Megatest

;;======================================================================
;; launch a task - this runs on the originating host, tests themselves
;;
;;======================================================================

(use regex regex-case base64 sqlite3 srfi-18 directory-utils posix-extras z3 call-with-environment-variables csv)
(use typed-records pathname-expand matchable)

(import (prefix base64 base64:))
(import (prefix sqlite3 sqlite3:))

(declare (unit launch))
(declare (uses common))
(declare (uses configf))

	       (csvt (let-values (((fmt-cell fmt-record fmt-csv) (make-format ",")))
		       (fmt-csv (map list->csv-record csvr))))
	       (status (configf:lookup dat "final" "exit-status"))
	       (msg     (configf:lookup dat "final" "message")))
          (if csvt  ;; this if blocked stack dump caused by .dat file from logpro being 0-byte.  fixed by upgrading logpro
              (rmt:csv->test-data run-id test-id csvt)
	      (debug:print 0 *default-log-port* "ERROR: no csvdat exists for run-id: " run-id " test-id: " test-id " stepname: " stepname ", check that logpro version is 1.15 or newer"))
	  ;;  (debug:print-info 13 *default-log-port* "Error: run-id/test-id/stepname="run-id"/"test-id"/"stepname" => bad csvr="csvr)
	  ;;  )
	  (cond
	   ((equal? status "PASS") "PASS") ;; skip the message part if status is pass
	   (status (conc (configf:lookup dat "final" "exit-status") ": " (if msg msg "no message")))
	   (else #f)))
	#f)))

	(let* ((testpath  (assoc/default 'testpath  cmdinfo))  ;; testpath is the test spec area
	       (top-path  (assoc/default 'toppath   cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))  ;; work-area is the test run area
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (ezsteps   (assoc/default 'ezsteps   cmdinfo))
	       ;; (runremote (assoc/default 'runremote cmdinfo))
	       ;; (transport (assoc/default 'transport cmdinfo))  ;; not used
	       ;; (serverinf (assoc/default 'serverinf cmdinfo))
	       ;; (port      (assoc/default 'port      cmdinfo))
	       (serverurl (assoc/default 'serverurl cmdinfo))
	       (homehost  (assoc/default 'homehost  cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (test-id   (assoc/default 'test-id   cmdinfo))
	       (target    (assoc/default 'target    cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))
	       (env-ovrd  (assoc/default 'env-ovrd  cmdinfo))
	       (set-vars  (assoc/default 'set-vars  cmdinfo)) ;; pre-overrides from -setvar
	       (runname   (assoc/default 'runname   cmdinfo))

                                                   (file-execute-access? fulln))
                                              fulln
                                              runscript))))) ;; assume it is on the path
	       ) ;; (rollup-status 0)

	  (if contour (setenv "MT_CONTOUR" contour))
	  
	  ;; On NFS it can be slow and unreliable to get needed startup information.
	  ;;  i. Check if we are on the homehost, if so, proceed
	  ;; ii. Check if host and port passed in via CMDINFO are valid and if
	  ;;     possible use them.
	  (let ((bestadrs (server:get-best-guess-address (get-host-name)))
		(needcare #f))
	    (if (equal? homehost bestadrs) ;; we are likely on the homehost
		(debug:print-info 0 *default-log-port* "test " test-name " appears to be running on the homehost " homehost)
		(let ((host-port (if serverurl (string-split serverurl ":") #f)))
		  (if (not *runremote*)(set! *runremote* (make-remote))) ;; init *runremote*
		  (if (string? homehost)
		      (if (and host-port
			       (> (length host-port) 1))
			  (let* ((host      (car host-port))
                                 (port      (cadr host-port))
                                 (start-res (http-transport:client-connect host port))
                                 (ping-res  (rmt:login-no-auto-client-setup start-res)))
			    (if (and start-res
				     ping-res)
				(let ((url  (http-transport:server-dat-make-url start-res)))
				  (remote-conndat-set! *runremote* start-res)
				  (remote-server-url-set! *runremote* url)
				  (debug:print-info 0 *default-log-port* "connected to " url " using CMDINFO data."))
				(debug:print-info 0 *default-log-port* "received " host ":" port " for url but could not connect.")
				))
			  (begin
			    (debug:print-info 0 *default-log-port* (if host-port
								       (conc "received invalid host-port information " host-port)
								       "no host-port information received"))
			    ;; potential for bad situation if simultaneous starting of hundreds of jobs on servers, set needcare.
			    (set! needcare #t)))
		      (begin
			(debug:print-info 0 *default-log-port* "received no homehost information. Please report this to support as it should not happen.")
			(set! needcare #t)))))
	    (if needcare  ;; due to very slow NFS we will do a brute force mkdir to ensure that the directory inode it truly available on this host
		(let ((logdir (conc top-path "/logs"))) ;; we'll try to create this directory
		  (handle-exceptions
		      exn
		      (debug:print 0 *default-log-port* "Failed to create directory " logdir " expect problems, message: " ((condition-property-accessor 'exn 'message) exn))
		    (create-directory logdir #t)))))
		  
	  ;; NFS might not have propagated the directory meta data to the run host - give it time if needed
	  (let loop ((count 0))
	    (if (or (file-exists? top-path)
		    (> count 10))
		(change-directory top-path)
		(begin
		  (debug:print 0 *default-log-port* "INFO: Not starting job yet - directory " top-path " not found")

		(exit 1)))
	  (change-directory *toppath*) 

	  ;; NOTE: Current order is to process runconfigs *before* setting the MT_ vars. This 
	  ;;       seems non-ideal but could well break stuff
	  ;;    BUG? BUG? BUG?
	  
	  (let ((rconfig (full-runconfigs-read)) ;; (read-config (conc  *toppath* "/runconfigs.config") #f #t sections: (list "default" target))))
		(wconfig (read-config "waivers.config" #f #t sections: `( "default" ,target )))) ;; read the waivers config if it exists
	    ;; (setup-env-defaults (conc *toppath* "/runconfigs.config") run-id (make-hash-table) keyvals target)
	    ;; (set-run-config-vars run-id keyvals target) ;; (db:get-target db run-id))
	    ;; Now have runconfigs data loaded, set environment vars
	    (for-each (lambda (section)
			(for-each (lambda (varval)
				    (let ((var (car varval))
					  (val (cadr varval)))

	      (list  "MT_TEST_RUN_DIR" work-area)
	      (list  "MT_TEST_NAME" test-name)
	      (list  "MT_ITEM_INFO" (conc itemdat))
	      (list  "MT_ITEMPATH"  item-path)
	      (list  "MT_RUNNAME"   runname)
	      (list  "MT_MEGATEST"  megatest)
	      (list  "MT_TARGET"    target)
	      (list  "MT_LINKTREE"  (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree"))
	      (list  "MT_TESTSUITENAME" (common:get-testsuite-name))))

	  (if mt-bindir-path (setenv "PATH" (conc (getenv "PATH") ":" mt-bindir-path)))
	  ;; (change-directory top-path)
	  ;; Can setup as client for server mode now
	  ;; (client:setup)

	  

(define (launch:cache-config)
  ;; if we have a linktree and -runtests and -target and the directory exists dump the config
  ;; to megatest-(current-seconds).cfg and symlink it to megatest.cfg
  (if (and *configdat* 
	   (or (args:get-arg "-run")
	       (args:get-arg "-runtests")
	       (args:get-arg "-execute")))
      (let* ((linktree (common:get-linktree)) ;; (get-environment-variable "MT_LINKTREE"))
	     (target   (common:args-get-target exit-if-bad: #t))
	     (runname  (or (args:get-arg "-runname")
			   (args:get-arg ":runname")
			   (getenv "MT_RUNNAME")))
	     (fulldir  (conc linktree "/"
			     target "/"
			     runname)))
	(if (and linktree (file-exists? linktree)) ;; can't proceed without linktree

;;   returns:
;;     *toppath*
;;   side effects:
;;     sets; *configdat*    (megatest.config info)
;;           *runconfigdat* (runconfigs.config info)
;;           *configstatus* (status of the read data)
;;
(define (launch:setup #!key (force #f) (areapath #f))
  (mutex-lock! *launch-setup-mutex*)
  (if (and *toppath*
	   (eq? *configstatus* 'fulldata)) ;; got it all
      (begin
	(debug:print 0 *default-log-port* "NOTE: skipping launch:setup-body call since we have fulldata")
	(mutex-unlock! *launch-setup-mutex*)
	*toppath*)
      (let ((res (launch:setup-body force: force areapath: areapath)))
	(mutex-unlock! *launch-setup-mutex*)
	res)))

(define (launch:setup-body #!key (force #f) (areapath #f))
  (let* ((toppath  (or *toppath* areapath (getenv "MT_RUN_AREA_HOME"))) ;; preserve toppath
	 (runname  (common:args-get-runname))
	 (target   (common:args-get-target exit-if-bad: #t))
	 (linktree (common:get-linktree))
	 (contour  #f) ;; NOT READY FOR THIS (args:get-arg "-contour"))
	 (sections (if target (list "default" target) #f)) ;; for runconfigs
	 (mtconfig (or (args:get-arg "-config") "megatest.config")) ;; allow overriding megatest.config 
	 (rundir   (if (and runname target linktree)(conc linktree (if contour (conc "/" contour) "") "/" target "/" runname) #f))
	 (mtcachef (and rundir (conc rundir "/" ".megatest.cfg-"  megatest-version "-" megatest-fossil-hash)))
	 (rccachef (and rundir (conc rundir "/" ".runconfigs.cfg-"  megatest-version "-" megatest-fossil-hash)))
	 (cancreate (and rundir (file-exists? rundir)(file-write-access? rundir)))
         (cxt       (hash-table-ref/default *contexts* toppath #f)))

		  (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         (rmt:get-keys))
		     (key-vals     (keys:target->keyval keys target))
		     (linktree     (common:get-linktree))
					; (or (getenv "MT_LINKTREE")
					;     (if *configdat*
					; 	   (configf:lookup *configdat* "setup" "linktree")
					; 	   (conc *toppath* "/lt"))))
		     (second-pass  (find-and-read-config
				    mtconfig
				    environ-patt: "env-override"
				    given-toppath: toppath
				    pathenvvar: "MT_RUN_AREA_HOME"))
		     (runconfigdat (begin     ;; this read of the runconfigs will see any adjustments made by re-reading megatest.config
				     (for-each (lambda (kt)

	      (set! *runconfigdat* rdat)
	      (set! *toppath*      toppath)
	      (set! *configstatus* 'partial))
	    (begin
	      (debug:print-error 0 *default-log-port* "No " mtconfig " file found. Giving up.")
	      (exit 2))))))
    ;; additional house keeping
    (let* ((linktree (common:get-linktree)))

      (if linktree
	  (begin
	    (if (not (file-exists? linktree))
		(begin
		  (handle-exceptions
		   exn
		   (begin

	  (begin
	    (debug:print-error 0 *default-log-port* "linktree not defined in [setup] section of megatest.config")
	    )))
    (if (and *toppath*
	     (directory-exists? *toppath*))
	(begin
	  (setenv "MT_RUN_AREA_HOME" *toppath*)
	  (setenv "MT_TESTSUITENAME" (common:get-testsuite-name)))
	(begin
	  (debug:print-error 0 *default-log-port* "failed to find the top path to your Megatest area.")
          ;;(exit 1)
          #f
          ))
    ;; 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")))))

(define (create-work-area run-id run-info keyvals test-id test-src-path disk-path testname itemdat #!key (remtries 2))
  (let* ((item-path (if (string? itemdat) itemdat (item-list->path itemdat))) ;; if pass in string - just use it
	 (runname   (if (string? run-info) ;; if we pass in a string as run-info use it as run-name.
			run-info
			(db:get-value-by-header (db:get-rows run-info)
						(db:get-header run-info)
						"runname")))
	 (contour   #f) ;; NOT READY FOR THIS (args:get-arg "-contour"))
	 ;; convert back to db: from rdb: - this is always run at server end
	 (target   (string-intersperse (map cadr keyvals) "/"))

	 (not-iterated  (equal? "" item-path))

	 ;; all tests are found at <rundir>/test-base or <linkdir>/test-base
	 (testtop-base (conc target "/" runname "/" testname))
	 (test-base    (conc testtop-base (if not-iterated "" "/") item-path))

	 ;; nb// if itempath is not "" then it is prefixed with "/"
	 (toptest-path (conc disk-path (if contour (conc "/" contour) "") "/" testtop-base))
	 (test-path    (conc disk-path (if contour (conc "/" contour) "") "/" test-base))

	 ;; ensure this exists first as links to subtests must be created there
	 (linktree  (common:get-linktree))
	 ;; WAS: (let ((rd (config-lookup *configdat* "setup" "linktree")))
	 ;;         (if rd rd (conc *toppath* "/runs"))))
	 ;; which seems wrong ...

	 (lnkbase   (conc linktree (if contour (conc "/" contour) "") "/" target "/" runname))
	 (lnkpath   (conc lnkbase "/" testname))
	 (lnkpathf  (conc lnkpath (if not-iterated "" "/") item-path))
	 (lnktarget (conc lnkpath "/" item-path)))

    ;; Update the rundir path in the test record for all, rundir=physical, shortdir=logical

;; 4. remotely run the test on allocated host
;;    - could be ssh to host from hosts table (update regularly with load)
;;    - could be netbatch
;;      (launch-test db (cadr status) test-conf))
(define (launch-test test-id run-id run-info keyvals runname test-conf test-name test-path itemdat params)
  (mutex-lock! *launch-setup-mutex*) ;; setting variables and processing the testconfig is NOT thread-safe, reuse the launch-setup mutex
  (let* ((item-path       (item-list->path itemdat))
	 (contour         #f)) ;; NOT READY FOR THIS (args:get-arg "-contour")))
    (let loop ((delta        (- (current-seconds) *last-launch*))
	       (launch-delay (string->number (or (configf:lookup *configdat* "setup" "launch-delay") "5"))))
      (if (> launch-delay delta)
	  (begin
	    (debug:print-info 0 *default-log-port* "Delaying launch of " test-name " for " (- launch-delay delta) " seconds")
	    (thread-sleep! (- launch-delay delta))
	    (loop (- (current-seconds) *last-launch*) launch-delay))))

	    (create-directory work-area #t)
	    (debug:print 0 *default-log-port* "WARNING: No disk work area specified - running in the test directory under tmp_run")))
      (set! cmdparms (base64:base64-encode 
		      (z3:encode-buffer 
		       (with-output-to-string
			 (lambda () ;; (list 'hosts     hosts)
			   (write (list (list 'testpath  test-path)
					;; (list 'transport (conc *transport-type*))
					;; (list 'serverinf *server-info*)
					(list 'homehost  (let* ((hhdat (common:get-homehost)))
							   (if hhdat
							       (car hhdat)
							       #f)))
					(list 'serverurl (if *runremote*
							     (remote-server-url *runremote*)
							     #f)) ;; 
					(list 'toppath   *toppath*)
					(list 'work-area work-area)
					(list 'test-name test-name) 
					(list 'runscript runscript) 
					(list 'run-id    run-id   )
					(list 'test-id   test-id  )
					;; (list 'item-path item-path )