Megatest

  (status 'starting)
  (trynum 0) ;; count the number of ports we've tried
  ) 

(define (servdat->url sdat)
  (conc (servdat-host sdat)":"(servdat-port sdat)))


;; generate entries for ~/.megatestrc with the following
;;
;;  grep define ../rmt.scm | grep rmt: |perl -pi -e 's/\(define\s+\((\S+)\W.*$/\1/'|sort -u

(defstruct rmt:remote
  (conns (make-hash-table)) ;; apath/dbname => rmt:conn
(defstruct remotedat
  (conns (make-hash-table)) ;; apath/dbname => conndat
  )

(defstruct rmt:conn
(defstruct conndat
  (apath    #f)
  (dbname   #f)
  (fullname #f)
  (hostport #f)
  (ipaddr   #f)
  (port     #f)
  (socket   #f)
  (srvpkt   #f)
  (srvkey   #f)
  (lastmsg  0)
  (expires  0)
  (inport   #f)
  (outport  #f))

(define *srvpktspec*
  `((server (host    . h)
	    (port    . p)
	    (servkey . k)
	    (pid     . i)
	    (ipaddr  . a)
	    (dbpath  . d))))

;;======================================================================
;;  S U P P O R T   F U N C T I O N S
;;======================================================================

;; replaces *runremote*
(define *rmt:remote* (make-rmt:remote))
(define *remotedat* (make-remotedat))

;; -> http://abc.com:900/<entrypoint>
;;
(define (rmt:conn->uri conn entrypoint)
  (conc "http://"(rmt:conn-ipaddr conn)":"(rmt:conn-port conn)"/"entrypoint))
(define (conndat->uri conn entrypoint)
  (conc "http://"(conndat-ipaddr conn)":"(conndat-port conn)"/"entrypoint))

;; set up the api proc, seems like there should be a better place for this?
(define api-proc (make-parameter conc))
(api-proc api:process-request)

;; do we have a connection to apath dbname and
;; is it not expired? then return it
;;
;; else setup a connection
;;
;; if that fails, return '(#f "some reason") ;; NB// convert to raising an exception
;;
(define (rmt:get-conn remote apath dbname)
  (let* ((fullname (db:dbname->path apath dbname)) ;; we'll switch to full name later
	 (conn     (hash-table-ref/default (rmt:remote-conns remote) dbname #f)))
(define (rmt:get-conn remdat apath dbname)
  (let* ((fullname (db:dbname->path apath dbname))
	 (conn     (hash-table-ref/default (remotedat-conns remdat) fullname #f)))
    (if (and conn
	     (< (current-seconds) (rmt:conn-expires conn)))
	     (< (current-seconds) (conndat-expires conn)))
	conn
	#f)))

(define (rmt:find-main-server apath dbname)
  (let* ((pktsdir     (get-pkts-dir apath))
	 (all-srvpkts (get-all-server-pkts pktsdir *srvpktspec*))
	 ;; (dbpath      (conc apath "/" dbname))
	 (viable-srvs (get-viable-servers all-srvpkts dbname)))
    (get-the-server apath viable-srvs)))


(define *connstart-mutex* (make-mutex))
(define *last-main-start* 0)

;; looks for a connection to main
;; looks for a connection to main, returns if have and not exired
;; creates new otherwise
;; 
;; connections for other servers happens by requesting from main
;;
;; TODO: This is unnecessarily re-creating the record in the hash table
;;
(define (rmt:open-main-connection remote apath)
  (let* ((dbname         (db:run-id->dbname #f))
	 (the-srv        (rmt:find-main-server apath dbname))
	 (start-main-srv (lambda ()
(define (rmt:open-main-connection remdat apath)
  (let* ((fullpath (db:dbname->path apath "/.db/main.db"))
	 (conns    (remotedat-conns remdat))
	 (conn     (hash-table-ref/default conns fullpath #f))) ;; TODO - create call for this
    (if (and conn                                             ;; conn is NOT a socket, just saying ...
	     (< (current-seconds) (conndat-expires conn)))
	#t ;; we are current and good to go - we'll deal elsewhere with a server that was killed or died 
	;; Below we will find or create and connect to main
	(let* ((dbname         (db:run-id->dbname #f))
	       (the-srv        (rmt:find-main-server apath dbname))
	       (start-main-srv (lambda () ;; call IF there is no the-srv found
			   ;; srv not ready, delay a little and try again
			   (api:run-server-process apath dbname)
			   (thread-sleep! 4)
			   (rmt:open-main-connection remote apath) ;; TODO: Add limit to number of tries
			   )))
    (if the-srv ;; yes, we have a server, now try connecting to it
	(let* ((srv-addr (server-address the-srv))
	       (ipaddr   (alist-ref 'ipaddr  the-srv))
	       (port     (alist-ref 'port    the-srv))
	       (srvkey   (alist-ref 'servkey the-srv))
	       (fullpath (db:dbname->path apath dbname))
	       (srvready (server-ready? ipaddr port srvkey)))
				 (mutex-lock! *connstart-mutex*)
				 (if (> (- (current-seconds) *last-main-start*) 5) ;; at least four seconds since last attempt to start main server
				     (begin
				       (api:run-server-process apath dbname)
				       (set! *last-main-start* (current-seconds))
				       (thread-sleep! 1)))
				 (mutex-unlock! *connstart-mutex*)
				 (rmt:open-main-connection remdat apath) ;; TODO: Add limit to number of tries
				 )))
	  (if (not the-srv) ;; have server, try connecting to it
	      (start-main-srv)
	      (let* ((srv-addr (server-address the-srv)) ;; need serv
		     (ipaddr   (alist-ref 'ipaddr  the-srv))
		     (port     (alist-ref 'port    the-srv))
		     (srvkey   (alist-ref 'servkey the-srv))
		     (fullpath (db:dbname->path apath dbname))
		     
	  (if srvready
	      (begin
		(hash-table-set! (rmt:remote-conns remote)
				 dbname ;; fullpath ;; yes, I'd prefer it to be fullpath - FIXME later
				 (make-rmt:conn
				  apath:   apath
				  dbname:  dbname
				  fullname: fullpath
				  hostport: srv-addr
				  ipaddr: ipaddr
				  port: port
				  srvpkt: the-srv
				  srvkey: srvkey ;; generated by rmt:get-signature on the server side
				  lastmsg: (current-seconds)
				  expires: (+ (current-seconds) 60) ;; this needs to be gathered during the ping
				  ))
		#t)
		     (new-the-srv (make-conndat
				   apath:   apath
				   dbname:  dbname
				   fullname: fullpath
				   hostport: srv-addr
				   socket: (open-nn-connection srv-addr)
				   ipaddr: ipaddr
				   port: port
				   srvpkt: the-srv
				   srvkey: srvkey ;; generated by rmt:get-signature on the server side
				   lastmsg: (current-seconds)
				   expires: (+ (current-seconds) 60) ;; this needs to be gathered during the ping
				   )))
		(hash-table-set! conns fullpath new-the-srv)))
	  #t))))
	      (start-main-srv)))
	(start-main-srv))))

;; NB// remote is a rmt:remote struct
;; NB// remdat is a remotedat struct
;;
(define (rmt:general-open-connection remote apath dbname #!key (num-tries 5))
  (let* ((mdbname (db:run-id->dbname #f))
	 (mconn   (rmt:get-conn remote apath mdbname)))
(define (rmt:general-open-connection remdat apath dbname #!key (num-tries 5))
  (assert (not (equal? dbname ".db/main.db")) "ERROR: general-open-connection should never be called with main as the db")
  (let* ((mdbname  (db:run-id->dbname #f))
	 (fullname (db:dbname->path apath dbname))
	 (conns    (remotedat-conns remdat))
	 (mconn    (rmt:get-conn remdat apath mdbname)))
    (if (and mconn
	     (not (debug:print-logger)))
	(if (equal? dbname ".db/main.db")
	    (debug:print-info 0 *default-log-port* "Not turning on logging to main, I am main!")
	    (begin
	      (debug:print-info 0 *default-log-port* "Turning on logging to main, look in logs dir for main log.")
	      (debug:print-logger rmt:log-to-main))))
    (cond
     ((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)
      (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))))
	  (< (conndat-expires mconn)(+ (current-seconds) 2))) ;; restablish connection if less than 2 seconds on the lease
      (rmt:open-main-connection remdat apath)
      (rmt:general-open-connection remdat apath mdbname))
     ((not (rmt:get-conn remdat apath dbname))                 ;; no channel open to dbname?     
      (let* ((res (rmt:send-receive-real remdat 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)))
		 (rmt:general-open-connection remdat 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
		  ((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
		   (hash-table-set! conns
				    fullname
				    (make-conndat
				     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)))))))
     

     )))
     #t))

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

;; FOR DEBUGGING SET TO #t
(define *localmode* #t)
;; (define *localmode* #t)
(define *localmode* #f)
(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)))
  ;; (if (not *remotedat*)(set! *remotedat* (make-remotedat)))
  (let* ((apath      *toppath*)
	 (conns      *rmt:remote*)
	 (remdat     *remotedat*)
	 (conns      (remotedat-conns remdat)) ;; just checking that remdat is a remotedat
	 (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)
	  ;; (api:process-request dbdat indat)
	  )
	(begin
	  (if rid
	  (rmt:general-open-connection conns apath dbname)
	  (rmt:send-receive-real conns apath dbname cmd params)))))
	      (rmt:general-open-connection remdat apath dbname) ;; was conns
	      (rmt:open-main-connection remdat apath))
	  (rmt:send-receive-real remdat 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))))
  (if (not (conndat-inport conn))
      (let-values (((i o) (tcp-connect (conndat-ipaddr conn)
				       (conndat-port conn))))
	(conndat-inport-set! conn i)
	(conndat-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)))
(define (rmt:send-receive-real remdat apath dbname cmd params)
  (let* ((conn (rmt:get-conn remdat apath dbname)))
    (assert conn "FATAL: rmt:send-receive-real called without the needed channels opened")
    (let* ((key     #f)
	   (host    (rmt:conn-ipaddr conn))
	   (port    (rmt:conn-port   conn))
	   (host    (conndat-ipaddr conn))
	   (port    (conndat-port   conn))
	   (payload `((cmd    . ,cmd)
		      (key    . ,(rmt:conn-srvkey conn))
		      (key    . ,(conndat-srvkey conn))
		      (params . ,params)))
	   (res      (open-send-receive-nn (conc host":"port)
					   (sexpr->string payload))))
      (string->sexpr 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)))
;; (define (rmt:send-receive-server-start remdat apath dbname)
;;   (let* ((conn (rmt:get-conn remdat apath dbname)))
;;     (assert conn "FATAL: Unable to connect to db "apath"/"dbname)
;;     #;(let* ((res      (with-input-from-request
;; 		      (rmt:conn->uri conn "api") 
;; 		      (conndat->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, "(seconds->year-week/day-time (current-seconds))"\n=====================")

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

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

(define (rmt:kill-server run-id)
  (rmt:send-receive 'kill-server run-id (list run-id)))
  (rmt:send-receive 'kill-server #f (list run-id)))

(define (rmt:start-server run-id)
  (rmt:send-receive 'start-server 0 (list run-id)))
  (rmt:send-receive 'start-server #f (list run-id)))

(define (rmt:get-server-info apath dbname)
  (rmt:send-receive 'get-server-info #f (list apath dbname)))

;;======================================================================
;;  M I S C
;;======================================================================

(define (rmt:login run-id)
  (rmt:send-receive 'login run-id (list *toppath* megatest-version *my-signature*)))

  ;; (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:log-to-main . params)
  (rmt:send-receive 'log-to-main #f (cons #f params)))

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

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

(define (rmt:set-var run-id varname value)

(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*)
	       (remdat     *remotedat*) ;; foundation for future fix
	       (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")
	  (if (sqlite3:database? db)
	  (sqlite3:finalize! db)
	  (sqlite3:finalize! inmem)
	      (sqlite3:finalize! db)
	      (debug:print-info 0 *default-log-port* "in rmt:server-shutdown, db is not a database, not finalizing..."))
	  (if (sqlite3:database? inmem)
	      (sqlite3:finalize! inmem)
	      (debug:print-info 0 *default-log-port* "in rmt:server-shutdown, inmem is not a database, not finalizing..."))
	  (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*?
			(let* ((res (rmt:deregister-server remdat
							   *toppath*
							   (servdat-host *server-info*)   ;; iface
							   (servdat-port *server-info*)
							   (servdat-uuid *server-info*)
							   (current-process-id)
							   dbfile ;; (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)

  (args:get-arg "-server"))

;; 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. 
;;
;; conn is a conndat record
;;
(define (server:ping host port server-id #!key (do-exit #f))
  (server-ready? host port server-id))
(define (server:ping conn #!key (do-exit #f))
  (let* ((req (conndat-socket conn))
	 (srvkey (conndat-srvkey conn))
	 (msg (sexpr->string '(ping ,srvkey))))
    (send-receive-nn req msg))) ;; (server-ready? host port server-id))

;;======================================================================
;; http-transportmod.scm contents moved here
;;======================================================================

(define (http-transport:make-server-url hostport)
  (if (not hostport)

	(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)))
;;(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)

		   (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))))
	   (nng-listen rep (conc "tcp://*:" portnum))
	   rep)))))
	   rep)))) ;;)

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

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

(define (http-transport:get-time-to-cleanup)
(define (rmt:get-time-to-cleanup)
  (let ((res #f))
    (mutex-lock! *http-mutex*)
    (set! res (> (current-seconds) *http-connections-next-cleanup*))
    (mutex-unlock! *http-mutex*)
    res))

(define (http-transport:inc-requests-count)
(define (rmt:inc-requests-count)
  (mutex-lock! *http-mutex*)
  (set! *http-requests-in-progress* (+ 1 *http-requests-in-progress*))
  ;; Use this opportunity to slow things down iff there are too many requests in flight
  (if (> *http-requests-in-progress* 5)
      (begin
	(debug:print-info 0 *default-log-port* "Whoa there buddy, ease up...")
	(thread-sleep! 1)))
  (mutex-unlock! *http-mutex*))

(define (http-transport:dec-requests-count proc) 
(define (rmt:dec-requests-count proc) 
  (mutex-lock! *http-mutex*)
  (proc)
  (set! *http-requests-in-progress* (- *http-requests-in-progress* 1))
  (mutex-unlock! *http-mutex*))

(define (http-transport:dec-requests-count-and-close-all-connections)
(define (rmt:dec-requests-count-and-close-all-connections)
  (set! *http-requests-in-progress* (- *http-requests-in-progress* 1))
  (let loop ((etime (+ (current-seconds) 5))) ;; give up in five seconds
    (if (> *http-requests-in-progress* 0)
	(if (> etime (current-seconds))
	    (begin
	      (thread-sleep! 0.052)
	      (loop etime))
	    (debug:print-error 0 *default-log-port*
			       "requests still in progress after 5 seconds of waiting. I'm going to pass on cleaning up http connections"))
	#;(close-idle-connections!)))
  (set! *http-connections-next-cleanup* (+ (current-seconds) 10))
  (mutex-unlock! *http-mutex*))

(define (http-transport:inc-requests-and-prep-to-close-all-connections)
(define (rmt:inc-requests-and-prep-to-close-all-connections)
  (mutex-lock! *http-mutex*)
  (set! *http-requests-in-progress* (+ 1 *http-requests-in-progress*)))

;; careful closing of connections stored in *runremote*
;;
(define (http-transport:close-connections #!key (area-dat #f))
  (debug:print-info 0 *default-log-port* "http-transport:close-connections doesn't do anything now!"))
(define (rmt:close-connections #!key (area-dat #f))
  (debug:print-info 0 *default-log-port* "rmt:close-connections doesn't do anything now!"))
;;   (let* ((runremote  (or area-dat *runremote*))
;; 	 (server-dat (if runremote
;;                          (remote-conndat runremote)
;;                          #f))) ;; (hash-table-ref/default *runremote* run-id #f)))
;;     (if (vector? server-dat)
;; 	(let ((api-dat (http-transport:server-dat-get-api-uri server-dat)))
;; 	  (handle-exceptions

;;======================================================================
;; NEW SERVER METHOD
;;======================================================================

;; only use for main.db - need to re-write some of this :(
;;
(define (get-lock-db sdat dbfile)
  (let* ((dbh (db:open-run-db dbfile db:initialize-db))
	 (res (db:get-iam-server-lock dbh dbfile)))
(define (get-lock-db sdat dbfile port)
  (let* ((dbh (db:open-run-db dbfile db:initialize-db)) ;; open-run-db creates a standard db with schema used by all situations
	 (res (db:get-iam-server-lock dbh dbfile port)))
    ;; res => list then already locked, check server is responsive
    ;;     => #t then sucessfully got the lock
    ;;     => #f reserved for future use as to indicate something went wrong
    (match res
      ((owner_pid owner_host owner_port event_time)
       (if (server-ready? owner_host owner_port "abc")
	   #f
	   (begin
	     (debug:print 0 *default-log-port* "WARNING: stale lock - have to steal it. This may fail.")
	     (db:steal-lock-db dbh dbfile port))))
      (#t  #t) ;; placeholder so that we don't touch res if it is #t
      (else (set! res #f)))
    (sqlite3:finalize! dbh)
    res))


(define (register-server pkts-dir pkt-spec host port servkey ipaddr dbpath)
  (let* ((pkt-dat `((host    . ,host)
		    (port    . ,port)

	 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* ((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

    (if (null? tail)
	res ;; NOTE: sort by age so oldest is considered first
	(let* ((spkt (car tail)))
	  (loop (cdr tail)
		(if (equal? dbpath (alist-ref 'dbpath spkt))
		    (cons spkt res)
		    res))))))

(define (remove-pkts-if-not-alive serv-pkts)
  (filter (lambda (pkt)
	    (let* ((host (alist-ref 'host pkt))
		   (port (alist-ref 'port pkt))
		   (key  (alist-ref 'servkey  pkt))
		   (pktz (alist-ref 'Z        pkt))
		   (res  (handle-exceptions
			     exn
			     #f
			   (server-ready? host port key))))
	      (if res
		  res
		  (let* ((pktsdir (get-pkts-dir *toppath*))
			 (pktpath (conc pktsdir"/"pktz".pkt")))
		    (debug:print 0 *default-log-port* "WARNING: pkt with no server "pktpath)
		    (delete-file* pktpath)
		    #f))))
	  serv-pkts))

;; from viable servers get one that is alive and ready
;;
(define (get-the-server apath serv-pkts)
  (let loop ((tail serv-pkts))
    (if (null? tail)
	#f

;;======================================================================
;; END NEW SERVER METHOD
;;======================================================================

;; if .db/main.db check the pkts
;; 
(define (http-transport:wait-for-server pkts-dir db-file server-key)
(define (rmt:wait-for-server pkts-dir db-file server-key)
  (let* ((sdat *server-info*))
    (let loop ((start-time (current-seconds))
	       (changed    #t)
	       (last-sdat  "not this"))
      (begin ;; let ((sdat #f))
	(thread-sleep! 0.01)
	(debug:print-info 0 *default-log-port* "Waiting for server alive signature")

				  (get-host-name)
				  (servdat-port sdat) server-key
				  (servdat-host sdat) db-file))
	      ;; (set! *my-signature* (servdat-uuid sdat)) ;; replace with Z, no, stick with proper key
	      ;; now read pkts and see if we are a contender
	      (let* ((all-pkts     (get-all-server-pkts pkts-dir *srvpktspec*))
		     (viables      (get-viable-servers all-pkts db-file))
		     (alive        (remove-pkts-if-not-alive viables))
		     (best-srv     (get-best-candidate viables db-file))
		     (best-srv-key (if best-srv (alist-ref 'servkey best-srv) #f)))
		(debug:print 0 *default-log-port* "best-srv-key: "best-srv-key", server-key: "server-key)
		     (best-srv     (get-best-candidate alive db-file))
		     (best-srv-key (if best-srv (alist-ref 'servkey best-srv) #f))
		     (i-am-srv     (equal? best-srv-key server-key))
		     (delete-pkt   (lambda ()
				     (let* ((pktfile (conc (get-pkts-dir *toppath*)
							 "/" (servdat-uuid *server-info*)
							 ".pkt")))
				       (debug:print-info 0 *default-log-port* "Attempting to remove bogus pkt file "pktfile)
				       (delete-file* pktfile))))) ;; remove immediately instead of waiting for on-exit
		(debug:print 0 *default-log-port* "best-srv-key: "best-srv-key", server-key: "server-key", i-am-srv: "i-am-srv)
		;; am I the best-srv, compare server-keys to know
		(if (equal? best-srv-key server-key)
		    (if (get-lock-db sdat db-file) ;; (db:get-iam-server-lock *dbstruct-db* *toppath* run-id)
		(if i-am-srv
		    (if (get-lock-db sdat db-file (servdat-port sdat)) ;; (db:get-iam-server-lock *dbstruct-db* *toppath* run-id)
			(begin
			  (debug:print 0 *default-log-port* "I'm the server!")
			  (debug:print-info 0 *default-log-port* "I'm the server!")
			  (servdat-dbfile-set! sdat db-file)
			  (servdat-status-set! sdat 'db-locked))
			(begin
			  (debug:print 0 *default-log-port* "I'm not the server, exiting.")
			  (debug:print-info 0 *default-log-port* "I'm not the server, exiting.")
			  (bdat-time-to-exit-set! *bdat* #t)
			  (delete-pkt)
			  (thread-sleep! 0.2)
			  (exit)))
		    (begin
		      (debug:print 0 *default-log-port*
		      (debug:print-info 0 *default-log-port*
				   "Keys do not match "best-srv-key", "server-key", exiting.")
		      (bdat-time-to-exit-set! *bdat* #t)
		      (delete-file* (conc (get-pkts-dir *toppath*)
		      (delete-pkt)
					  "/" (servdat-uuid *server-info*)
					  ".pkt")) ;; remove immediately instead of waiting for on-exit
		      (thread-sleep! 0.2)
		      (exit)))
		sdat))
	    (begin ;; sdat not yet contains server info
	      (debug:print-info 0 *default-log-port* "Still waiting, last-sdat=" last-sdat)
	      (sleep 4)
	      (if (> (- (current-seconds) start-time) 120) ;; been waiting for two minutes
		  (begin
		    (debug:print-error 0 *default-log-port* "transport appears to have died, exiting server")
		    (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
(define (rmt:register-server remdat apath iface port server-key dbname)
  (remotedat-conns remdat) ;; just checking types
  (rmt:open-main-connection remdat apath) ;; we need a channel to main.db
  (rmt:send-receive-real remdat 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
(define (rmt:get-count-servers remdat apath)
  (remotedat-conns remdat) ;; just checking types
  (rmt:open-main-connection remdat apath) ;; we need a channel to main.db
  (rmt:send-receive-real remdat apath      ;; params: host port servkey pid ipaddr dbpath
			 (db:run-id->dbname #f)
			 'get-count-servers `(,apath
					      )))
			 'get-count-servers `(,apath)))

(define (rmt:get-servers-info apath)
  (rmt:send-receive 'get-servers-info #f `(,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
(define (rmt:deregister-server remdat apath iface port server-key dbname)
  (remotedat-conns remdat) ;; just checking types
  (rmt:open-main-connection remdat apath) ;; we need a channel to main.db
  (rmt:send-receive-real remdat 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))
(define (rmt: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))
      (let* ((curr-host (and *server-info* (servdat-host *server-info*)))
	     (curr-port (and *server-info* (servdat-port *server-info*))))
	;; first we verify port and interface, update *server-info* in need be.
	(cond
	 ((> tries num-tries-allowed)
	  (debug:print 0 *default-log-port* "http-transport:keep-running, giving up after trying for several minutes.")
	  (debug:print 0 *default-log-port* "rmt:keep-running, giving up after trying for several minutes.")
	  (exit 1))
	 ((not *server-info*)
	  (thread-sleep! 0.25)
	  (loop curr-host curr-port (+ tries 1)))
	 ((or (not last-host)(not last-port))
	  (debug:print 0 *default-log-port* "http-transport:keep-running, still no interface, tries="tries)
	  (debug:print 0 *default-log-port* "rmt:keep-running, still no interface, tries="tries)
	  (thread-sleep! 0.25)
	  (loop curr-host curr-port (+ tries 1)))
	 ((or (not (equal? last-host curr-host))
	      (not (equal? last-port curr-port)))
	  (debug:print-info 0 *default-log-port* "WARNING: interface changed, refreshing iface and port info")
	  (thread-sleep! 0.25)
	  (loop curr-host curr-port (+ tries 1)))

		       "SERVER STARTED: " curr-host
		       ":" curr-port
		       " AT " (current-seconds) " server signature: " *my-signature*
		       " with "(servdat-trynum *server-info*)" port changes")
	  (flush-output *default-log-port*)
	  #t))))))

;; run http-transport:keep-running in a parallel thread to monitor that the db is being 
;; run rmt:keep-running in a parallel thread to monitor that the db is being 
;; used and to shutdown after sometime if it is not.
;;
(define (rmt:keep-running dbname) 
  ;; if none running or if > 20 seconds since 
  ;; server last used then start shutdown
  ;; This thread waits for the server to come alive
  (debug:print-info 0 *default-log-port* "Starting the sync-back, keep alive thread in server")

  (let* ((remdat            *remotedat*)
  (let* ((server-start-time (current-seconds))
	 (server-start-time (current-seconds))
	 (pkts-dir          (get-pkts-dir))
	 (server-key        (rmt:get-signature)) ;; This servers key
	 (is-main           (equal? (args:get-arg "-db") ".db/main.db"))
	 (last-access       0)
	 (server-timeout    (server:expiration-timeout)))
	 (server-timeout    (server:expiration-timeout))
	 (shutdown-server-sequence (lambda (port)
				     (set! *unclean-shutdown* #f)
				     (debug:print-info 0 *default-log-port* "Starting to shutdown the server. pid="(current-process-id))
				     (rmt:server-shutdown)
				     (portlogger:open-run-close portlogger:set-port port "released")
				     (exit)))
	 (timed-out?        (lambda ()
			      (<= (+ last-access server-timeout)
				 (current-seconds)))))
    ;; 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))
	(rmt:wait-for-server pkts-dir dbname server-key)
	(rmt:wait-for-stable-interface))
    ;; this is our forever loop
    (let* ((iface             (servdat-host *server-info*))
	   (port              (servdat-port *server-info*)))
      (let loop ((count          0)
		 (bad-sync-count 0)
		 (start-time     (current-milliseconds)))
	
	(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")
	      (db:setup dbname) ;; sets *dbstruct-db* as side effect
	      (servdat-status-set! *server-info* 'db-opened)
	      ;; IFF I'm not main, call into main and register self
	      (if (not is-main)
		  (let ((res (rmt:register-server *rmt:remote*
		  (let ((res (rmt:register-server remdat
						  *toppath* iface port
						  server-key dbname)))
		    (if res ;; we are the server
			(servdat-status-set! *server-info* 'have-interface-and-db)
			(let* ((serv-info (rmt:get-server-info *toppath* dbname)))
			  (match serv-info
			    ((host port servkey pid ipaddr apath dbpath)
			     (if (not (server-ready? host port servkey))
			(begin 
			  (debug:print 0 *default-log-port* "We are not the server for "dbname", exiting.")
			  (exit)))))
				 (begin
				   (debug:print-info 0 *default-log-port* "Server registered but not alive. Removing and trying again.")
				   (rmt:deregister-server remdat apath host port servkey dbpath) ;; servkey pid ipaddr apath dbpath)
				   (loop (+ count 1) bad-sync-count start-time))))
			    (else
			     (debug:print 0 *default-log-port* "We are not the server for "dbname", exiting. Server info is: "serv-info)
			     (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 (common:low-noise-print 60 "dbstats")
	    (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
	   ((not *server-run*)
	    (debug:print-info 0 *default-log-port* "*server-run* set to #f. Shutting down.")
	    (shutdown-server-sequence port))
	   ((timed-out?)
	    (debug:print-info 0 *default-log-port* "Server timed out. seconds since last db access: " (- (current-seconds) last-access))
	    (shutdown-server-sequence port))
	   ((and *server-run*
		 (> (+ last-access server-timeout)
		 (not (timed-out?))
		    (current-seconds))
		 (if is-main
		     (> (rmt:get-count-servers *rmt:remote* *toppath*) 1)
		 #;(if is-main ;; intention here was to exit main server quickly. 
		     (> (rmt:get-count-servers remdat *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))
	    (debug:print-info 0 *default-log-port* "Starting to shutdown the server. pid="(current-process-id))
	    (rmt:server-shutdown)
	    (shutdown-server-sequence port)
	    (portlogger:open-run-close portlogger:set-port port "released")
	    (exit)
	    #;(debug:print-info 0 *default-log-port* "Sending 'quit to server, received: "
			      (open-send-receive-nn (conc iface":"port)      ;; do this here and not in server-shutdown
						    (sexpr->string 'quit)))
	    )))))))

;; Call this to start the actual server
;;

     (let ((emsg ((condition-property-accessor 'exn 'message) exn)))
       (print "ERROR: Failed to start server \"" emsg "\"")
       (exit 1))
      
     (nng-dial #;nn-bind rep (conc "tcp://*:" portnum)))
    rep))

;; open connection to server, send message, close connection
(define (open-nn-connection host-port)
;;
(define (open-send-close-nn host-port msg #!key (timeout 3) ) ;; default timeout is 3 seconds
  (let ((req  (make-req-socket 'req))
        (uri  (conc "tcp://" host-port))
  (let ((req  (make-req-socket))
        (uri  (conc "tcp://" host-port)))
        (res  #f)
        ;; (contacts (alist-ref 'contact attrib))
        ;; (mode (alist-ref 'mode attrib))
	)
    (socket-set! req 'nng/recvtimeo 2000)
    (nng-dial req uri)
    req))

(define (send-receive-nn req msg)
  (nng-send req msg)
  (nng-recv req))

(define (close-nn-connection req)
  (nng-close! req))
  
;; ;; 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  (make-req-socket 'req))
;;         (uri  (conc "tcp://" host-port))
;;         (res  #f)
;;         ;; (contacts (alist-ref 'contact attrib))
;;         ;; (mode (alist-ref 'mode attrib))
;; 	)
;;     (socket-set! req 'nng/recvtimeo 2000)
    (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)
     (nng-dial req uri)
     ;; (print "Connected to the server " )
     (nng-send req msg)
     ;; (print "Request Sent")  
     (let* ((th1  (make-thread (lambda ()
                                 (let ((resp (nng-recv req)))
                                   (nng-close! req)
                                   (set! res (if (equal? resp "ok")
;;     (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)
;;      (nng-dial req uri)
;;      ;; (print "Connected to the server " )
;;      (nng-send req msg)
;;      ;; (print "Request Sent")  
;;      (let* ((th1  (make-thread (lambda ()
;;                                  (let ((resp (nng-recv req)))
;;                                    (nng-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))))

;;                                                  #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  (make-req-socket))
        (uri  (conc "tcp://" host-port))
        (res  #f)
        (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)
     (nng-dial req uri)
     ;; (print "Connected to the server " )
     (nng-send req msg)
     ;; (print "Request Sent")  
     ;; receive code here
     ;;(print (nn-recv req))
     (let* ((th1  (make-thread (lambda ()
                                 (let ((resp (nng-recv req)))
                                   (nng-close! req)
                                   (print resp)
                                   ;; (print resp)
                                   (set! res resp)))
                               "recv thread"))
            (th2 (make-thread (lambda ()
                                (thread-sleep! timeout)
                                (thread-terminate! th1))
                             "timer thread")))
       (thread-start! th1)

;; 						  headers: '((content-type text/plain))))
;; 				  ((equal? (uri-path (request-uri (current-request))) 
;; 					   '(/ "loop-test"))
;; 				   (send-response body: (alist-ref 'data ($))
;; 						  headers: '((content-type text/plain))))
;; 				  ((equal? (uri-path (request-uri (current-request))) 
;; 					   '(/ ""))
;; 				   (send-response body: ((http-get-function 'http-transport:main-page))))
;; 				   (send-response body: ((http-get-function 'rmt:main-page))))
;; 				  ((equal? (uri-path (request-uri (current-request))) 
;; 					   '(/ "json_api"))
;; 				   (send-response body: ((http-get-function 'http-transport:main-page))))
;; 				   (send-response body: ((http-get-function 'rmt:main-page))))
;; 				  ((equal? (uri-path (request-uri (current-request))) 
;; 					   '(/ "runs"))
;; 				   (send-response body: ((http-get-function 'http-transport:main-page))))
;; 				   (send-response body: ((http-get-function 'rmt:main-page))))
;; 				  ((equal? (uri-path (request-uri (current-request))) 
;; 					   '(/ any))
;; 				   (send-response body: "hey there!\n"
;; 						  headers: '((content-type text/plain))))
;; 				  ((equal? (uri-path (request-uri (current-request))) 
;; 					   '(/ "hey"))
;; 				   (send-response body: "hey there!\n" 
;; 						  headers: '((content-type text/plain))))
;;                               ((equal? (uri-path (request-uri (current-request))) 
;; 					   '(/ "jquery3.1.0.js"))
;; 				   (send-response body: ((http-get-function 'http-transport:show-jquery))
;; 				   (send-response body: ((http-get-function 'rmt:show-jquery))
;; 						  headers: '((content-type application/javascript))))
;;                               ((equal? (uri-path (request-uri (current-request))) 
;; 					   '(/ "test_log"))
;; 				   (send-response body: ((http-get-function 'http-transport:html-test-log) $) 
;; 				   (send-response body: ((http-get-function 'rmt:html-test-log) $) 
;; 						  headers: '((content-type text/HTML))))    
;;                               ((equal? (uri-path (request-uri (current-request))) 
;; 					   '(/ "dashboard"))
;; 				   (send-response body: ((http-get-function 'http-transport:html-dboard) $)
;; 				   (send-response body: ((http-get-function 'rmt:html-dboard) $)
;; 						  headers: '((content-type text/HTML)))) 
;; 				  (else (continue))))))))