Megatest

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

;; Call this to start the actual server
;;

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

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

;; Get the transport
(define (server:get-transport)
  (if *transport-type*

(define (server:get-server-id)
  (if *server-id* *server-id*
      (let ((sig (server:mk-signature))) ;; clients re-use the server:mk-signature logic
        (set! *server-id* sig)
        *server-id*)))

;; When using zmq this would send the message back (two step process)
;; with spiffy or rpc this simply returns the return data to be returned
;; 
(define (server:reply return-addr query-sig success/fail result)
  (debug:print-info 11 *default-log-port* "server:reply return-addr=" return-addr ", result=" result)
  ;; (send-message pubsock target send-more: #t)
  ;; (send-message pubsock 
  (case (server:get-transport)
    ((rpc)  (db:obj->string (vector success/fail query-sig result)))
    ((http) (db:obj->string (vector success/fail query-sig result)))
    ((fs)   result)
    (else 
     (debug:print-error 0 *default-log-port* "unrecognised transport type: " *transport-type*)
     result)))
;; ;; When using zmq this would send the message back (two step process)
;; ;; with spiffy or rpc this simply returns the return data to be returned
;; ;; 
;; (define (server:reply return-addr query-sig success/fail result)
;;   (debug:print-info 11 *default-log-port* "server:reply return-addr=" return-addr ", result=" result)
;;   ;; (send-message pubsock target send-more: #t)
;;   ;; (send-message pubsock 
;;   (case (server:get-transport)
;;     ((rpc)  (db:obj->string (vector success/fail query-sig result)))
;;     ((http) (db:obj->string (vector success/fail query-sig result)))
;;     ((fs)   result)
;;     (else 
;;      (debug:print-error 0 *default-log-port* "unrecognised transport type: " *transport-type*)
;;      result)))

;; Given an area path,  start a server process    ### NOTE ### > file 2>&1 
;; if the target-host is set 
;; try running on that host
;;   incidental: rotate logs in logs/ dir.
;;
(define  (server:run areapath) ;; areapath is *toppath* for a given testsuite area
  (let* (;; (curr-host   (get-host-name))
         ;; (attempt-in-progress (server:start-attempted? areapath))
         ;; (dot-server-url (server:check-if-running areapath))
	 ;; (curr-ip     (server:get-best-guess-address curr-host))
	 ;; (curr-pid    (current-process-id))
	 ;; (homehost    (server:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" ))
	 ;; (target-host (car homehost))
	 (testsuite   (common:get-testsuite-name))
  (let* ((testsuite   (common:get-testsuite-name))
	 (logfile     (conc areapath "/logs/server.log")) ;; -" curr-pid "-" target-host ".log"))
	 (profile-mode (or (configf:lookup *configdat* "misc" "profilesw")
			   ""))
	 (cmdln (conc (common:get-megatest-exe)
		      " -server - ";; (or target-host "-")
		      (if (equal? (configf:lookup *configdat* "server" "daemonize") "yes")
			  " -daemonize "

(define (server:logf-get-start-info logf)
  (let ((server-rx    (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+) server-id: (\\S+) pid: (\\d+)")) ;; SERVER STARTED: host:port AT timesecs server id
        (dbprep-rx    (regexp "^SERVER: dbprep"))
        (dbprep-found 0)
	(bad-dat      (list #f #f #f #f #f)))
    (handle-exceptions
	exn
      (begin
	(debug:print-info 0 *default-log-port* "Unable to get server info from " logf ", exn=" exn)
	bad-dat) ;; no idea what went wrong, call it a bad server
      (with-input-from-file
	  logf
	(lambda ()
	  (let loop ((inl  (read-line))
		     (lnum 0))
	    (if (not (eof-object? inl))
		(let ((mlst (string-match server-rx inl))
                      (dbprep (string-match dbprep-rx inl)))
                  (if dbprep (set! dbprep-found 1))
		  (if (not mlst)
		      (if (< lnum 500) ;; give up if more than 500 lines of server log read
			  (loop (read-line)(+ lnum 1))
			  (begin 
     exn
     (begin
       ;; WARNING: this is potentially dangerous to blanket ignore the errors
       (if (file-exists? logf)
	   (debug:print-info 2 *default-log-port* "Unable to get server info from "logf", exn=" exn))
       bad-dat) ;; no idea what went wrong, call it a bad server
     (with-input-from-file
	 logf
       (lambda ()
	 (let loop ((inl  (read-line))
		    (lnum 0))
	   (if (not (eof-object? inl))
	       (let ((mlst (string-match server-rx inl))
		     (dbprep (string-match dbprep-rx inl)))
		 (if dbprep (set! dbprep-found 1))
		 (if (not mlst)
		     (if (< lnum 500) ;; give up if more than 500 lines of server log read
			 (loop (read-line)(+ lnum 1))
			 (begin 
                           (debug:print-info 0 *default-log-port* "Unable to get server info from first 500 lines of " logf )
                           bad-dat))
		      (match mlst
			((_ host port start server-id pid)
			 (list host
			       (string->number port)
			       (string->number start)
			       server-id
			       (string->number pid)))
			(else
			 (debug:print 0 *current-log-port* "ERROR: did not recognise SERVER line info "mlst)
			 bad-dat))))
                (begin 
                  (if dbprep-found
                      (begin
                         (debug:print-info 2 *default-log-port* "Server is in dbprep at " (common:human-time))
                         (thread-sleep! 0.5)) ;; was 25 sec but that blocked things from starting?
                      (debug:print-info 0 *default-log-port* "Unable to get server info from " logf " at " (seconds->time-string (current-seconds))))
		  bad-dat))))))))
		     (match mlst
			    ((_ host port start server-id pid)
			     (list host
				   (string->number port)
				   (string->number start)
				   server-id
				   (string->number pid)))
			    (else
			     (debug:print 0 *current-log-port* "ERROR: did not recognise SERVER line info "mlst)
			     bad-dat))))
	       (begin 
		 (if dbprep-found
		     (begin
		       (debug:print-info 2 *default-log-port* "Server is in dbprep at " (common:human-time))
		       (thread-sleep! 0.5)) ;; was 25 sec but that blocked things from starting?
		     (debug:print-info 0 *default-log-port* "Unable to get server info from " logf " at " (seconds->time-string (current-seconds))))
		 bad-dat))))))))

;; ;; get a list of servers from the log files, with all relevant data
;; ;; ( mod-time host port start-time pid )
;; ;;
;; (define (server:get-list areapath #!key (limit #f))
;;   (let ((fname-rx    (regexp "^(|.*/)server-(\\d+)-(\\S+).log$"))
;; 	(day-seconds (* 24 60 60)))

		 (thread-sleep! ( + 1 idletime))
		 (server:wait-for-server-start-last-flag areapath)))))))

;; oldest server alive determines host then choose random of youngest
;; five servers on that host
;;
(define (server:get-servers-info areapath)
  ;; (assert *toppath* "FATAL: server:get-servers-info called before *toppath* has been set.")
  (let* ((servinfodir (conc *toppath*"/.servinfo")))
  (let* ((servinfodir (server:get-servinfo-dir areapath))) ;; (conc *toppath*"/.servinfo")))
    (if (not (file-exists? servinfodir))
	(create-directory servinfodir))
    (let* ((allfiles    (glob (conc servinfodir"/*")))
	   (res         (make-hash-table)))
      (for-each
       (lambda (f)
	 (let* ((hostport  (pathname-strip-directory f))
		(serverdat (server:logf-get-start-info f)))
	   (match serverdat
	     ((host port start server-id pid)
	      (if (and host port start server-id pid)
		  (hash-table-set! res hostport serverdat)
		  (debug:print-info 0 *default-log-port* "bad server info for "f": "serverdat)))
		  (debug:print-info 2 *default-log-port* "bad server info for "f": "serverdat)))
	     (else
	      (debug:print-info 0 *default-log-port* "bad server info for "f": "serverdat)))))
	      (debug:print-info 2 *default-log-port* "bad server info for "f": "serverdat)))))
       allfiles)
      res)))

;; check the .servinfo directory, are there other servers running on this
;; or another host?
;;
;; returns #t => ok to start another server
;;         #f => not ok to start another server
;;
(define (server:minimal-check areapath)
  (server:clean-up-old areapath)
  (let* ((srvdir      (server:get-servinfo-dir areapath)) ;; (conc areapath"/.servinfo"))
	 (servrs      (glob (conc srvdir"/*")))
	 (thishostip  (server:get-best-guess-address (get-host-name)))
	 (thisservrs  (glob (conc srvdir"/"thishostip":*")))
	 (homehostinf (server:choose-server areapath 'homehost))
	 (havehome    (car homehostinf))
	 (wearehome   (cdr homehostinf)))
    (debug:print-info 0 *default-log-port* thishostip", have homehost: "havehome", we are homehost: "wearehome
		      ", numservers: "(length thisservrs))
    (cond
     ((not havehome) #t) ;; no homehost yet, go for it
     ((and havehome wearehome (< (length thisservrs) 20)) #t) ;; we are home and less than 20 servers, ok to start another
     ((and havehome (not wearehome)) #f)     ;; we are not the home host
     ((and havehome wearehome (>= (length thisservrs) 20)) #f) ;; have enough running
     (else
      (debug:print 0 *default-log-port* "WARNING: Unrecognised scenario, servrs="servrs", thishostip="thishostip", thisservrs="thisservrs)
      #t))))
	 

(define server-last-start 0)


;; oldest server alive determines host then choose random of youngest
;; five servers on that host
;;
;; mode:
;;   best - get best server (random of newest five)
;;   home - get home host based on oldest server
;;   info - print info
(define (server:choose-server areapath #!optional (mode 'best))
  ;; age is current-starttime
  ;; find oldest alive
  ;;   1. sort by age ascending and ping until good
  ;; find alive rand from youngest
  ;;   1. sort by age descending
  ;;   2. take five
  ;;   3. check alive, discard if not and repeat
  ;; first we clean up old server files
  (assert (eq? (rmt:transport-mode) 'http) "FATAL: server:run called with rmt:transport-mode="(rmt:transport-mode))
  (server:clean-up-old areapath)
  (let* ((since-last (- (current-seconds) server-last-start))
        (server-start-delay 10))     
    (if ( < (- (current-seconds) server-last-start) 10 )
      (begin
        (debug:print 2 *default-log-port* "server:choose-server: seconds since last server start: " (- (current-seconds) server-last-start))
        (debug:print 2 *default-log-port* "server:choose-server: last server start less than " server-start-delay " seconds ago. Sleeping " server-start-delay " seconds")
        (thread-sleep! server-start-delay)
      )
      (debug:print 2 *default-log-port* "server:choose-server: seconds since last server start: " (- (current-seconds) server-last-start))
    )
  )
  (let* ((serversdat  (server:get-servers-info areapath))
	 (servkeys    (hash-table-keys serversdat))
	 (by-time-asc (if (not (null? servkeys))
	 (by-time-asc (if (not (null? servkeys)) ;; NOTE: Oldest is last
			  (sort servkeys ;; list of "host:port"
				(lambda (a b)
				  (>= (list-ref (hash-table-ref serversdat a) 2)
				      (list-ref (hash-table-ref serversdat b) 2))))
			  '())))
    (debug:print 2 *default-log-port* "server:choose-server: serversdat: " serversdat)
    (debug:print 2 *default-log-port* "server:choose-server: servkeys: " servkeys)
    (if (not (null? by-time-asc))
	(let* ((oldest     (last by-time-asc))
	       (oldest-dat (hash-table-ref serversdat oldest))
	       (host       (list-ref oldest-dat 0))
	       (all-valid  (filter (lambda (x)
				     (equal? host (list-ref (hash-table-ref serversdat x) 0)))
				   by-time-asc))
	       (best-five  (lambda ()
			     (if (> (length all-valid) 5)
				 (take all-valid 5)
				 all-valid)))
	       (best-ten  (lambda ()
			     (if (> (length all-valid) 11)
				 (take (drop-right all-valid 1) 10) ;; remove the oldest from consideration so it can age out
				 (if (> (length all-valid) 8)
				     (drop-right all-valid 1)
				     all-valid))))
	       (names->dats (lambda (names)
			      (map (lambda (x)
				     (hash-table-ref serversdat x))
				   names)))
	       (am-home?    (lambda ()
			      (let* ((currhost (get-host-name))
				     (bestadrs (server:get-best-guess-address currhost)))
				(or (equal? host currhost)
				    (equal? host bestadrs))))))
	  (case mode
	    ((info)
	     (print "oldest: "oldest-dat", selected host: "host", all-valid: "all-valid)
	     (print "youngest: "(hash-table-ref serversdat (car all-valid))))
	     (debug:print 0 *default-log-port* "oldest: "oldest-dat", selected host: "host", all-valid: "all-valid)
	     (debug:print 0 *default-log-port* "youngest: "(hash-table-ref serversdat (car all-valid))))
	    ((home)     host)
	    ((homehost) (cons host (am-home?))) ;; shut up old code
	    ((home?)    (am-home?))
	    ((best-five)(names->dats (best-five)))
	    ((best-ten)(names->dats (best-ten)))
	    ((all-valid)(names->dats all-valid))
	    ((best)     (let* ((best-five (best-five))
			       (len       (length best-five)))
			  (hash-table-ref serversdat (list-ref best-five (random len)))))
	    ((best)     (let* ((best-ten (best-ten))
			       (len       (length best-ten)))
			  (hash-table-ref serversdat (list-ref best-ten (random len)))))
	    ((count)(length all-valid))
	    (else
	     (debug:print 0 *default-log-port* "ERROR: invalid command "mode)
	     #f)))
	(begin
	  (server:run areapath)
          (set! server-last-start (current-seconds))
	  (thread-sleep! 3)
	  ;; (thread-sleep! 3)
	  (case mode
	    ((homehost) (cons #f #f))
	    (else	#f))))))

(define (server:get-servinfo-dir areapath)
  (let* ((spath (conc areapath"/.servinfo")))
    (if (not (file-exists? spath))
	(create-directory spath #t))
    spath))

(define (server:clean-up-old areapath)
  ;; any server file that has not been touched in ten minutes is effectively dead
  (let* ((sfiles (glob (conc (server:get-servinfo-dir areapath)"/*"))))
    (for-each
     (lambda (sfile)
       (let* ((modtime (handle-exceptions
			   exn
			 (begin
			   (debug:print 0 *default-log-port* "WARNING: failed to get modification file for "sfile)
			   (current-seconds))
			 (file-modification-time sfile))))
	 (if (and (number? modtime)
		  (> (- (current-seconds) modtime)
		     600))
	     (begin
	       (debug:print 0 *default-log-port* "WARNING: found old server info file "sfile", removing it.")
	       (handle-exceptions
		   exn
		 (debug:print 0 *default-log-port* "WARNING: failed to delete old server info file "sfile)
		 (delete-file sfile))))))
     sfiles)))

;; would like to eventually get rid of this
;;
(define (common:on-homehost?)
  (if (eq? (rmt:transport-mode) 'http)
  (server:choose-server *toppath* 'home?))
      (server:choose-server *toppath* 'home?)
      #t)) ;; there is no homehost for tcp and nfs is always on home so #t should work

;; kind start up of server, wait before allowing another server for a given
;; area to be launched
;;
(define (server:kind-run areapath)
  ;; look for $MT_RUN_AREA_HOME/logs/server-start-last
  ;; and wait for it to be at least <server idletime> seconds old
  ;; (server:wait-for-server-start-last-flag areapath)
  (let loop ()
    (if (> (alist-ref 'adj-proc-load (common:get-normalized-cpu-load #f)) 2)
	(begin
	  (if (common:low-noise-print 30 "our-host-load")
	      (debug:print 0 *default-log-port* "WARNING: system load is high, waiting to start server."))
	  (loop))))
  (if (< (server:choose-server areapath 'count) 10)
  (if (< (server:choose-server areapath 'count) 20)
      (server:run areapath))
  #;(if (not (server:check-if-running areapath)) ;; why try if there is already a server running?
      (let* ((lock-file    (conc areapath "/logs/server-start.lock")))
	(let* ((start-flag (conc areapath "/logs/server-start-last")))
	  (common:simple-file-lock-and-wait lock-file expire-time: 25)
	  (debug:print-info  2 *default-log-port* "server:kind-run: touching " start-flag)
	  (system (conc "touch " start-flag)) ;; lazy but safe

(define (server:start-and-wait areapath #!key (timeout 60))
  (let ((give-up-time (+ (current-seconds) timeout)))
    (let loop ((server-info (server:check-if-running areapath))
	       (try-num    0))
      (if (or server-info
	      (> (current-seconds) give-up-time)) ;; server-url will be #f if no server available.
	  (server:record->url server-info)
	  (let* ( (servers (server:choose-server areapath 'all-valid))
	  (let ((num-ok (length (server:choose-server areapath 'all-valid))))
                (num-ok (if servers (length (server:choose-server areapath 'all-valid)) 0)))
	    (if (and (> try-num 0)  ;; first time through simply wait a little while then try again
		     (< num-ok 1))  ;; if there are no decent candidates for servers then try starting a new one
		(server:run areapath))
	    (thread-sleep! 5)
	    (loop (server:check-if-running areapath)
		  (+ try-num 1)))))))

(define (server:get-num-servers #!key (numservers 2))
  (let ((ns (string->number
	     (or (configf:lookup *configdat* "server" "numservers") "notanumber"))))
    (or ns numservers)))

;; no longer care if multiple servers are started by accident. older servers will drop off in time.
;;
(define (server:check-if-running areapath) ;;  #!key (numservers "2"))
  (let* ((ns            (server:get-num-servers)) ;; get the setting the for maximum number of servers allowed
	 (servers       (server:choose-server areapath 'best-five))) ;; (server:get-best (server:get-list areapath))))
	 (servers       (server:choose-server areapath 'best-ten))) ;; (server:get-best (server:get-list areapath))))
    (if (or (and servers
		 (null? servers))
	    (not servers))
	    ;; (and (list? servers)
	    ;;	 (< (length servers) (+ 1 (random ns))))) ;; somewhere between 1 and numservers
        #f
        (let loop ((hed (car servers))

(define (server:kill servr)
  (handle-exceptions
    exn
    (begin 
      (debug:print-info 0 *default-log-port*  "Unable to get host and/or port from " servr ", exn=" exn)     
    #f)
  (match-let (((mod-time hostname port start-time server-id pid)
  (match-let (((hostname port start-time server-id pid)
	       servr))
    (tasks:kill-server hostname pid))))

;; called in megatest.scm, host-port is string hostname:port
;;
;; NOTE: This is NOT called directly from clients as not all transports support a client running
;;       in the same process as the server.
;;
(define (server:ping host-port-in server-id #!key (do-exit #f))
(define (server:ping host:port server-id #!key (do-exit #f))
  (let ((host:port (if (not host-port-in) ;; use read-dotserver to find
		       #f ;; (server:check-if-running *toppath*)
		;; (if (number? host-port-in) ;; we were handed a server-id
		;; 	   (let ((srec (tasks:get-server-by-id (db:delay-if-busy (tasks:open-db)) host-port-in)))
		;; 	     ;; (print "srec: " srec " host-port-in: " host-port-in)
		;; 	     (if srec
		;; 		 (conc (vector-ref srec 3) ":" (vector-ref srec 4))
		;; 		 (conc "no such server-id " host-port-in)))
		       host-port-in))) ;; )
    (let* ((host-port (if host:port
			  (let ((slst (string-split   host:port ":")))
			    (if (eq? (length slst) 2)
				(list (car slst)(string->number (cadr slst)))
				#f))
			  #f)))
  (let* ((host-port (cond
		     ((string? host:port)
		      (let ((slst (string-split   host:port ":")))
			(if (eq? (length slst) 2)
			    (list (car slst)(string->number (cadr slst)))
			    #f)))
		     (else
		      #f))))
;;	   (toppath       (launch:setup)))
      ;; (print "host-port=" host-port)
      (if (not host-port)
    (cond
     ((and (list? host-port)
	  (begin
	    (if host-port-in
	   (eq? (length host-port) 2))
		(debug:print 0 *default-log-port*  "ERROR: bad host:port"))
	    (if do-exit (exit 1))
	    #f)
	  (let* ((iface      (car host-port))
		 (port       (cadr host-port))
		 (server-dat (http-transport:client-connect iface port server-id))
		 (login-res  (rmt:login-no-auto-client-setup server-dat)))
	    (if (and (list? login-res)
		     (car login-res))
		(begin
		  ;; (print "LOGIN_OK")
		  (if do-exit (exit 0))
		  #t)
		(begin
		  ;; (print "LOGIN_FAILED")
		  (if do-exit (exit 1))
		  #f)))))))
      (let* ((myrunremote (make-and-init-remote *toppath*))
	     (iface       (car host-port))
	     (port        (cadr host-port))
	     (server-dat  (client:connect iface port server-id myrunremote))
	     (login-res   (rmt:login-no-auto-client-setup myrunremote)))
	(http-transport:close-connections myrunremote)
	(if (and (list? login-res)
		 (car login-res))
	    (begin
	      ;; (print "LOGIN_OK")
	      (if do-exit (exit 0))
	      #t)
	    (begin
	      ;; (print "LOGIN_FAILED")
	      (if do-exit (exit 1))
	      #f))))
     (else 
      (if host:port
	  (debug:print 0 *default-log-port*  "ERROR: bad host:port "host:port))
      (if do-exit
	  (exit 1)
	  #f)))))

;; 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 ()

;; Default is 60 seconds.
;;
(define (server:expiration-timeout)
  (let ((tmo (configf:lookup *configdat* "server" "timeout")))
    (if (and (string? tmo)
	     (common:hms-string->seconds tmo)) ;; BUG: hms-string->seconds is broken, if given "10" returns 0. Also, it doesn't belong in this logic unless the string->number is changed below
        (* 3600 (string->number tmo))
	60)))
	600)))

(define (server:get-best-guess-address hostname)
  (let ((res #f))
    (for-each 
     (lambda (adr)
       (if (not (eq? (u8vector-ref adr 0) 127))
	   (set! res adr)))