Megatest

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

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




(include "common_records.scm")
(include "key_records.scm")
(include "db_records.scm")

;;======================================================================
;; ezsteps

		    (if (steprun-good? logpro-used (launch:einf-exit-code exit-info))
			(if (not (null? tal))
			    (loop (car tal) (cdr tal) stepname))
			(debug:print 4 *default-log-port* "WARNING: step " (car ezstep) " failed. Stopping")))
		  (debug:print 4 *default-log-port* "WARNING: a prior step failed, stopping at " ezstep)))))))

(define (launch:monitor-job run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m work-area runtlim misc-flags)
  (let* ((update-period (string->number (or (configf:lookup *configdat* "setup" "test-stats-update-period") "30")))
         (start-seconds (current-seconds))
	 (calc-minutes  (lambda ()
			  (inexact->exact 
			   (round 
			    (- 
			     (current-seconds) 
			     start-seconds)))))
	 (kill-tries 0))
    ;; (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area)
    ;; (tests:set-full-meta-info test-id run-id (calc-minutes) work-area)
    (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area 10)
    (let loop ((minutes   (calc-minutes))
	       (cpu-load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
	       (disk-free (get-df (current-directory)))
               (last-sync (current-seconds)))
      (let* ((over-time     (> (current-seconds) (+ last-sync update-period)))
             (new-cpu-load  (let* ((load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
                                   (delta (abs (- load cpu-load))))
                              (if (> delta 0.1) ;; don't bother updating with small changes
                                  load
                                  #f)))
             (new-disk-free (let* ((df    (if over-time ;; only get df every 30 seconds
                                              (get-df (current-directory))
                                              disk-free))
                                   (delta (abs (- df disk-free))))
                              (if (and (> df 0)
                                       (> (/ delta df) 0.1)) ;; (> delta 200) ;; ignore changes under 200 Meg
                                  df
                                  #f)))
             (do-sync       (or new-cpu-load new-disk-free over-time)))
        (debug:print 4 *default-log-port* "cpu: " new-cpu-load " disk: " new-disk-free " last-sync: " last-sync " do-sync: " do-sync)
	(set! kill-job? (or (test-get-kill-request run-id test-id) ;; run-id test-name itemdat))
			    (and runtlim (let* ((run-seconds   (- (current-seconds) start-seconds))
						(time-exceeded (> run-seconds runtlim)))
					   (if time-exceeded
					       (begin
						 (debug:print-info 0 *default-log-port* "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" run-seconds " seconds, limit=" runtlim)
						 #t)
					       #f)))))
        (if do-sync
            (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f))
	(if kill-job? 
	    (begin
	      (mutex-lock! m)
	      ;; NOTE: The pid can change as different steps are run. Do we need handshaking between this
	      ;;       section and the runit section? Or add a loop that tries three times with a 1/4 second
	      ;;       between tries?
	      (let* ((pid1 (launch:einf-pid exit-info)) ;; (vector-ref exit-info 0))

	      (mutex-unlock! m)
	      ;; no point in sticking around. Exit now.
	      (exit)))
	(if (hash-table-ref/default misc-flags 'keep-going #f)
	    (begin
	      (thread-sleep! 3) ;; (+ 3 (random 6))) ;; add some jitter to the call home time to spread out the db accesses
	      (if (hash-table-ref/default misc-flags 'keep-going #f)  ;; keep originals for cpu-load and disk-free unless they change more than the allowed delta
		  (loop (calc-minutes)
                        (or new-cpu-load cpu-load)
                        (or new-disk-free disk-free)
                        (if do-sync (current-seconds) last-sync)))))))
    (tests:update-central-meta-info run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f))) ;; NOTE: Checking twice for keep-going is intentional


(define (launch:execute encoded-cmd)
  (let* ((cmdinfo    (common:read-encoded-string encoded-cmd))
	 (tconfigreg #f))
    (setenv "MT_CMDINFO" encoded-cmd)

	    (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 (rmt:get-test-info-by-id run-id test-id))
		 (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
	     ((member (db:test-get-state test-info) '("INCOMPLETE" "KILLED" "UNKNOWN" "KILLREQ" "STUCK")) ;; prior run of this test didn't complete, go ahead and try to rerun
	      (debug:print 0 *default-log-port* "INFO: test is INCOMPLETE or KILLED, treat this execute call as a rerun request")
	      ;; (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")
	      (rmt:test-set-state-status run-id test-id "REMOTEHOSTSTART" "n/a" #f)
	      ) ;; prime it for running

	      (debug:print-error 0 *default-log-port* "test state is " (db:test-get-state test-info) ", cannot proceed")
	      (exit))))
	  
	  (debug:print 2 *default-log-port* "Exectuing " test-name " (id: " test-id ") on " (get-host-name))
	  (set! keys       (rmt:get-keys))
	  ;; (runs:set-megatest-env-vars run-id inkeys: keys inkeyvals: keyvals) ;; these may be needed by the launching process
	  ;; one of these is defunct/redundant ...
	  (if (not (launch:setup force-reread: #t))
	      (begin
		(debug:print 0 *default-log-port* "Failed to setup, exiting") 
		;; (sqlite3:finalize! db)
		;; (sqlite3:finalize! tdb)
		(exit 1)))
	  (change-directory *toppath*) 

		  (tests:summarize-items run-id test-id test-name #f))
	      (tests:summarize-test run-id test-id)  ;; don't force - just update if no
	      (rmt:update-run-stats run-id (rmt:get-raw-run-stats run-id)))
	    (mutex-unlock! m)
	    (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: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")

;;   returns:
;;     *toppath*
;;   side effects:
;;     sets; *configdat*    (megatest.config info)
;;           *runconfigdat* (runconfigs.config info)
;;           *configstatus* (status of the read data)
;;
(define (launch:setup #!key (force-reread #f) (areapath #f))
  (mutex-lock! *launch-setup-mutex*)
  (if (and *toppath*
	   (eq? *configstatus* 'fulldata) (not force-reread)) ;; got it all
      (begin
	(debug:print 2 *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-reread: force-reread areapath: areapath)))
	(mutex-unlock! *launch-setup-mutex*)
	res)))

;; return paths depending on what info is available.
;;
(define (launch:get-cache-file-paths areapath toppath target mtconfig)
  (let* ((use-cache (common:use-cache?))
         (runname  (common:args-get-runname))
         (linktree (common:get-linktree))
         (testname (common:get-full-test-name))
         (rundir   (if (and runname target linktree)
                       (common:directory-writable? (conc linktree "/" target "/" runname))
                       #f))
         (testdir  (if (and rundir testname)
                       (common:directory-writable? (conc rundir "/" testname))
                       #f))
         (cachedir (or testdir rundir))
         (mtcachef (and cachedir (conc cachedir "/" ".megatest.cfg-"  megatest-version "-" megatest-fossil-hash)))
         (rccachef (and cachedir (conc cachedir "/" ".runconfigs.cfg-"  megatest-version "-" megatest-fossil-hash))))
    (debug:print-info 6 *default-log-port* 
                      "runname=" runname 
                      "\n  linktree=" linktree
                      "\n  testname=" testname
                      "\n  rundir=" rundir 
                      "\n  testdir=" testdir 
                      "\n  cachedir=" cachedir
                      "\n  mtcachef=" mtcachef
                      "\n  rccachef=" rccachef)
    (cons mtcachef rccachef)))

(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  (or *toppath* areapath (getenv "MT_RUN_AREA_HOME"))) ;; preserve toppath


	     (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

			     (car cachefiles))) ;; (and cachedir (conc cachedir "/" ".megatest.cfg-"  megatest-version "-" megatest-fossil-hash)))
	     (rccachef   (if (null? cachefiles)
			     #f
			     (cdr cachefiles)))) ;; (and cachedir (conc cachedir "/" ".runconfigs.cfg-"  megatest-version "-" megatest-fossil-hash)))


	      ;; (cancreate (and cachedir (common:file-exists? cachedir)(file-write-access? cachedir) (not (common:in-running-test?)))))




	(set! *toppath* toppath) ;; This is needed when we are running as a test using CMDINFO as a datasource
        ;;(BB> "launch:setup-body -- cachefiles="cachefiles)
	(cond
	 ;; if mtcachef exists just read it, however we need to assume toppath is available in $MT_RUN_AREA_HOME
	 ((and (not force-reread) mtcachef (common:file-exists? mtcachef) (get-environment-variable "MT_RUN_AREA_HOME") use-cache)
          ;;(BB> "launch:setup-body -- cond branch 1 - use-cache")
          (set! *configdat*    (configf:read-alist mtcachef))
          ;;(BB> "launch:setup-body -- 1 set! *configdat*="*configdat*)
	  (set! *runconfigdat* (configf:read-alist rccachef))
	  (set! *configinfo*   (list *configdat*  (get-environment-variable "MT_RUN_AREA_HOME")))
	  (set! *configstatus* 'fulldata)
	  (set! *toppath*      (get-environment-variable "MT_RUN_AREA_HOME"))
	  *toppath*)
	 ;; we have all the info needed to fully process runconfigs and megatest.config
	 ((and (not force-reread) mtcachef) ;; BB- why are we doing this without asking if caching is desired?
          ;;(BB> "launch:setup-body -- cond branch 2")
	  (let* ((first-pass    (find-and-read-config        ;; NB// sets MT_RUN_AREA_HOME as side effect
				 mtconfig
				 environ-patt: "env-override"
				 given-toppath: toppath
				 pathenvvar: "MT_RUN_AREA_HOME"))
		 (first-rundat  (let ((toppath (if toppath 
						   toppath
						   (car first-pass))))
				  (read-config ;; (conc toppath "/runconfigs.config") ;; this should be converted to runconfig:read but it is non-trivial, leaving it for now.
				   (conc (if (string? toppath)
					     toppath
					     (get-environment-variable "MT_RUN_AREA_HOME"))
					 "/runconfigs.config")
				   *runconfigdat* #t 
				   sections: sections))))
	    (set! *runconfigdat* first-rundat)
	    (if first-pass  ;; 
		(begin
                  ;;(BB> "launch:setup-body -- \"first-pass\"=first-pass")
		  (set! *configdat*  (car first-pass))
                  ;;(BB> "launch:setup-body -- 2 set! *configdat*="*configdat*)
		  (set! *configinfo* first-pass)
		  (set! *toppath*    (or toppath (cadr first-pass))) ;; use the gathered data unless already have it
		  (set! toppath      *toppath*)
		  (if (not *toppath*)
		      (begin
			(debug:print-error 0 *default-log-port* "you are not in a megatest area!")
			(exit 1)))

					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)
						     (setenv (car kt) (cadr kt)))
						   key-vals)
					 (read-config (conc toppath "/runconfigs.config") *runconfigdat* #t ;; consider using runconfig:read some day ...
						      sections: sections)))
                         (cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
                         (mtcachef     (car cachefiles))
                         (rccachef     (cdr cachefiles)))
		    (if rccachef (configf:write-alist runconfigdat rccachef))
		    (set! *runconfigdat* runconfigdat)
		    (if mtcachef (configf:write-alist *configdat* mtcachef))
		    (if (and rccachef mtcachef) (set! *configstatus* 'fulldata))))
		;; no configs found? should not happen but let's try to recover gracefully, return an empty hash-table
		(begin (set! *configdat* (make-hash-table))
                       ;;(BB> "launch:setup-body -- 3 set! *configdat*="*configdat*)
                       )
		)))
	 ;; else read what you can and set the flag accordingly
	 (else
          ;;(BB> "launch:setup-body -- cond branch 3 - else")
	  (let* ((cfgdat   (find-and-read-config 
			    (or (args:get-arg "-config") "megatest.config")
			    environ-patt: "env-override"
			    given-toppath: (get-environment-variable "MT_RUN_AREA_HOME")
			    pathenvvar: "MT_RUN_AREA_HOME")))

            (if (and cfgdat (list? cfgdat) (> (length cfgdat) 0) (hash-table? (car cfgdat)))
		(let* ((toppath  (or (get-environment-variable "MT_RUN_AREA_HOME")(cadr cfgdat)))
		       (rdat     (read-config (conc toppath  ;; convert this to use runconfig:read!
						    "/runconfigs.config") *runconfigdat* #t sections: sections)))
		  (set! *configinfo*   cfgdat)
		  (set! *configdat*    (car cfgdat))
		  (set! *runconfigdat* rdat)
		  (set! *toppath*      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)
	      (set! *toppath* #f) ;; force it to be false so we return #f
	      #f
	      ))
        ;; one more attempt to cache the configs for future reading
        (let* ((cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
               (mtcachef     (car cachefiles))
               (rccachef     (cdr cachefiles)))
          (if (and rccachef *runconfigdat*) (configf:write-alist *runconfigdat* rccachef))
          (if (and mtcachef *configdat*)    (configf:write-alist *configdat* mtcachef))
          (if (and rccachef mtcachef *runconfigdat* *configdat*)
              (set! *configstatus* 'fulldata)))

	;; 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*)))

;;    - 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 (configf:lookup-number *configdat* "setup" "launch-delay" default: 1)))
      (if (> launch-delay delta)
	  (begin
	    (if (common:low-noise-print 1200 "test launch delay") ;; every two hours or so remind the user about launch delay.
		(debug:print-info 0 *default-log-port* "NOTE: test launches are delayed by " launch-delay " seconds. See megatest.config launch-delay setting to adjust.")) ;; launch of " test-name " for " (- launch-delay delta) " seconds"))
	    (thread-sleep! (- launch-delay delta))
	    (loop (- (current-seconds) *last-launch*) launch-delay))))
    (change-directory *toppath*)
    (alist->env-vars ;; consolidate this code with the code in megatest.scm for "-execute", *maybe* - the longer they are set the longer each launch takes (must be non-overlapping with the vars)
     (append
      (list
       (list "MT_RUN_AREA_HOME" *toppath*)