Megatest

(define (configf:assoc-safe-add alist key val #!key (metadata #f))
  (let ((newalist (filter (lambda (x)(not (equal? key (car x)))) alist)))
    (append newalist (list (if metadata
			       (list key val metadata)
			       (list key val))))))

;; this is used in megatestqa/ext.scm.
;; remove it from here and there by 12/31/21
(define config:assoc-safe-add configf:assoc-safe-add)

(define (configf:section-var-set! cfgdat section-name var value #!key (metadata #f))
  (hash-table-set! cfgdat section-name
		   (configf:assoc-safe-add
		    (hash-table-ref/default cfgdat section-name '())
		    var value metadata: metadata)))

(define (configf:eval-string-in-environment str)

		       (lambda ()
			 (set! result ((eval (read)) ht))))
		     (set! result (conc "#{(" cmdtype ") "  cmd "}"))))
		(case cmdsym
		  ((system shell scheme)
		   (let ((delta (- (current-seconds) start-time)))
		     (if (> delta 2)
			 (debug:print-info 2 *default-log-port* "for line \"" l "\"\n command:  " cmd " took " delta " seconds to run with output:\n   " result)
			 (debug:print-info 9 *default-log-port* "for line \"" l "\"\n command:  " cmd " took " delta " seconds to run with output:\n   " result)))))
		(loop (conc prestr result poststr)))
	      res))
	res)))

;; Run a shell command and return the output as a string
(define (shell cmd)

;; runs:run-tests is called from megatest.scm and itself
;;======================================================================
;;
;;  test-names: Comma separated patterns same as test-patts but used in selection 
;;              of tests to run. The item portions are not respected.
;;              FIXME: error out if /patt specified
;;            
;; run-count is passed from megatest.scm as configf:lookup *configdat* "setup" "reruns", or defaults to 1.
(define (runs:run-tests target runname test-patts user flags #!key (run-count 1)) ;; test-names
  (let* ((keys               (keys:config-get-fields *configdat*))
	 (keyvals            (keys:target->keyval keys target))
	 (run-id             (rmt:register-run keyvals runname "new" "n/a" user (args:get-arg "-contour")))  ;;  test-name)))
	 ;; (deferred          '()) ;; delay running these since they have a waiton clause
	 (runconfigf         (conc  *toppath* "/runconfigs.config"))
         (dbfile             (conc  *toppath* "/megatest.db"))

    ;; (set! required-tests     (lset-intersection equal? test-names all-test-names))
    
    ;; look up all tests matching the comma separated list of globs in
    ;; test-patts (using % as wildcard)

    ;; (set! test-names (delete-duplicates (tests:get-valid-tests *toppath* test-patts)))
    (debug:print-info 0 *default-log-port* "tests search path: " (string-intersperse (tests:get-tests-search-path *configdat*) " "))
    (debug:print-info 2 *default-log-port* "all tests:         " (string-intersperse (sort all-test-names string<) " "))
    (debug:print-info 0 *default-log-port* "test names:        " (string-intersperse (sort test-names string<) " "))
    (debug:print-info 0 *default-log-port* "required tests:    " (string-intersperse (sort required-tests string<) " "))

    ;; on the first pass or call to run-tests set FAILS to NOT_STARTED if
    ;; -keepgoing is specified
    (if (eq? *passnum* 0)
	(begin

		    ;; seek and kill in flight -runtests with % as testpatt here
		    ;; (if (equal? testpatt "%")
		    (tasks:kill-runner target run-name testpatt)
		    ;; (debug:print 0 *default-log-port* "not attempting to kill any run launcher processes as testpatt is " testpatt))
		    (debug:print 1 *default-log-port* "Removing tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((set-state-status)
		    ;; (if (tasks:need-server run-id)(tasks:start-and-wait-for-server tdbdat run-id 10))
		    (debug:print 1 *default-log-port* "Modifying state and status for tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((print-run)
		    (debug:print 1 *default-log-port* "Printing info for run " runkey ", run=" run ", tests=" tests ", header=" header)
		    action)
		   ((run-wait)
		    (debug:print 1 *default-log-port* "Waiting for run " runkey ", run=" runnamepatt " to complete"))
		   ((archive)
		    (debug:print 1 *default-log-port* "Archiving/restoring (" (args:get-arg "-archive") ") data for run: " runkey " " (db:get-value-by-header run header "runname"))

    ((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))

	     (not (equal? curr-ip target-host)))
	(begin
	  (debug:print-info 0 *default-log-port* "Starting server on " target-host ", logfile is " logfile)
	  (setenv "TARGETHOST" target-host)))
      
    (setenv "TARGETHOST_LOGF" logfile)
    (thread-sleep! (/ (random 5000) 1000)) ;; add about a random (up to 5 seconds) initial delay. It seems pretty common that many running tests request a server at the same time
    (debug:print 0 *default-log-port* "INFO: starting server at " (common:human-time))

    (system (conc "nbfake " cmdln))
    (unsetenv "TARGETHOST_LOGF")
    (if (get-environment-variable "TARGETHOST")(unsetenv "TARGETHOST"))
    (thread-join! log-rotate)
    (pop-directory)))

;; given a path to a server log return: host port startseconds server-id
;; any changes to number of elements returned by this fuction will dirctly affect server:record->url,server:record->id,server:kill,server:get-num-alive which use match let 
;; example of what it's looking for in the log file:
;;     SERVER STARTED: 10.38.175.67:50216 AT 1616502350.0 server-id: 4907e90fc55c7a09694e3f658c639cf4 

(define (server:logf-get-start-info logf)
  (let ((server-rx (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+) server-id: (\\S+)")) ;; SERVER STARTED: host:port AT timesecs server id
        (dbprep-rx (regexp "^SERVER: dbprep"))
        (dbprep-found 0)) 
    (handle-exceptions
	exn

			(list (car dat) ;; host
			      (string->number (cadr dat)) ;; port
			      (string->number (caddr dat))
                              (cadr (cddr dat))))))
                (begin 
                   (if dbprep-found
                      (begin
                         (debug:print-info 0 *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)))
                   )
		    (list #f #f #f #f)))))))))

;; 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)))
    ;; if the directory exists continue to get the list
    ;; otherwise attempt to create the logs dir and then

        ;; Get the list of server logs that do not contain "exiting". Ignore logs for servers that have exited.
	(let* ((server-logs-cmd  (conc "grep -iL exiting " areapath "/logs/server-*-*.log"))
               (server-logs   (string-split (string-chomp (call-with-input-pipe server-logs-cmd read-all))))
	       (num-serv-logs (length server-logs)))
	  (if (or (null? server-logs) (= num-serv-logs 0))
              (let ()
                 (debug:print 1  *default-log-port* "There are no servers running at " (common:human-time))
	         '()
              )
	      (let loop ((hed  (string-chomp (car server-logs)))
			 (tal  (cdr server-logs))
			 (res '()))
		(let* ((mod-time  (handle-exceptions
				   exn

(define (server:get-client-signature) ;; BB> why is this proc named "get-"?  it returns nothing -- set! has not return value.
  (if *my-client-signature* *my-client-signature*
      (let ((sig (server:mk-signature)))
        (set! *my-client-signature* sig)
        *my-client-signature*)))


;; if server-start-last exists, and wasn't old enough, wait <idle time>, then call this function recursively until it is old enough.
;; if it is old enough, overwrite it and wait 0.25 seconds.
;; if it then has the wrong server key, wait <idle time> and call this function recursively.
;;
(define (server:wait-for-server-start-last-flag areapath)
  (let* ((start-flag (conc areapath "/logs/server-start-last"))
	 ;;; THIS INTERACTS WITH [server] timeout. Suggest using 0.1 or above for timeout (6 seconds)
	 (idletime    (configf:lookup-number *configdat* "server" "idletime" default: 4))
	 (server-key (conc (get-host-name) "-" (current-process-id))))
    (if (file-exists? start-flag)
	(let* ((fmodtime (file-modification-time start-flag))
	       (delta    (- (current-seconds) fmodtime))
	       (old-enough   (> delta idletime))
              )

          ;; write start-flag file, wait 0.25s, then if previously the start-flag file was older than <idletime> seconds, and the new file still has the same server key as you just wrote, return #t.
          ;; 
	  (if (and old-enough
		   (begin
                     (debug:print-info 0 *default-log-port* "Writing " start-flag)
		     (with-output-to-file start-flag (lambda () (print server-key)))


		     (thread-sleep! 0.25)
		     (let ((res (with-input-from-file start-flag (lambda () (read-line)))))


		       (equal? server-key res)))
                )
	      #t

           ;; If either of the above conditions is not true, print a "Gating server start" message, wait <idle-time>, then call this function recursively. 
	      (begin
		(debug:print-info 0 *default-log-port* "Gating server start, last start: "
				  (seconds->time-string fmodtime) ", time since last start: " delta ", required idletime: " idletime ", gating reason:" (if old-enough "server key does not match" "too soon to start another server"))

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


        
;; 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)
  (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  0 *default-log-port* "server:kind-run: touching " start-flag)
	  (system (conc "touch " start-flag)) ;; lazy but safe
	  (server:run areapath)
	  (thread-sleep! 18) ;; don't release the lock for at least a few seconds. And allow time for the server startup to get to "SERVER STARTED".
	  (common:simple-file-release-lock lock-file)))

      (debug:print-info 0 *default-log-port* "Found server already running. NOT trying to start another.")
   )
)

;; this one seems to be the general entry point
;;
(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))