normal")
					   " 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)))
				    (thread-sleep! 2)))
      			      (debug:print 4 *default-log-port* " ... done")
      			      )
			    "clean exit")))
      (thread-start! th1)
      (thread-start! th2)
      (thread-join! th1)
      )

      #f
      (conc "http://" (car hostport) ":" (cadr hostport))))

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

;; NOTE: http-transport:launch is the entry point
;;          -> http-transport:run
;;             -> http-transport:try-start-server -> http-transport:try-start-server (until success)

(define (http-get-function fnkey)
  (hash-table-ref/default *http-functions* fnkey (lambda () "nothing here yet")))

#;(define (rmt:launch-server hostn port)
   (if *server-info*
	(begin
	  (servdat-host-set! *server-info* hostn)
	  (servdat-port-set! *server-info* port)
	  (servdat-status-set! *server-info* 'trying-port)
	  (servdat-trynum-set! *server-info* (+ (servdat-trynum *server-info*) 1)))
	(set! *server-info* (make-servdat host: ipaddrstr port: portnum)))
   (let* ((l        (tcp-listen port))
	 (dbstruct #f))
    (let-values (((i o) (tcp-accept l)))
      ;; (write-line "Hello!" o)
      (let loop ((indat (read i)))
	(let* ((res (api:process-request dbstruct indat)))
	  (case res
	    ((quit)
	     (close-input-port i)
	     (close-output-port o))
	    (else
	     (write res o))))))))

(define (rmt:run hostn)
  ;;  ;; Configurations for server
  ;;  (tcp-buffer-size 2048)
  ;;  (max-connections 2048) 
  (debug:print 2 *default-log-port* "Attempting to start the server ...")
  (let* ((db              #f) ;;        (open-db)) ;; we don't want the server to be opening and closing the db unnecesarily
	 (hostname        (get-host-name))
	 (ipaddrstr       (let ((ipstr (if (string=? "-" hostn)
					   ;; (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
					   (server:get-best-guess-address hostname)
					   #f)))
			    (if ipstr ipstr hostn))) ;; hostname))) 
	 (port            (portlogger:open-run-close portlogger:find-port))
	 (link-tree-path  (common:get-linktree))
	 (tmp-area        (common:get-db-tmp-area))
	 ;; (tmp-area        (common:get-db-tmp-area))
	 #;(start-file      (conc tmp-area "/.server-start")))
    (debug:print-info 0 *default-log-port* "portlogger recommended port: " port)
    (if *server-info*
	(begin
	  (servdat-host-set! *server-info* ipaddrstr)
	  (servdat-port-set! *server-info* port)
	  (servdat-status-set! *server-info* 'trying-port)
	  (servdat-trynum-set! *server-info* (+ (servdat-trynum *server-info*) 1)))
	(set! *server-info* (make-servdat host: ipaddrstr port: port)))
    (let* ((l        (rmt:try-start-server ipaddrstr port)))
      (let oloop ()
	(let-values (((i o) (tcp-accept l)))
	  ;; (write-line "Hello!" o)
	  (let loop ((indat (read i)))
	    (if (eof-object? indat)
		(begin
    (let* ((rep (rmt:try-start-server ipaddrstr port)))
      (let loop ((instr (nn-recv rep)))
	(let* ((data   (string->sexpr instr))
	       (res    (case data
			 ((quit) 'quit)
			 (else    (api:process-request *dbstruct-db* data))))
	       (resdat (sexpr->string res)))
	  (if (not (eq? res 'quit))
	      (begin
		  (close-input-port i)
		  (close-output-port o)
		  (oloop))
		(let* ((res (api:process-request *dbstruct-db* indat)))
		  (set! *db-last-access* (current-seconds))
		  (write res o)
		  (loop (read i))))))))
		(set! *db-last-access* (current-seconds))
		(nn-send rep resdat)
		(loop (nn-recv rep)))))))
    (debug:print-info 0 *default-log-port* "After server, should never see this")
    ;; server exit stuff here
    (let* ((portnum (servdat-port *server-info*)))
      (portlogger:open-run-close portlogger:set-port portnum "released")
      (rmt:server-shutdown)
      ;; (bdat-time-to-exit-set! *bdat* #t) ;; tell on-exit to be fast as we've already cleaned up
      (portlogger:open-run-close portlogger:set-port port "released") ;; done in rmt:run
      ;; (debug:print-info 0 *default-log-port* "Max cached queries was    " *max-cache-size*)
      ;; (debug:print-info 0 *default-log-port* "Number of cached writes   " *number-of-writes*)
      ;; (debug:print-info 0 *default-log-port* "Average cached write time "
      ;; 		      (if (eq? *number-of-writes* 0)
      ;; 			  "n/a (no writes)"
      ;; 			  (/ *writes-total-delay*
      ;; 			     *number-of-writes*))
      ;; 		      " ms")
      ;; (debug:print-info 0 *default-log-port* "Number non-cached queries "  *number-non-write-queries*)
      ;; (debug:print-info 0 *default-log-port* "Average non-cached time   "
      ;; 		      (if (eq? *number-non-write-queries* 0)
      ;; 			  "n/a (no queries)"
      ;; 			  (/ *total-non-write-delay* 
      ;; 			     *number-non-write-queries*))
      ;; 		      " ms")
      (debug:print 1 *default-log-port* "INFO: server has been stopped"))))
      
      (db:print-current-query-stats)
      (debug:print-info 0 *default-log-port* "Server shutdown complete. Exiting")
      )))

(define (rmt:try-start-server ipaddrstr portnum)
  (if *server-info*
  (if *server-info* ;; update the server info as we might be trying next port
      (begin
	(servdat-host-set! *server-info* ipaddrstr)
	(servdat-port-set! *server-info* portnum)
	(servdat-status-set! *server-info* 'trying-port)
	(servdat-trynum-set! *server-info* (+ (servdat-trynum *server-info*) 1)))
	(servdat-trynum-set! *server-info*
			     (+ (servdat-trynum *server-info*) 1)))
      (set! *server-info* (make-servdat host: ipaddrstr port: portnum)))
  (debug:print-info 0 *default-log-port* "rmt:try-start-server time="
		    (seconds->time-string (current-seconds))
		    " ipaddrsstr=" ipaddrstr
		    " portnum=" portnum)
  (if (is-port-in-use portnum)
      (begin
	(portlogger:open-run-close portlogger:set-failed portnum)
	(debug:print 0 *default-log-port* "WARNING: failed to start on portnum: " portnum ", trying next port")
	;; (thread-sleep! 0.1)
	(rmt:try-start-server ipaddrstr
			      (portlogger:open-run-close
			       portlogger:find-port)))
      (begin
	(if (not *server-info*)
	    (set! *server-info* (make-servdat
				 host: ipaddrstr
				 port: portnum)))
	(servdat-status-set! *server-info* 'starting)
	(servdat-port-set!   *server-info* portnum)
	(if (not (servdat-rep *server-info*))
	    (servdat-rep-set!    *server-info*  (nn-socket 'rep)))
	(let* ((rep (servdat-rep *server-info*)))
	  (debug:print 0 *default-log-port* "INFO: Trying to start server on " ipaddrstr ":" portnum)
  (handle-exceptions
   exn
   (begin
     (print-error-message exn)
     (if (< portnum 64000)
	 (begin 
	   (debug:print 0 *default-log-port* "WARNING: attempt to start server failed. Trying again ...")
	   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	   (debug:print 5 *default-log-port* "exn=" (condition->list exn))
	   (portlogger:open-run-close portlogger:set-failed portnum)
	   (debug:print 0 *default-log-port* "WARNING: failed to start on portnum: " portnum ", trying next port")
	   ;; (thread-sleep! 0.1)
	   (rmt:try-start-server ipaddrstr
				 (portlogger:open-run-close portlogger:find-port)))
	 (begin
	   (print "ERROR: Tried and tried but could not start the server"))))
	  (handle-exceptions
	   exn
	   (begin
	     (print-error-message exn)
	     (if (< portnum 64000)
		 (begin 
		   (debug:print 0 *default-log-port* "WARNING: attempt to start server failed. Trying again ...")
		   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
		   (debug:print 5 *default-log-port* "exn=" (condition->list exn))
		   (portlogger:open-run-close portlogger:set-failed portnum)
		   (debug:print 0 *default-log-port* "WARNING: failed to start on portnum: " portnum ", trying next port")
		   ;; (thread-sleep! 0.1)
		   (rmt:try-start-server ipaddrstr
					 (portlogger:open-run-close portlogger:find-port)))
		 (begin
		   (print "ERROR: Tried and tried but could not start the server, stopping at port "portnum))))
   ;; any error in following steps will result in a retry
   (if *server-info*
       (servdat-status-set! *server-info* 'starting)
       (set! *server-info* (make-servdat host: ipaddrstr port: portnum)))
   
	   (nn-bind rep (conc "tcp://*:" portnum))
   (debug:print 0 *default-log-port* "INFO: Trying to start server on " ipaddrstr ":" portnum)
   (tcp-listen portnum)))
	   rep)))))

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

;;======================================================================
;; C L I E N T S

	 all-pkt-files)))

(define (server-address srv-pkt)
  (conc (alist-ref 'host srv-pkt) ":"
	(alist-ref 'port srv-pkt)))
	
(define (server-ready? host port key) ;; server-address is host:port
  (let-values (((i o)(handle-exceptions
		      exn
		      (values #f #f)
		      (tcp-connect host port))))
    (if (and i o)
;;  (let-values (((i o)(handle-exceptions
;;		      exn
;;		      (values #f #f)
;;		      (tcp-connect host port))))
;;    (if (and i o)
	(begin
	  (write `((cmd . ping)
		   (key . ,key)
		   (params . ())) o)
	  (let ((res (with-input-from-port i
		       read)))
	    (close-output-port o)
	    (close-input-port i)
	    res))
  (let* ((data (sexpr->string  `((cmd . ping)
				 (key . ,key)
				 (params . ()))))
	 (res  (open-send-receive-nn (conc host ":" port) data)))
    (string->sexpr res)))

;; (let ((res (with-input-from-port i
;; 		       read)))
;; 	    (close-output-port o)
;; 	    (close-input-port i)
;; 	    res))
;; 	    (if (string? res)
;; 		(string->sexpr res)
;; 		res)))
	(begin ;; connection failed
	  (debug:print-info 0 *default-log-port* "Server at "host":"port" is not responding.")
	  #f))))
 	      
;;     (begin ;; connection failed
;; 	  (debug:print-info 0 *default-log-port* "Server at "host":"port" is not responding.")
;; 	  #f))))

;; (define (loop-test host port data) ;; server-address is host:port
;;   ;; ping the server and ask it
;;   ;; if it ready
;;   ;; (let* ((sdat (servdat-init #f host port #f)))
;;   ;;   (http-transport:send-receive sdat "abc" 'ping '())))
;;   (let* ((payload (sexpr->string data))
;; 	 (res     (with-input-from-request

		    (exit))
		  (loop start-time
			(equal? sdat last-sdat)
			sdat))))))))

(define (rmt:register-server remote apath iface port server-key dbname)
  (rmt:open-main-connection remote apath) ;; we need a channel to main.db
  (rmt:send-receive-real remote apath ;; params: host port servkey pid ipaddr dbpath
			 (db:run-id->dbname #f) 'register-server `(,iface
								   ,port
								   ,server-key
								   ,(current-process-id)
								   ,iface
								   ,apath
								   ,dbname)))
  (rmt:send-receive-real remote apath      ;; params: host port servkey pid ipaddr dbpath
			 (db:run-id->dbname #f)
			 'register-server `(,iface
					    ,port
					    ,server-key
					    ,(current-process-id)
					    ,iface
					    ,apath
					    ,dbname)))

(define (rmt:get-count-servers remote apath)
  (rmt:open-main-connection remote apath) ;; we need a channel to main.db
  (rmt:send-receive-real remote apath      ;; params: host port servkey pid ipaddr dbpath
			 (db:run-id->dbname #f)
			 'get-count-servers `(,apath
					      )))

(define (rmt:deregister-server remote apath iface port server-key dbname)
  (rmt:open-main-connection remote apath) ;; we need a channel to main.db
  (rmt:send-receive-real remote apath      ;; params: host port servkey pid ipaddr dbpath
                         (db:run-id->dbname #f)
                         'deregister-server `(,iface
                                              ,port
                                              ,server-key
                                              ,(current-process-id)
                                              ,iface
                                              ,apath
                                              ,dbname)))

(define (http-transport:wait-for-stable-interface #!optional (num-tries-allowed 100))
  ;; wait until *server-info* stops changing
  (let* ((stime (current-seconds)))
    (let loop ((last-host  #f)
	       (last-port  #f)
	       (tries 0))

    ;; main and run db servers have both got wait logic (could/should merge it)
    (if is-main
	(http-transport:wait-for-server pkts-dir dbname server-key)
	(http-transport:wait-for-stable-interface))
    ;; this is our forever loop
    (let* ((iface             (servdat-host *server-info*))
	   (port              (servdat-port *server-info*)))
      (let loop ((count         0)
      (let loop ((count          0)
		 (bad-sync-count 0)
		 (start-time     (current-milliseconds)))
	
	(if (not is-main)
	(if (and (not is-main)
		 (common:low-noise-print 60 "servdat-status"))
	    (debug:print-info 0 *default-log-port* "servdat-status is " (servdat-status *server-info*)))

	;; set up the database handle
	(mutex-lock! *heartbeat-mutex*)
	(if (not *dbstruct-db*) ;; no db opened yet, open the db and register with main if appropriate
	    (let ((watchdog (bdat-watchdog *bdat*)))		 
	      (debug:print 0 *default-log-port* "SERVER: dbprep")

			(begin 
			  (debug:print 0 *default-log-port* "We are not the server for "dbname", exiting.")
			  (exit)))))
	      (debug:print 0 *default-log-port*
			   "SERVER: running, db "dbname" opened, megatest version: "
			   (common:get-full-version))
	      ;; start the watchdog

	      ;; is this really needed?
	      
	      (if watchdog
	      #;(if watchdog
		  (if (not (member (thread-state watchdog)
				   '(ready running blocked
					   sleeping dead)))
		      (begin
			(debug:print-info 0 *default-log-port* "Starting watchdog thread (in state "(thread-state watchdog)")")
			(thread-start! watchdog))
		      (debug:print-info 0 *default-log-port* "Not starting watchdog thread (in state "(thread-state watchdog)")"))
		  (debug:print 0 *default-log-port* "ERROR: *watchdog* not setup, cannot start it."))
	      #;(loop (+ count 1) bad-sync-count start-time)))
	      #;(loop (+ count 1) bad-sync-count start-time)
	      ))

	(debug:print-info 0 *default-log-port* "syncing "*toppath*" "dbname" at "(current-seconds))
	(db:sync-inmem->disk *dbstruct-db* *toppath* dbname force-sync: #t)
	
	(mutex-unlock! *heartbeat-mutex*)
	
	;; 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)))

	    (begin
	      (debug:print 0 *default-log-port* "Server stats:")
	      (db:print-current-query-stats)))
	(let* ((hrs-since-start  (/ (- (current-seconds) server-start-time) 3600)))
	  (cond
	   ((and *server-run*
		 (> (+ last-access server-timeout)
		    (current-seconds)))
		    (current-seconds))
		 (if is-main
		     (> (rmt:get-count-servers *rmt:remote* *toppath*) 1)
		     #t))
	    (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)))
	    (loop 0 bad-sync-count (current-milliseconds)))
	   (else
	    (set! *unclean-shutdown* #f)
	    (debug:print-info 0 *default-log-port* "Server timed out. seconds since last db access: " (- (current-seconds) last-access))
	    (http-transport:server-shutdown port))))))))

(define (http-transport:server-shutdown port)
  (begin
    ;;(BB> "http-transport:server-shutdown called")
    (debug:print-info 0 *default-log-port* "Starting to shutdown the server. pid="(current-process-id))
	    (debug:print-info 0 *default-log-port* "Starting to shutdown the server. pid="(current-process-id))
    ;;
    ;; start_shutdown
	    (rmt:server-shutdown)
    ;;

    ;; deregister the server

    
    (bdat-time-to-exit-set! *bdat* #t) ;; tell on-exit to be fast as we've already cleaned up
    (portlogger:open-run-close portlogger:set-port port "released")
	    (portlogger:open-run-close portlogger:set-port port "released")
    (thread-sleep! 1)

	    (exit)
    ;; (debug:print-info 0 *default-log-port* "Max cached queries was    " *max-cache-size*)
    ;; (debug:print-info 0 *default-log-port* "Number of cached writes   " *number-of-writes*)
    ;; (debug:print-info 0 *default-log-port* "Average cached write time "
	    #;(debug:print-info 0 *default-log-port* "Sending 'quit to server, received: "
    ;; 		      (if (eq? *number-of-writes* 0)
    ;; 			  "n/a (no writes)"
    ;; 			  (/ *writes-total-delay*
    ;; 			     *number-of-writes*))
    ;; 		      " ms")
    ;; (debug:print-info 0 *default-log-port* "Number non-cached queries "  *number-non-write-queries*)
    ;; (debug:print-info 0 *default-log-port* "Average non-cached time   "
    ;; 		      (if (eq? *number-non-write-queries* 0)
    ;; 			  "n/a (no queries)"
    ;; 			  (/ *total-non-write-delay* 
    ;; 			     *number-non-write-queries*))
    ;; 		      " ms")
    
			      (open-send-receive-nn (conc iface":"port)      ;; do this here and not in server-shutdown
    (db:print-current-query-stats)
    (common:save-pkt `((action . exit)
                       (T      . server)
						    (sexpr->string 'quit)))
                       (pid    . ,(current-process-id)))
                     *configdat* #t)
    (debug:print-info 0 *default-log-port* "Server shutdown complete. Exiting")
    (exit)))
	    )))))))

;; Call this to start the actual server
;;
;; all routes though here end in exit ...
;;
;; This is the point at which servers are started
;;
(define (rmt:server-launch dbname)
  (debug:print-info 0 *default-log-port* "Entered rmt:server-launch")
  (let* ((th2 (make-thread (lambda ()
			     (debug:print-info 0 *default-log-port* "Server run thread started")
			     (rmt: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")
			     (rmt:keep-running dbname)
			     "Keep running"))))
    (thread-start! th2)
    (thread-sleep! 0.252) ;; give the server time to settle before starting the keep-running monitor.
    (thread-start! th3)
    (set! *didsomething* #t)
    (thread-join! th2)
    (thread-join! th3)
  (exit))

    ;; (exit))
  )
  #f
  )
	    
;; Generate a unique signature for this process, used at both client and
;; server side
(define (rmt:mk-signature)
  (message-digest-string (md5-primitive) 
			 (with-output-to-string
			   (lambda ()
			     (write (list (current-directory)
                                          (current-process-id)
					  (argv)))))))

(define (rmt:get-signature) 
  (if *my-signature* *my-signature*
      (let ((sig (rmt:mk-signature)))
        (set! *my-signature* sig)
        *my-signature*)))

;;======================================================================
;; Nanomsg transport
;;======================================================================

(define (is-port-in-use port-num)
  (let* ((ret #f))
    (let-values (((inp oup pid)
		  (process "netstat" (list  "-tulpn" ))))
      (let loop ((inl (read-line inp)))
        (if (not (eof-object? inl))
            (begin 
	      (if (string-search (regexp (conc ":" port-num)) inl)
		  (begin
					;(print "Output: "  inl)
		    (set! ret  #t))
		  (loop (read-line inp)))))))
    ret))

;;start a server, returns the connection
;;
(define (start-nn-server portnum )
  (let ((rep (nn-socket 'rep)))
    (handle-exceptions
     exn
     (let ((emsg ((condition-property-accessor 'exn 'message) exn)))
       (print "ERROR: Failed to start server \"" emsg "\"")
       (exit 1))
      
     (nn-bind rep (conc "tcp://*:" portnum)))
    rep))

;; open connection to server, send message, close connection
;;
(define (open-send-close-nn host-port msg #!key (timeout 3) ) ;; default timeout is 3 seconds
  (let ((req  (nn-socket 'req))
        (uri  (conc "tcp://" host-port))
        (res  #f)
        ;; (contacts (alist-ref 'contact attrib))
        ;; (mode (alist-ref 'mode attrib))
	) 
    (handle-exceptions
     exn
     (let ((emsg ((condition-property-accessor 'exn 'message) exn)))
       ;; Send notification       
       (debug:print 0 *default-log-port* "ERROR: Failed to connect / send to " uri " message was \"" emsg "\"" )
       #f)
     (nn-connect req uri)
     ;; (print "Connected to the server " )
     (nn-send req msg)
     ;; (print "Request Sent")  
     (let* ((th1  (make-thread (lambda ()
                                 (let ((resp (nn-recv req)))
                                   (nn-close req)
                                   (set! res (if (equal? resp "ok")
                                                 #t
                                                 #f))))
                               "recv thread"))
            (th2 (make-thread (lambda ()
                                (thread-sleep! timeout)
                                (thread-terminate! th1))
			      "timer thread")))
       (thread-start! th1)
       (thread-start! th2)
       (thread-join! th1)
       res))))

(define (open-send-receive-nn host-port msg #!key (timeout 3) ) ;; default timeout is 3 seconds
  (let ((req  (nn-socket 'req))
        (uri  (conc "tcp://" host-port))
        (res  #f)
	;;        (contacts (alist-ref 'contact attrib))
        ;; (mode (alist-ref 'mode attrib))
	) 
    (handle-exceptions
     exn
     (let ((emsg ((condition-property-accessor 'exn 'message) exn)))
       ;; Send notification      
       (debug:print 0 *default-log-port* "ERROR: Failed to connect / send to " uri " message was \"" emsg "\", exn=" exn)
       #f)
     (nn-connect req uri)
     ;; (print "Connected to the server " )
     (nn-send req msg)
     ;; (print "Request Sent")  
     ;; receive code here
     ;;(print (nn-recv req))
     (let* ((th1  (make-thread (lambda ()
                                 (let ((resp (nn-recv req)))
                                   (nn-close req)
                                   (print resp)
                                   (set! res resp)))
                               "recv thread"))
            (th2 (make-thread (lambda ()
                                (thread-sleep! timeout)
                                (thread-terminate! th1))
                             "timer thread")))
       (thread-start! th1)
       (thread-start! th2)
       (thread-join! th1)
       res))))

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

;; run ping in separate process, safest way in some cases
;;