Megatest

		      (begin
			(tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "dbprep")
			(thread-sleep! 0.5) ;; give some margin for queries to complete before switching from file based access to server based access
			(set! *dbstruct-db*  (db:setup)) ;;  run-id))
			(set! server-going #t)
			(tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "running")
			(server:write-dotserver *toppath* (conc iface ":" port))
                        (server:dotserver-starting-remove))
                      (begin ;; gotta exit nicely
			(tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "collision")
			(http-transport:server-shutdown server-id port))))))

      ;; 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)))

    (exit)))

;; all routes though here end in exit ...
;;
;; start_server? 
;;
(define (http-transport:launch run-id)

  (server:dotserver-starting)


  (let* ((tdbdat (tasks:open-db)))
    (set! *run-id*   run-id)
    (if (args:get-arg "-daemonize")
	(begin
	  (daemon:ize)
	  (if *alt-log-file* ;; we should re-connect to this port, I think daemon:ize disrupts it
	      (begin

		(thread-sleep! 2)
		(loop (tasks:server-lock-slot (db:delay-if-busy tdbdat) run-id 'http)
		      (- remtries 1)))
	      (begin
		;; since we didn't get the server lock we are going to clean up and bail out
		(debug:print-info 2 *default-log-port* "INFO: server pid=" (current-process-id) ", hostname=" (get-host-name) " not starting due to other candidates ahead in start queue")
		(tasks:server-delete-records-for-this-pid (db:delay-if-busy tdbdat) " http-transport:launch")
                
                (server:dotserver-starting-remove)
		))
	  (let* ((th2 (make-thread (lambda ()
				     (debug:print-info 0 *default-log-port* "Server run thread started")
				     (http-transport:run 
				      (if (args:get-arg "-server")
					  (args:get-arg "-server")
					  "-")

(let ((targ (or (args:get-arg "-reqtarg")(args:get-arg "-target"))))
  (if targ (setenv "MT_TARGET" targ)))

;; The watchdog is to keep an eye on things like db sync etc.
;;
(define *watchdog* (make-thread common:watchdog "Watchdog thread"))

;;(thread-start! *watchdog*)

(if (args:get-arg "-log")
    (let ((oup (open-output-file (args:get-arg "-log"))))
      (debug:print-info 0 *default-log-port* "Sending log output to " (args:get-arg "-log"))
      (set! *default-log-port* oup)))

(if (or (args:get-arg "-h")

     ;; if not on homehost ensure we have a connection to a live server
     ;; NOTE: we *have* a homehost record by now
     ((and (not (cdr (remote-hh-dat *runremote*)))        ;; are we on a homehost?
           (not (remote-conndat *runremote*)))            ;; and no connection
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  6  hh-dat: " (remote-hh-dat *runremote*) " conndat: " (remote-conndat *runremote*))
      (mutex-unlock! *rmt-mutex*)
      (tasks:start-and-wait-for-server (tasks:open-db) 0 15)
      (let* ((cinfo (rmt:get-connection-info 0))
            (transport (vector-ref cinfo 6))) ;; TODO: replace with tasks:server-dat-accessor-?? for transport
        (remote-conndat-set! *runremote* cinfo) ;; calls client:setup which calls client:setup-http
        (remote-transport-set! *runremote* transport))
      (rmt:send-receive cmd rid params attemptnum: attemptnum))
     ;; all set up if get this far, dispatch the query
     ((cdr (remote-hh-dat *runremote*)) ;; we are on homehost
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  7")
      (rmt:open-qry-close-locally cmd (if rid rid 0) params))
     ;; not on homehost, do server query

                                  ((commfail)(vector #f "communications fail"))
                                  ((exn)(vector #f "other fail" (print-call-chain)))))
                         ((rpc) (condition-case ;; handling here has caused a lot of problems. However it is needed to deal with attemtped communication to servers that have gone away
                                  (rpc-transport:client-api-send-receive 0 conninfo cmd params)
                                  ((commfail)(vector #f "communications fail"))
                                  ((exn)(vector #f "other fail" (print-call-chain)))))
			 (else
			  (debug:print 0 *default-log-port* "ERROR: transport " (remote-transport *runremote*) " not supported (1)")
			  (exit))))
	     (success  (if (vector? dat) (vector-ref dat 0) #f))
	     (res      (if (vector? dat) (vector-ref dat 1) #f)))
	(if (vector? conninfo)(http-transport:server-dat-update-last-access conninfo)) ;; refresh access time
        (debug:print-info 12 *default-log-port* "rmt:send-receive, case  9. conninfo=" conninfo " dat=" dat)
	(if success
	    (case (remote-transport *runremote*)

         (begin
           (apply (eval procsym) params))))
    res))


;; rpc receiver
(define (rpc-transport:api-exec cmd params)
  (let* ( (resdat  (api:execute-requests *dbstruct-db* (vector cmd params))) ;; #( flag result )
          (flag    (vector-ref resdat 0))
          (res     (vector-ref resdat 1)))

    (mutex-lock! *heartbeat-mutex*)

    (set! *last-db-access* (current-seconds)) ;; bump *last-db-access*; this will renew keep-running thread's lease on life for another (server:get-timeout) seconds
    ;;(BB> "in api-exec; last-db-access updated to "*last-db-access*)

  ;;(tasks:server-set-state! (db:delay-if-busy (tasks:open-db)) server-id "stopped")


  ;; TODO: (low) the following is extraordinaritly slow.  Maybe we don't even need portlogger for rpc anyway??  the exception-based failover when ports are taken is fast!
  ;;(portlogger:open-run-close portlogger:set-port (rpc:default-server-port) "released")
  
  (set! *time-to-exit* #t)
  ;;(if *dbstruct-db* (db:sync-touched *dbstruct-db* *run-id* force-sync: #t))
  
  (server:remove-dotserver-file *toppath* "anyhost:anyport" force: #t)
  (tasks:server-delete-record (db:delay-if-busy (tasks:open-db)) server-id " rpc-transport:keep-running complete")
  

  ;;(BB> "Before (exit) (from-on-exit="from-on-exit")")
  ;;(unless from-on-exit (exit))  ;; sometimes we hang (around) here with 100% cpu.
  ;;(BB> "After")
  ;; strace reveals endless:

  ;; (when (args:get-arg "-daemonize")
  ;;     (daemon:ize)
  ;;     (when *alt-log-file* ;; we should re-connect to this port, I think daemon:ize disrupts it
  ;;       (current-error-port *alt-log-file*)
  ;;       (current-output-port *alt-log-file*)))

  ;; double check we dont alrady have a running server for this run-id
  (when (and (server:read-dotserver *toppath*)
             (server:check-if-running run-id))
    (debug:print 0 *default-log-port* "INFO: Server for run-id " run-id " already running")
    (exit 0))

  ;; did not find server running, let's clean up the table of dead servers

  (tasks:server-force-clean-running-records-for-run-id  (db:delay-if-busy (tasks:open-db)) run-id "notresponding")  



























  (server:dotserver-starting)

  



  
  ;; let's get a server-id for this server
  ;;   if at first we do not suceed, try 3 more times.
  (let ((server-id (retry-thunk
                    (lambda () (tasks:server-lock-slot (db:delay-if-busy (tasks:open-db)) run-id 'rpc))
                    chatty: #f
                    final-failure-returns-actual: #t
                    retries: 4)))
    (when (not server-id) ;; dang we couldn't get a server-id.
      ;; since we didn't get the server lock we are going to clean up and bail out


      (debug:print-info 2 *default-log-port* "INFO: server pid=" (current-process-id) ", hostname=" (get-host-name) " not starting due to other candidates ahead in start queue")
      (tasks:server-delete-records-for-this-pid (db:delay-if-busy (tasks:open-db)) " rpc-transport:launch")
      (server:dotserver-starting-remove)
      (exit 1))

    ;; we got a server-id (and a corresponding entry in servers table in globally shared mdb)
    ;; all systems go.  Proceed to setup rpc server.  
    (rpc-transport:run
     (if (args:get-arg "-server")
         (args:get-arg "-server")

         ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
         ;; rpc:listener is the tcp-listen result from inside the find-free-port-and-open complex.
         ;;   It is our handle on the listening tcp port
         ;;   We will attach this to our rpc server with rpc:make-server in thread th1 .
	 (rpc:listener    (rpc-transport:find-free-port-and-open start-port)) 
	 (th1             (make-thread
			   (lambda ()
                             ;;(BB> "BEFORE rpc:make-server")
			     ((rpc:make-server rpc:listener) #t)
                             ;;(BB> "BEFORE rpc:make-server")
                             )
			   "rpc:server"))


         (hostname        (if (string=? "-" hostn)
			      (get-host-name) 
			      hostn))
	 (ipaddrstr       (if (string=? "-" hostn)
			      (server:get-best-guess-address hostname) ;; (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
                              (string-intersperse 
                               (map number->string
                                    (u8vector->list
                                      (hostname->ip hostn))) ".")
                              ))
	 (portnum         (let ((res (rpc:default-server-port)))  res))
	 (host:port       (conc (if ipaddrstr ipaddrstr hostname) ":" portnum)))

    (when (not (equal? ipaddrstr (server:get-best-guess-address (get-host-name))))
        
      (debug:print 0 *default-log-port* "Error: This host "(ip->string (hostname->ip (get-host-name)))" ("(get-host-name)") is not the homehost "ipaddrstr" ("(ip->hostname (string->ip ipaddrstr))"; Cannot proceed.")
      (server:dotserver-starting-remove)
      (tcp-close rpc:listener) ;; gotta exit nicely and free up that tcp port
      (exit))
    
    (tasks:server-set-interface-port (db:delay-if-busy (tasks:open-db)) server-id ipaddrstr portnum)

    ;;============================================================
    ;;  activate thread th1 to attach opened tcp port to rpc server
    ;;=============================================================
    (thread-start! th1)
    (set! db *dbstruct-db*)

    (debug:print 0 *default-log-port* "Server started on " host:port)
    ;;(BB> "before SELF-TEST")


    (if (retry-thunk (lambda ()
                       (rpc-transport:self-test run-id ipaddrstr portnum))
                     final-failure-returns-actual: #t ;; TODO: remove this line
                     )
        (debug:print 0 *default-log-port* "INFO: rpc self test passed!")
        (begin
          (debug:print 0 *default-log-port* "Error: rpc listener did not pass self test.  Shutting down.  On: " host:port)
          (tasks:server-set-state! (db:delay-if-busy (tasks:open-db)) server-id "dead")
          (tcp-close rpc:listener) ;; gotta exit nicely and free up that tcp port
          (rpc-transport:server-shutdown server-id rpc:listener)
          (server:dotserver-starting-remove)
          (exit)))




    ;;(on-exit (lambda ()
    ;;           (rpc-transport:server-shutdown server-id rpc:listener from-on-exit: #t)))
    
    ;; check again for running servers for this run-id in case one has snuck in since we checked last in rpc-transport:launch
    (if (not (equal? server-id (tasks:server-am-i-the-server? (db:delay-if-busy (tasks:open-db)) run-id)));; try to ensure no double registering of servers
        (begin ;; i am not the server, another server snuck in and beat this one to the punch
          (tcp-close rpc:listener) ;; gotta exit nicely and free up that tcp port
          (tasks:server-set-state! (db:delay-if-busy (tasks:open-db)) server-id "collision")
          (server:dotserver-starting-remove))

        (begin ;; i am the server
          ;; setup the in-memory db
          (set! *dbstruct-db*  (db:setup run-id))
          (db:get-db *dbstruct-db* run-id)

          ;; at this point, satisfied server has started
          ;; let's make it official
          (server:write-dotserver *toppath* (conc ipaddrstr ":" portnum))
          (mutex-lock! *heartbeat-mutex*)
          (set! *last-db-access* (current-seconds))
          (mutex-unlock! *heartbeat-mutex*)
          (set! *rpc:listener* rpc:listener) 
          (tasks:server-set-state! (db:delay-if-busy (tasks:open-db)) server-id "running") ;; update our mdb servers entry

          
          
          ;; this let loop will hold open this thread until we want the server to shut down.
          ;;   if no requests received within the last 20 seconds :
          ;;   database hasnt changed in ??
          ;;


          
          ;; keep-running loop: polls last-db-access to see if we have timed out.  keep running if not. 
          (let loop ((count          0)
                     (bad-sync-count 0))
            (BB> "keep running: count = "count)
            ;; Use this opportunity to sync the inmemdb to db

            (let ((start-time (current-milliseconds))
                  (sync-time  #f)
                  (rem-time   #f))

              ;; following is now done in common:watchdog, commenting out.  sync-time will now be 0; can live with that.
              ;; ;; inmemddb is a dbstruct
              ;; (condition-case
              ;;  (db:sync-touched *dbstruct-db* *run-id* force-sync: #t)
              ;;  ((sync-failed)(cond
              ;;                 ((> bad-sync-count 10) ;; time to give up
              ;;                  (rpc-transport:server-shutdown server-id rpc:listener))
              ;;                 (else ;; (> bad-sync-count 0)  ;; we've had a fail or two, delay and loop
              ;;                  (thread-sleep! 5)
              ;;                  (loop count (+ bad-sync-count 1)))))
              ;;  ((exn)
              ;;   (debug:print-error 0 *default-log-port* "error from sync code other than 'sync-failed. Attempting to gracefully shutdown the server ")
              ;;   (rpc-transport:server-shutdown server-id rpc:listener)))
              
              (set! sync-time  (- (current-milliseconds) start-time))
              (set! rem-time (quotient (- 4000 sync-time) 1000))
              (debug:print 4 *default-log-port* "SYNC: time= " sync-time ", rem-time=" rem-time)
              
              (if (and (<= rem-time 4)
                       (> rem-time 0))
                  (thread-sleep! rem-time)

                    ;;
                    (if (tasks:server-am-i-the-server? (db:delay-if-busy (tasks:open-db)) run-id)
                        (tasks:server-set-state! (db:delay-if-busy (tasks:open-db)) server-id "running"))
                    ;;
                    (loop 0 bad-sync-count))
                  (begin
                    ;;(BB> "SERVER SHUTDOWN CALLED!  last-access="last-access" current-seconds="(current-seconds)" server-timeout="server-timeout)
                    
                    (rpc-transport:server-shutdown server-id rpc:listener)))))
          ;; end new loop
          ))))


(define (rpc-transport:find-free-port-and-open port #!key )
  (handle-exceptions

;;
(define (server:launch run-id transport-type-raw)
  (let ((transport-type
         (cond
          ((string? transport-type-raw) (string->symbol transport-type-raw))
          (else transport-type-raw))))
         
    ;;(BB> "server:launch fired for run-id="run-id" transport-type="transport-type)

    (case transport-type
      ((http)(http-transport:launch run-id))
      ;;((nmsg)(nmsg-transport:launch run-id))
      ((rpc)  (rpc-transport:launch run-id))
      (else (debug:print-error 0 *default-log-port* "unknown server type " transport-type)))))
  

	      (file-read-access? dotfile))
	 (with-input-from-file
	     dotfile
	   (lambda ()
	     (read-line)))
	 #f))))


(define (server:dotserver-starting)
  (with-output-to-file
      (conc *toppath* "/.starting-server")
    (lambda ()
      (print (current-process-id) " on " (get-host-name)))))

(define (server:dotserver-starting-remove)
  (delete-file* (conc *toppath* "/.starting-server")))
  

  
;; write a .server file in *toppath* with hostport
;; return #t on success, #f otherwise
;;
(define (server:write-dotserver areapath hostport)
  (let ((lock-file   (conc areapath "/.server.lock"))
	(server-file (conc areapath "/.server")))
    (if (common:simple-file-lock lock-file)
	(let ((res (handle-exceptions
		    exn
		    #f ;; failed for some reason, for the moment simply return #f
		    (with-output-to-file server-file
		      (lambda ()
			(print hostport)))
		    #t)))
	  (debug:print-info 0 *default-log-port* "server file " server-file " for " hostport " created")
	  (common:simple-file-release-lock lock-file)
	  res)
	#f)))

(define (server:remove-dotserver-file areapath hostport #!key (force #f))
  (let ((dotserver   (server:read-dotserver areapath))
	(server-file (conc areapath "/.server"))
	(lock-file   (conc areapath "/.server.lock")))
    (if (or force (and dotserver (string-match (conc ".*:" hostport "$") dotserver))) ;; port matches, good enough info to decide to remove the file
	(if (common:simple-file-lock lock-file)
	    (begin
	      (handle-exceptions
	       exn
	       #f
	       (delete-file* server-file))
	      (debug:print-info 0 *default-log-port* "server file " server-file " for " hostport " removed")