Megatest

;;     GNU General Public License for more details.
;; 
;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;;======================================================================

(use srfi-1 data-structures posix regex-case (prefix base64 base64:)
     format dot-locking csv-xml z3 udp ;; sql-de-lite
     hostinfo md5 message-digest typed-records directory-utils stack
     matchable regex posix (srfi 18) extras ;; tcp 
     (prefix nanomsg nmsg:)
     (prefix sqlite3 sqlite3:)
     pkts (prefix dbi dbi:)
;; (use srfi-1 data-structures posix regex-case (prefix base64 base64:)
;;      format dot-locking csv-xml z3 udp ;; sql-de-lite
;;      hostinfo md5 message-digest typed-records directory-utils stack
;;      matchable regex posix (srfi 18) extras ;; tcp 
;;      (prefix nanomsg nmsg:)
;;      (prefix sqlite3 sqlite3:)
;;      pkts (prefix dbi dbi:)
     )

(declare (unit common))
;; (declare (uses commonmod))
;; (import commonmod)

(include "common_records.scm")
;;      )
;; 
;; (declare (unit common))
;; ;; (declare (uses commonmod))
;; ;; (import commonmod)
;; 
;; (include "common_records.scm")


;; (require-library margs)
;; (include "margs.scm")

;; (define old-exit exit)
;; 

(define *already-seen-runconfig-info* #f)

(define *test-meta-updated* (make-hash-table))
(define *globalexitstatus*  0) ;; attempt to work around possible thread issues
(define *passnum*           0) ;; when running track calls to run-tests or similar
;; (define *alt-log-file* #f)  ;; used by -log
(define *common:denoise*    (make-hash-table)) ;; for low noise printing
(define *default-log-port*  (current-error-port))
;; (define *default-log-port*  (current-error-port))
(define *time-zero* (current-seconds)) ;; for the watchdog
(define *default-area-tag* "local")

;; DATABASE
(define *dbstruct-db*         #f) ;; used to cache the dbstruct in db:setup. Goal is to remove this.
;; db stats
(define *db-stats*            (make-hash-table)) ;; hash of vectors < count duration-total >

(define *server-kind-run*   (make-hash-table))
(define *home-host*         #f)
;; (define *total-non-write-delay* 0)
(define *heartbeat-mutex*   (make-mutex))
(define *api-process-request-count* 0)
(define *max-api-process-requests* 0)
(define *server-overloaded*  #f)
(define *writes-total-delay*  0)

;; client
(define *rmt-mutex*         (make-mutex))     ;; remote access calls mutex 

;; RPC transport
(define *rpc:listener*      #f)

(define *launch-setup-mutex* (make-mutex))     ;; need to be able to call launch:setup often so mutex it and re-call the real deal only if *toppath* not set
(define *homehost-mutex*     (make-mutex))

;; Miscellaneous
(define *triggers-mutex*     (make-mutex))     ;; block overlapping processing of triggers
(define *numcpus-cache* (make-hash-table))

(use posix-extras pathname-expand files)
;; (use posix-extras pathname-expand files)

;; this plugs a hole in posix-extras in recent chicken versions > 4.9)
(let-values (( (chicken-release-number chicken-major-version)
               (apply values
                      (map string->number
                           (take
                            (string-split (chicken-version) ".")
                            2)))))
  (let ((resolve-pathname-broken?
         (or (> chicken-release-number 4)
             (and (eq? 4 chicken-release-number) (> chicken-major-version 9)))))
    (if resolve-pathname-broken?
        (define ##sys#expand-home-path pathname-expand))))
;; (let-values (( (chicken-release-number chicken-major-version)
;;                (apply values
;;                       (map string->number
;;                            (take
;;                             (string-split (chicken-version) ".")
;;                             2)))))
;;   (let ((resolve-pathname-broken?
;;          (or (> chicken-release-number 4)
;;              (and (eq? 4 chicken-release-number) (> chicken-major-version 9)))))
;;     (if resolve-pathname-broken?
;;         (define ##sys#expand-home-path pathname-expand))))
      
(define (realpath x) (resolve-pathname  (pathname-expand (or x "/dev/null")) ))
;; (define (realpath x) (resolve-pathname  (pathname-expand (or x "/dev/null")) ))
;; (define (realpath x)(pathname-expand (or x "/dev/null")) )
(define (realpath x)
  (with-input-from-pipe
   (string-append "readlink -f \""x"\"")
   read-line))

(define (common:get-this-exe-fullpath #!key (argv (argv)))
  (let* ((this-script
          (cond
           ((and (> (length argv) 2)
                 (string-match "^(.*/csi|csi)$" (car argv))
                 (string-match "^-(s|ss|sx|script)$" (cadr argv)))

;; Do NOT check if not on homehost!
;;
(define (common:exit-on-version-changed)
  (if (common:on-homehost?)
      (if (common:api-changed?)
	  (let* ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config"))
                (dbfile  (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.db"))
                (read-only (not (file-write-access? dbfile)))
                (read-only (not (file-writable? dbfile)))
                (dbstruct (db:setup #t)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (common:get-last-run-version))
            (cond
             ((get-environment-variable "MT_SKIP_DB_MIGRATE") #t)

    (begin
      (debug:print-error 0 *default-log-port* "received bad encoded string \"" instr "\", message: " ((condition-property-accessor 'exn 'message) exn) ", exn=" exn)
      (print-call-chain (current-error-port))
      #f)
    (read (open-input-string (base64:base64-decode instr))))
   (read (open-input-string (z3:decode-buffer (base64:base64-decode instr))))))

;; dot-locking egg seems not to work, using this for now
;; if lock is older than expire-time then remove it and try again
;; to get the lock
;;
(define (common:simple-file-lock fname #!key (expire-time 300))
  (let ((fmod-time (handle-exceptions
		       ext
		     (current-seconds)
		     (file-modification-time fname))))
    (if (common:file-exists? fname)
	(if (> (- (current-seconds) fmod-time) expire-time)
	    (begin
	      (handle-exceptions exn #f (delete-file* fname))	
	      (common:simple-file-lock fname expire-time: expire-time))
	    #f)
	(let ((key-string (conc (get-host-name) "-" (current-process-id))))
	  (with-output-to-file fname
	    (lambda ()
	      (print key-string)))
	  (thread-sleep! 0.25)
	  (if (common:file-exists? fname)
	      (handle-exceptions exn
                #f 
                (with-input-from-file fname
	  	  (lambda ()
		    (equal? key-string (read-line)))))
	      #f)))))

(define (common:simple-file-lock-and-wait fname #!key (expire-time 300))
  (let ((end-time (+ expire-time (current-seconds))))
    (let loop ((got-lock (common:simple-file-lock fname expire-time: expire-time)))
      (if got-lock
	  #t
	  (if (> end-time (current-seconds))
	      (begin
		(thread-sleep! 3)
		(loop (common:simple-file-lock fname expire-time: expire-time)))
	      #f)))))

(define (common:simple-file-release-lock fname)
  (handle-exceptions
      exn
      #f ;; I don't really care why this failed (at least for now)
    (delete-file* fname)))

;;======================================================================
;; S T A T E S   A N D   S T A T U S E S
;;======================================================================

;; BBnote: *common:std-states* - dashboard filter control and test control state buttons defined here; used in set-fields-panel and dboard:make-controls
(define *common:std-states*   ;; for toggle buttons in dashboard
  '(

  (time->string (seconds->local-time (current-seconds)) "%Y-%m-%d %H:%M:%S"))

;;======================================================================
;; currently the primary job of the watchdog is to run the sync back to megatest.db from the db in /tmp
;; if we are on the homehost and we are a server (by definition we are on the homehost if we are a server)
;;
(define (common:readonly-watchdog dbstruct)
  (let ((just-testing 0.0501))
  (thread-sleep! 0.05) ;; delay for startup
    (thread-sleep! just-testing)) ;; (/ 1 20)) ;; 0.051) ;; delay for startup
  (debug:print-info 13 *default-log-port* "common:readonly-watchdog entered.")
  ;; sync megatest.db to /tmp/.../megatst.db
  (let* ((sync-cool-off-duration   3)
        (golden-mtdb     (dbr:dbstruct-mtdb dbstruct))
        (golden-mtpath   (db:dbdat-get-path golden-mtdb))
        (tmp-mtdb        (dbr:dbstruct-tmpdb dbstruct))
        (tmp-mtpath      (db:dbdat-get-path tmp-mtdb)))

				  (let ((db (cdr *task-db*)))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! *task-db* 0 #f)
					  (set! *task-db* #f)))))
                              (http-client#close-all-connections!)
                              (http-client#close-idle-connections!)
                              ;; (if (and *runremote*
                              ;;          (remote-conndat *runremote*))
                              ;;     (begin
                              ;;       (http-client#close-all-connections!))) ;; for http-client
                              (if (not (eq? *default-log-port* (current-error-port)))
                                  (close-output-port *default-log-port*))
			      (set! *default-log-port* (current-error-port))) "Cleanup db exit thread"))

;;
(define (common:get-create-writeable-dir dirs)
  (if (null? dirs)
      #f
      (let loop ((hed (car dirs))
		 (tal (cdr dirs)))
	(let ((res (or (and (directory? hed)
			    (file-write-access? hed)
			    (file-writable? hed)
			    hed)
		       (handle-exceptions
			   exn
			   (begin
			     (debug:print-info 0 *default-log-port* "could not create " hed
					       ", this might cause problems down the road. exn=" exn)
			     #f)

(define (common:directory-writable? path-string)
  (handle-exceptions
      exn
    (begin
      (debug:print 0 *default-log-port* "Failed to identify access to " path-string ", exn=" exn)
      #f)
   (if (and (directory-exists? path-string)
            (file-write-access? path-string))
            (file-writable? path-string))
       path-string
       #f)))

(define (common:get-linktree)
  (or (getenv "MT_LINKTREE")
      (if *configdat*
	  (configf:lookup *configdat* "setup" "linktree")

				   (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)
						"] Failed to read .homehost file after trying five times. Giving up and exiting, message: "
						((condition-property-accessor 'exn 'message) exn))
				   (exit 1)))
			   (let ((hhf (conc *toppath* "/.homehost")))
			     (if (common:file-exists? hhf)
				 (with-input-from-file hhf read-line)
				 (if (file-write-access? *toppath*)
				 (if (file-writable? *toppath*)
				     (begin
				       (with-output-to-file hhf
					 (lambda ()
					   (print bestadrs)))
				       (begin
					 (mutex-unlock! *homehost-mutex*)
					 (car (common:get-homehost))))

  (if *toppath*
      (let* ((fullpath (conc *toppath* "/.sysdata/" key "-" dtype ".log"))
	     (delfile  (lambda (exn)
			 (debug:print-info 1 *default-log-port* " removing bad file " fullpath ", exn=" exn)
			 (delete-file* fullpath)
			 #f)))
	(if (and (file-exists? fullpath)
		 (file-read-access? fullpath))
		 (file-readable? fullpath))
	    (handle-exceptions
		exn
	      (begin
		(debug:print 0 *default-log-port* "failed to get cached info from " fullpath ", exn=" exn)
		#f)
	     (debug:print 2 *default-log-port* "reading file " fullpath)
	     (let ((real-age (- (current-seconds)

         (hh     (if hh-dat (car hh-dat) #f)))
    (common:wait-for-normalized-load maxnormload msg hh)))

(define (common:get-num-cpus remote-host)
  (let* ((actual-host (or remote-host (get-host-name))))
    ;; hosts had better not be changing the number of cpus too often!
    (or (hash-table-ref/default *numcpus-cache* actual-host #f)
	(let* ((numcpus (or (common:get-cached-info actual-host "num-cpus" age: (+ 2592000 (random 3600)))
	(let* ((numcpus (or (common:get-cached-info actual-host "num-cpus" age: (+ 2592000 (pseudo-random-integer 3600)))
			    (let* ((proc   (lambda ()
					     (let loop ((numcpu 0)
							(inl    (read-line)))
					       (if (eof-object? inl)
						   (if (> numcpu 0)
						       numcpu
						       #f) ;; if zero return #f so caller knows that things are not working

;; wait for normalized cpu load to drop below maxload
;;
(define (common:wait-for-normalized-load maxnormload msg remote-host #!optional (rem-tries 5))
  (let ((num-cpus (common:get-num-cpus remote-host)))
    (if num-cpus
	(common:wait-for-cpuload maxnormload num-cpus 15 msg: msg remote-host: remote-host)
	(begin
	  (thread-sleep! (random 60)) ;; we failed to get num cpus. wait a bit and try again
	  (thread-sleep! (pseudo-random-integer 60)) ;; we failed to get num cpus. wait a bit and try again
	  (if (> rem-tries 0)
	      (common:wait-for-normalized-load maxnormload msg remote-host (- rem-tries 1))
	      #f)))))

;;======================================================================
;; DO NOT CALL THIS DIRECTLY. It is called from common:wait-for-normalized-load
;;   count     - count down to zero, at some point we'd give up if the load never drops

			", normalized effective load: " normalized-effective-load
			))
     ;; overloaded and count expired (i.e. went to zero)
     (else
      (if (> num-tries 0) ;; should be "num-tries-left". 
	  (if (common:low-noise-print 30 (conc (round effective-load) "-load-acceptable-" effective-host))
	      (debug:print 0 *default-log-port* "Load on " effective-host " is acceptable at effective normalized load of "
			   effective-normalized-load "  continuing."))
			   normalized-effective-load "  continuing."))
	  (debug:print 0 *default-log-port* "Load on " effective-host ", "
		       first" could not be retrieved. Giving up and continuing."))))))

;;======================================================================
;; DO NOT CALL THIS DIRECTLY. It is called from common:wait-for-normalized-load
;;
;; (define (common:wait-for-cpuload maxload-in numcpus-in waitdelay #!key (count 1000) (msg #f)(remote-host #f)(force-maxload #f)(num-tries 5))

;; 					       ;; at least use 1
;; 	 (loadjmp (- first (if (> next (* numcpus 0.7)) ;; could do something with average of first and next?
;; 			       0
;; 			       next))) ;; we will force a conservative calculation any time next is large.
;; 	 (first-next-avg    (/ (+ first next) 2))
;; 	 ;; add some randomness to the time to break any alignment
;; 	 ;; where netbatch dumps many jobs to machines simultaneously
;;          (adjwait           (min (+ 300 (random 10)) (abs (* (+ (random 10)
;;          (adjwait           (min (+ 300 (pseudo-random-integer 10)) (abs (* (+ (pseudo-random-integer 10)
;; 								(/ (- 1000 count) 10)
;; 								waitdelay)
;; 							     (- first adjmaxload) ))))
;; 	 (load-jump-limit   (configf:lookup-number *configdat* "setup" "load-jump-limit"))
;; 	 ;; effective load accounts for load jumps, this should elminate all the first-next-avg, adjwait, load-jump-limit
;; 	 ;; etc.
;; 	 (effective-load    (common:get-intercept first next))
;; 	 (effective-host    (or remote-host "localhost"))
;; 	 (normalized-effective-load (/ effective-load numcpus))
;; 	 (will-wait                 (> normalized-effective-load maxload)))
;; 	 
;;     ;; let's let the user know once in a long while that load checking
;;     ;; is happening but not constantly report it
;;     #;(if (common:low-noise-print 30 (conc "cpuload" (or remote-host "localhost"))) ;; (> (random 100) 75) ;; about 25% of the time
;;     #;(if (common:low-noise-print 30 (conc "cpuload" (or remote-host "localhost"))) ;; (> (pseudo-random-integer 100) 75) ;; about 25% of the time
;; 	(debug:print-info 1 *default-log-port* "Checking cpuload on " (or remote-host "localhost") ", maxload: " maxload
;;     ", load: " first ", adjmaxload: " adjmaxload ", loadjmp: " loadjmp))
;; 
;;     (debug:print-info 1 *default-log-port*
;; 		      "On host: " effective-host
;; 		      ", effective load: " effective-load
;; 		      ", numcpus: " numcpus

     (lambda (disk-num)
       (let* ((dirpath    (cadr (assoc disk-num disks)))
	      (freespc    (cond
			   ((not (directory? dirpath))
			    (if (common:low-noise-print 300 "disks not a dir " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a directory - ignoring it."))
			    -1)
			   ((not (file-write-access? dirpath))
			   ((not (file-writable? dirpath))
			    (if (common:low-noise-print 300 "disks not writeable " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not writeable - ignoring it."))
			    -1)
			   ((not (eq? (string-ref dirpath 0) #\/))
			    (if (common:low-noise-print 300 "disks not a proper path " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a fully qualified path - ignoring it."))
			    -1)
			   (else
			    (get-df dirpath))))
	      (free-inodes (cond
			   ((not (directory? dirpath))
			    (if (common:low-noise-print 300 "disks not a dir " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a directory - ignoring it."))
			    -1)
			   ((not (file-write-access? dirpath))
			   ((not (file-writable? dirpath))
			    (if (common:low-noise-print 300 "disks not writeable " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not writeable - ignoring it."))
			    -1)
			   ((not (eq? (string-ref dirpath 0) #\/))
			    (if (common:low-noise-print 300 "disks not a proper path " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a fully qualified path - ignoring it."))
			    -1)

     (for-each
      (lambda (pktsdir) ;; look at all
	(cond
	 ((not (common:file-exists? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory " pktsdir " does not exist."))
	 ((not (directory? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not a directory."))
	 ((not (file-read-access? pktsdir))
	 ((not (file-readable? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not readable."))
	 (else
	  (debug:print-info 0 *default-log-port* "Loading packets found in " pktsdir)
	  (let ((pkts (glob (conc pktsdir "/*.pkt"))))
	    (for-each
	     (lambda (pkt)
	       (let* ((uuid    (cadr (string-match ".*/([0-9a-f]+).pkt" pkt)))

	     (begin
	       (debug:print 0 *default-log-port* "joining threads failed. exn=" exn)
               #t) ;; just ignore it, it might have died in the meantime so joining it will throw an exception
             (thread-join! thread))
           )))
   (hash-table-keys *common:thread-punchlist*)))

(define (dtests:get-pre-command #!key (default-override #f))
  (let* ((orig-pre-command "export CMD='")
         (viewscreen-pre-command  "viewscreen ")
         (use-viewscreen (configf:lookup *configdat* "dashboard" "use-viewscreen"))
         (default-pre-command (if use-viewscreen viewscreen-pre-command orig-pre-command))
         (cfg-ovrd (configf:lookup *configdat* "dashboard" "pre-command")))
    (or cfg-ovrd default-override default-pre-command))) ;; "xterm -geometry 180x20 -e \""))

  
(define (dtests:get-post-command #!key (default-override #f))
  (let* ((orig-post-command (conc "';xterm -geometry 180x20 -e \"(echo; echo -n START:;date +ww%U.%w-$H:%M:%S;echo;echo $CMD;echo;$CMD)|&"
                                 "tee -a runlog-`date +ww%U.%w-%H:%M`.log;echo Press any key to continue;bash -c 'read -n 1 -s'\" &"))
         (viewscreen-post-command  "")
         (use-viewscreen (configf:lookup *configdat* "dashboard" "use-viewscreen"))
         (default-post-command (if use-viewscreen viewscreen-post-command orig-post-command))
         (cfg-ovrd (configf:lookup *configdat* "dashboard" "post-command")))
    (or cfg-ovrd default-override default-post-command))) ;; ";echo Press any key to continue;bash -c 'read -n 1 -s'\" &")))

;;======================================================================
;; (define *common:telemetry-log-state* 'startup)
;; (define *common:telemetry-log-socket* #f)
;; 
;; (define (common:telemetry-log-open)
;;   (if (eq? *common:telemetry-log-state* 'startup)
;;       (let* ((serverhost (configf:lookup *configdat* "telemetry" "host"))