Megatest

  (let* ((stepname       (car ezstep))  ;; do stuff to run the step
	 (stepinfo       (cadr ezstep))
	;; (let ((info (cadr ezstep)))
	;; 		   (if (proc? info) "" info)))
	;; (stepproc       (let ((info (cadr ezstep)))
	;; 		   (if (proc? info) info #f)))
	 (stepparts      (string-match (regexp "^(\\{([^\\}\\{]*)\\}\\s*|)(.*)$") stepinfo))
	 (stepparams     (if (and (list? stepparts)
				  (> (length stepparts) 1))
			     (list-ref stepparts 2)
			     #f)) ;; for future use, {VAR=1,2,3}, run step for each
	 (paramparts     (if (string? stepparams)
			     (map (lambda (x)(string-split x "=")) (string-split-fields "[^;]*=[^;]*" stepparams))
			     '()))
	 (subrun         (alist-ref "subrun" paramparts equal?))
	 (stepcmd        (if (and (list? stepparts)
				  (> (length stepparts) 2))
			     (list-ref stepparts 3)
			     (conc "# error, no command for step "stepname)))
	 (script         "") ; "#!/bin/bash\n") ;; yep, we depend on bin/bash FIXME!!!\
	 (logpro-file    (conc stepname ".logpro"))
	 (html-file      (conc stepname ".html"))
	 (dat-file       (conc stepname ".dat"))
	 (tconfig-logpro (configf:lookup testconfig "logpro" stepname))
	 (logpro-used    (common:file-exists? logpro-file)))

	    (set-signal-handler! signal/int sighand)
	    (set-signal-handler! signal/term sighand)
	    ) ;; (set-signal-handler! signal/stop sighand)
	  
	  ;; Do not run the test if it is REMOVING, RUNNING, KILLREQ or REMOTEHOSTSTART,
	  ;; Mark the test as REMOTEHOSTSTART *IMMEDIATELY*
	  ;;
	  (let* ((test-info (let loop ((tries 0))
			      (let ((tinfo (rmt:get-test-info-by-id run-id test-id)))
				(if tinfo
				    tinfo
				    (if (> tries 5)
					#f
					(begin
					  (thread-sleep! (+ 1 (* tries 10)))
					  (loop (+ tries 1))))))))
		 (test-host (if test-info
				(db:test-get-host        test-info)
				(begin
				  (debug:print 0 *default-log-port* "ERROR: failed to find a record for test-id " test-id ", exiting.")
				  (exit))))
		 (test-pid  (db:test-get-process_id  test-info)))
	    (cond

	  ;; We are about to actually kick off the test
	  ;; so this is a good place to remove the records for 
	  ;; any previous runs
	  ;; (db:test-remove-steps db run-id testname itemdat)
	  ;; now is also a good time to write the .testconfig file
	  (let* ((tconfig-fname   (conc work-area "/.testconfig"))
		 (tconfig-tmpfile (conc tconfig-fname ".tmp"))
		 (tconfig         (tests:get-testconfig test-name item-path tconfigreg #t force-create: #t)) ;; 'return-procs)))
		 (scripts (configf:get-section tconfig "scripts")))
	    ;; create .testconfig file
	    (configf:write-alist tconfig tconfig-tmpfile)
	    (file-move tconfig-tmpfile tconfig-fname #t)
	    (delete-file* ".final-status")

	    ;; extract scripts from testconfig and write them to files in test run dir
	    (for-each
	     (lambda (scriptdat)
	       (match scriptdat
		      ((name content)
		       (with-output-to-file name
			 (lambda ()
			   (print content)
			   (change-file-mode name (bitwise-ior perm/irwxg perm/irwxu)))))
		      (else
		       (debug:print-info 0 "Invalid script definiton found in [scripts] section of testconfig. \"" scriptdat "\""))))
	     scripts))
	  ;; 
	  (let* ((m            (make-mutex))
		 (kill-job?    #f)
		 (exit-info    (make-launch:einf pid: #t exit-status: #t exit-code: #t rollup-status: 0)) ;; pid exit-status exit-code (i.e. process was successfully run) rollup-status
		 (job-thread   #f)
		 ;; (keep-going   #t)
		 (misc-flags   (let ((ht (make-hash-table)))

				      (if (equal? (db:test-get-status testinfo) "AUTO") "AUTO-WARN" "WARN"))
				     ((eq? (launch:einf-rollup-status exit-info) 3) "CHECK")
				     ((eq? (launch:einf-rollup-status exit-info) 4) "WAIVED")
				     ((eq? (launch:einf-rollup-status exit-info) 5) "ABORT")
				     ((eq? (launch:einf-rollup-status exit-info) 6) "SKIP")
				     (else "FAIL")))) ;; (db:test-get-status testinfo)))
		    (debug:print-info 1 *default-log-port* "Test exited in state=" (db:test-get-state testinfo) ", setting state/status based on exit code of " (launch:einf-exit-status exit-info) " and rollup-status of " (launch:einf-rollup-status exit-info))
   
        ;; Leave a .final-status file for each sub-test
        (tests:save-final-status run-id test-id)

		    (tests:test-set-status! run-id 
					    test-id 
					    new-state
					    new-status
					    (args:get-arg "-m") #f)
		    ;; need to update the top test record if PASS or FAIL and this is a subtest
		    ;; NO NEED TO CALL set-state-status-and-roll-up-items HERE, THIS IS DONE IN set-state-status-and-roll-up-items called by tests:test-set-status!
		    ))
	      ;; for automated creation of the rollup html file this is a good place...
	      (if (not (equal? item-path ""))
		      (tests:summarize-items run-id test-id test-name #f))
	      (tests:summarize-test run-id test-id)  ;; don't force - just update if no
        ;; Leave a .final-status file for the top level test
        (tests:save-final-status run-id test-id)
	      (rmt:update-run-stats run-id (rmt:get-raw-run-stats run-id)))
	    (mutex-unlock! m)
            (launch:end-of-run-check run-id )
	    (debug:print 2 *default-log-port* "Output from running " fullrunscript ", pid " (launch:einf-pid exit-info) " in work area " 
			 work-area ":\n====\n exit code " (launch:einf-exit-code exit-info) "\n" "====\n")
	    (if (not (launch:einf-exit-status exit-info))
		(exit 4))))

(define (launch:end-of-run-check run-id )
    (let*	((not-completed-cnt (rmt:get-not-completed-cnt run-id))  
           (running-cnt (rmt:get-count-tests-running-for-run-id run-id))
           (all-test-launched (rmt:get-var (conc "lunch-complete-" run-id)))
           (current-state (rmt:get-run-state run-id))
           (current-status (rmt:get-run-status run-id)))
     ;;get-vars run-id to query metadata table to check if all completed. if all-test-launched = yes then only not-completed-cnt = 0 means everyting is completed if no entry found in the table do nothing 
     (debug:print 0 *default-log-port* "Running test cnt :" running-cnt)                      
     (rmt:set-state-status-and-roll-up-run  run-id current-state current-status)
     (runs:update-junit-test-reporter-xml run-id) 
     (cond 
       ((and all-test-launched (eq? not-completed-cnt 0) (equal? all-test-launched "yes" ))
                (if (and (equal? (rmt:get-var (conc "end-of-run-" run-id)) "no") (common:simple-lock (conc "endOfRun" run-id)))
                (begin
           	(debug:print 4 *default-log-port* "look for  post hook. currseconds: " (current-seconds) " EOR " (rmt:get-var (conc "end-of-run-" run-id)))
                (debug:print 0 *default-log-port* "End of Run Detected.")
                (rmt:set-var (conc "end-of-run-" run-id) "yes")
                ;(thread-sleep! 10)
          	(runs:run-post-hook run-id)
                (debug:print 4 *default-log-port* "currseconds: " (current-seconds)" eor: " (rmt:get-var (conc "end-of-run-" run-id)))
                (common:simple-unlock (conc "endOfRun" run-id)))
                 (debug:print 0 *default-log-port* "End of Run Detected but not running post hook. This should happen when eor is set to yes. This will happen only when 2 tests exit at smae time. eor= " (rmt:get-var (conc "end-of-run-" run-id)))))
        ((> running-cnt 3) 
        	  (debug:print 0 *default-log-port* "There are " running-cnt " tests running." ))
        ((> running-cnt 0)
            (debug:print 0 *default-log-port* "running cnt > 0 but <= 3 kill-running-tests-if-dead" )
   				  (let ((kill-cnt (launch:kill-tests-if-dead run-id)))
           			(if (and all-test-launched  (equal? all-test-launched "yes") (eq? kill-cnt running-cnt))
           					(launch:end-of-run-check run-id)))) ;;todo
        (else  (debug:print 0 *default-log-port* "Should it get here?? May be everything is not launched yet. Running test cnt:" running-cnt " Not completed test cnt:" not-completed-cnt)
         (let* ((not-completed-tests (rmt:get-tests-for-run run-id "%" `("NOT_STARTED" "RUNNING" "LAUNCHED" "REMOTEHOSTSTART") `() #f #f #f #f #f #f #f #f)))
       (if (> (length not-completed-tests) 0) 
           (let loop ((running-test (car not-completed-tests))
			     (tal    (cdr not-completed-tests)))
		       (let* ((test-name (vector-ref running-test 2))
                 (item-path (vector-ref running-test 11)))
			       	(debug:print 0 *default-log-port* "test " test-name "/" item-path " not completed")
              (if (not (null? tal))
				  (loop (car tal) (cdr tal)))))))))))        
 
(define (launch:is-test-alive host pid)
  (if (and host pid (not (equal? host "n/a")))
      (let* ((cmd (conc "ssh " host " pstree -A " pid))
	     (output (with-input-from-pipe cmd read-lines)))
	(debug:print 2 *default-log-port* "Running " cmd " received " output)
	(if (eq? (length output) 0)
	   #f
	   #t))
      #t))
 
(define (launch:kill-tests-if-dead run-id)
  (let* ((running-tests (rmt:get-tests-for-run run-id "%" `("RUNNING" "LAUNCHED" "REMOTEHOSTSTART") `() #f #f #f #f #f #f #f #f)))
       (let loop ((running-test (car running-tests))
			     (tal    (cdr running-tests))
			     (kill-cnt 0))
		       (let* ((test-name (vector-ref running-test 2))

(define (launch:setup-body #!key (force-reread #f) (areapath #f))
  (if (and (eq? *configstatus* 'fulldata)
	   *toppath*
	   (not force-reread)) ;; no need to reprocess
      *toppath*   ;; return toppath
      (let* ((use-cache (common:use-cache?)) ;; BB- use-cache checks *configdat* for use-cache setting.  We do not have *configdat*.  Bootstrapping problem here.
	     (toppath  (common:get-toppath areapath))
	     (target   (common:args-get-target))
	     (sections (if target (list "default" target) #f)) ;; for runconfigs
	     (mtconfig (or (args:get-arg "-config") "megatest.config")) ;; allow overriding megatest.config 
             (cachefiles (launch:get-cache-file-paths areapath toppath target mtconfig))
	     ;; checking for null cachefiles should not be necessary, I was seeing error car of '(), might be a chicken bug or a red herring ...
	     (mtcachef   (if (null? cachefiles)
			     #f

		  (set! toppath      *toppath*)
		  (if (not *toppath*)
		      (begin
			(debug:print-error 0 *default-log-port* "you are not in a megatest area!")
			(exit 1)))
		  (setenv "MT_RUN_AREA_HOME" *toppath*)
		  ;; the seed read is done, now read runconfigs, cache it then read megatest.config one more time and cache it
		  (let* ((keys         (common:list-or-null (rmt:get-keys)
							    message: "Failed to retrieve keys in launch.scm. Please report this to the developers."))
			 (key-vals     (keys:target->keyval keys target))
			 (linktree     (common:get-linktree)) ;; (or (getenv "MT_LINKTREE")(if *configdat* (configf:lookup *configdat* "setup" "linktree") #f)))
					;     (if *configdat*
					; 	   (configf:lookup *configdat* "setup" "linktree")
					; 	   (conc *toppath* "/lt"))))
			 (second-pass  (find-and-read-config
					mtconfig

	;; if have -append-config then read and append here
	(let ((cfname (args:get-arg "-append-config")))
	  (if (and cfname
		   (file-read-access? cfname))
	      (read-config cfname *configdat* #t))) ;; values are added to the hash, no need to do anything special.
	*toppath*)))


(define (get-best-disk confdat testconfig)
  (let* ((disks   (or (and testconfig (hash-table-ref/default testconfig "disks" #f))
		      (hash-table-ref/default confdat "disks" #f)))
	 (minspace (let ((m (configf:lookup confdat "setup" "minspace")))
		     (string->number (or m "10000")))))
    (if disks 
	(let ((res (common:get-disk-with-most-free-space disks minspace))) ;; min size of 1000, seems tad dumb
	  (if res
	      (cdr res)
	      (begin ;; DEAD CODE PATH - REVISIT!
;;		(if (common:low-noise-print 20 "No valid disks or no disk with enough space")
;;		    (debug:print-error 0 *default-log-port* "No valid disks found in megatest.config. Please add some to your [disks] section and ensure the directory exists and has enough space!\n    You can change minspace in the [setup] section of megatest.config. Current setting is: " minspace))
		;;(exit 1)
                 (if (null? disks)
                     (cons 1 (conc *toppath* "/runs"))
                     (let ((paths (sort disks (lambda (x y) (> (string-length (cadr x)) (string-length (cadr y)))))))
                       (let loop ((head (car paths)) (tail (cdr paths)))
                         (let ((result (handle-exceptions exn #f (create-directory (cadr head) #t))))
                           (if result
                               result
                               (if (null? tail)
                                   (cons 1 (conc *toppath* "/runs"))
                                   (loop (car tail) (cdr tail)))))))))))
	;; no disks definition - use mtrah/runs, fall back to currdir/runs
	(let* ((toppath (or *toppath*
			    (common:get-toppath *toppath*)
			    (begin
			      (debug:print-error 0 *default-log-port* "Creating runs dir in current directory, this is probably not what you wanted. Please check your setup.")
			      (current-directory))))
	       (runsdir (conc toppath "/runs")))
	  (if (not (file-exists? runsdir))(create-directory runsdir))
	  runsdir)
	))) ;; the code creates the necessary directories if it does not exist and returns the path.

(define (launch:test-copy test-src-path test-path)
  (let* ((ovrcmd (let ((cmd (configf:lookup *configdat* "setup" "testcopycmd")))
		   (if cmd
		       ;; substitute the TEST_SRC_PATH and TEST_TARG_PATH
		       (string-substitute "TEST_TARG_PATH" test-path
					  (string-substitute "TEST_SRC_PATH" test-src-path cmd #t) #t)

    ;; level
    (if (not not-iterated) ;; i.e. iterated
	(let ((iterated-parent  (pathname-directory (conc lnkpath "/" item-path))))
	  (debug:print-info 2 *default-log-port* "Creating iterated parent " iterated-parent)
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print-error 0 *default-log-port* " Failed to create directory " iterated-parent ((condition-property-accessor 'exn 'message) exn) ", continuing but link tree may be corrupted")
	     #;(exit 1))
	   (create-directory iterated-parent #t))))

    (if (symbolic-link? lnkpath) 
	(handle-exceptions
	 exn
	 (begin
	   (debug:print-error 0 *default-log-port* " Failed to remove symlink " lnkpath ((condition-property-accessor 'exn 'message) exn) ", continuing but link tree may be corrupted.")
	   #;(exit 1))
	 (delete-file lnkpath)))

    (if (not (or (common:file-exists? lnkpath)
		 (symbolic-link? lnkpath)))
	(handle-exceptions
	 exn
	 (begin
	   (debug:print-error 0 *default-log-port* " Failed to create symlink " lnkpath ((condition-property-accessor 'exn 'message) exn) ", continuing but link tree may be corrupted.")
	   #;(exit 1))
	 (create-symbolic-link toptest-path lnkpath)))
    
    ;; NB - This was not working right - some top tests are not getting the path set!!!
    ;;
    ;; Do the setting of this record after the paths are created so that the shortdir can 
    ;; be set to the real directory location. This is safer for future clean up if the link
    ;; tree is damaged or lost.