Megatest

;; 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 remdat apath)
  (let* ((fullpath (db:dbname->path apath ".db/main.db"))
	 (conns    (servdat-conns remdat))
	 (conn     (rmt:get-conn remdat apath ".db/main.db"))) ;; (hash-table-ref/default conns fullpath #f)) ;; TODO - create call for this

    (cond



     ((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 
     ((and conn
	   (>= (current-seconds)(conndat-expires conn)))
      (debug:print-info 0 *default-log-port* "connection to "fullpath" server expired. Reconnecting.")
      (rmt:drop-conn remdat apath ".db/main.db") ;;

	   dconn
	   (< (current-seconds)(conndat-expires dconn)))
      #t) ;; good to go
     ((not mconn) ;; no channel open to main? open it...
      (rmt:open-main-connection sinfo apath)
      (rmt:general-open-connection sinfo apath dbname num-tries: (- num-tries 1)))
     ((not dconn)                 ;; no channel open to dbname?     
      (let* ((res (rmt:send-receive-real sinfo apath mdbname 'get-server `(,apath ,dbname))))
	(case res
	  ((server-started)
	   (if (> num-tries 0)
	       (begin
		 (thread-sleep! 2)
		 (rmt:general-open-connection sinfo apath dbname num-tries: (- num-tries 1)))
	       (begin

;; Defaults to current area
;;
(define (rmt:send-receive cmd rid params #!key (attemptnum 1)(area-dat #f))
  (let* ((apath      *toppath*)
	 (sinfo      *db-serv-info*)
	 (dbname     (db:run-id->dbname rid)))
    (if *localmode*
	(api:execute-requests *dbstruct* cmd params)
	(begin
	  (rmt:open-main-connection sinfo apath)
	  (if rid (rmt:general-open-connection sinfo apath dbname))
	  #;(if (not (member cmd '(log-to-main)))
	      (debug:print-info 0 *default-log-port* "rmt:send-receive "cmd" params="params))
	  (assert (not (eq? 'primordial (thread-name (current-thread)))) "FATAL: Do not call rmt:send-receive-real in the primodial thread.")
	  (let* ((cdat (rmt:get-conn sinfo apath dbname)))
	    (assert cdat "FATAL: rmt:send-receive-real called without the needed channels opened")
	    (let* ((uconn    (servdat-uconn sinfo)) ;; get the interface to ulex
		   ;; then send-receive using the ulex layer to host-port stored in cdat
		   (res      (send-receive uconn (conndat-hostport cdat) cmd params)))
	      (conndat-expires-set! cdat (+ (current-seconds)
					    (server:expiration-timeout)
					    -2)) ;; two second margin for network time misalignments etc.
	      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 (common:run-sync?)
    ;; (and (common:on-homehost?)
  (args:get-arg "-server"))

(define *rmt:run-mutex* (make-mutex))
(define *rmt:run-flag* #f)





;; Main entry point to start a server. was start-server
(define (rmt:run hostn)
  (mutex-lock! *rmt:run-mutex*)
  (if *rmt:run-flag*
      (begin
	(debug:print-warn 0 *default-log-port* "rmt:run already running.")
	(mutex-unlock! *rmt:run-mutex*))
      (begin
	(set! *rmt:run-flag* #t)
	(mutex-unlock! *rmt:run-mutex*)
	;;  ;; Configurations for server
	;;  (tcp-buffer-size 2048)
	;;  (max-connections 2048) 
	(debug:print 2 *default-log-port* "PID: "(current-process-id)". Attempting to start the server ...")
	(if (and *db-serv-info*
		 (servdat-uconn *db-serv-info*))
	    (let* ((uconn (servdat-uconn *db-serv-info*)))
	      (wait-and-close uconn))
	    (let* ((port            (portlogger:open-run-close portlogger:find-port))
		   (handler-proc    (lambda (rem-host-port qrykey cmd params) ;;
				      (set! *db-last-access* (current-seconds))
				      (assert (list? params) "FATAL: handler called with non-list params")
				      (assert (args:get-arg "-server") "FATAL: handler called on non-server side. cmd="cmd", params="params)

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

(define (rmt:register-server sinfo apath iface port server-key dbname)
  (servdat-conns sinfo) ;; just checking types
  (rmt:open-main-connection sinfo apath) ;; we need a channel to main.db
  (rmt:send-receive-real sinfo 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 sinfo apath)
  (servdat-conns sinfo) ;; just checking types
  (rmt:open-main-connection sinfo apath) ;; we need a channel to main.db
  (rmt:send-receive-real sinfo apath      ;; params: host port servkey pid ipaddr dbpath
			 (db:run-id->dbname #f)
			 'get-count-servers `(,apath)))

(define (rmt:get-servers-info apath)
  (rmt:send-receive 'get-servers-info #f `(,apath)))

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

					  (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 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)
	      ))
	
	(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

  (let* ((th2 (make-thread (lambda ()
			     (debug:print-info 0 *default-log-port* "Server run thread started")
			     (rmt:run (rmt:get-reasonable-hostname)))
			   "Server run"))
	 (th3 (make-thread (lambda ()
			     (debug:print-info 0 *default-log-port* "Server monitor thread started")
			     (if (args:get-arg "-server")
				 (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))

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

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

;;======================================================================
;; 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))

#;(define (open-nn-connection host-port)
  (let ((req  (make-req-socket))
        (uri  (conc "tcp://" host-port)))
    (nng-dial req uri)
    (socket-set! req 'nng/recvtimeo 2000)
    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")
;;                                                  #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)) 
    (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)
     (nng-send req msg)
     (let* ((th1  (make-thread (lambda ()
                                 (let ((resp (nng-recv req)))
                                   (nng-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
;;
#;(define (server:ping-server ifaceport)
  (with-input-from-pipe 
   (conc (common:get-megatest-exe) " -ping " ifaceport)
   (lambda ()
     (let loop ((inl (read-line))
		(res "NOREPLY"))
       (if (eof-object? inl)
	   (case (string->symbol res)
	     ((NOREPLY)  #f)
	     ((LOGIN_OK) #t)
	     (else       #f))
	   (loop (read-line) inl))))))

;; NOT USED (well, ok, reference in rpc-transport but otherwise not used).
;;
#;(define (server:login toppath)
  (lambda (toppath)
    (set! *db-last-access* (current-seconds)) ;; might not be needed.
    (if (equal? *toppath* toppath)
	#t
	#f)))

;; (define server:sync-lock-token "SERVER_SYNC_LOCK")
;; (define (server:release-sync-lock)
;;   (db:no-sync-del! *no-sync-db* server:sync-lock-token))
;; (define (server:have-sync-lock?)
;;   (let* ((have-lock-pair (db:no-sync-get-lock *no-sync-db* server:sync-lock-token))
;;          (have-lock?     (car have-lock-pair))
;;          (lock-time      (cdr have-lock-pair))
;;          (lock-age       (- (current-seconds) lock-time)))
;;     (cond
;;      (have-lock? #t)
;;      ((>lock-age
;;        (* 3 (configf:lookup-number *configdat* "server" "minimum-intersync-delay" default: 180)))
;;       (server:release-sync-lock)
;;       (server:have-sync-lock?))
;;      (else #f))))

)