Megatest

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

(declare (unit common))
(declare (uses commonmod))
(declare (uses rmtmod))
(declare (uses debugprint))
(declare (uses mtargs))


(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 posix-extras pathname-expand files)


(import commonmod
	debugprint
	rmtmod
	(prefix mtargs args:))




(include "common_records.scm")


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

;; (define old-exit exit)
;; 
;; (define (exit . code)
;;   (if (null? code)
;;       (old-exit)
;;       (old-exit code)))

(define (stop-the-train)
  (thread-start! (make-thread (lambda ()
				(let loop ()
				  (if (and *toppath*
					   (file-exists? (conc *toppath*"/stop-the-train")))
				      (let* ((msg (conc "ERROR: found file "*toppath*"/stop-the-train, exiting immediately")))
					;; yes, print to current-output-port AND *default-log-port*, annoying but necessary I think
					(print msg)

					(debug:print 0 *default-log-port* msg)



					(exit 1)))
				  (thread-sleep! 5)
				  (loop))))))

;; execute thunk, return value.  If exception thrown, trap exception, return #f, and emit nonfatal condition note to *default-log-port* .
;; arguments - thunk, message
(define (common:fail-safe thunk warning-message-on-exception)

;; (define *common:denoise*    (make-hash-table)) ;; for low noise printing
(define *default-log-port*  (current-error-port))
(define *time-zero* (current-seconds)) ;; for the watchdog
(define *on-exit-procs* '()) ;; add procs to this list to be executed on exit
(define *default-area-tag* "local")

;; DATABASE
;; (define *dbstruct-dbs*         #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 *db-stats-mutex*      (make-mutex))
;; db access
(define *db-last-access*      (current-seconds)) ;; last db access, used in server
;; (define *db-write-access*     #t)
;; db sync
;; (define *db-last-sync*        0)                 ;; last time the sync to megatest.db happened
(define *db-sync-in-progress* #f)                ;; if there is a sync in progress do not try to start another
;; (define *db-multi-sync-mutex* (make-mutex))      ;; protect access to *db-sync-in-progress*, *db-last-sync*

(define *transport-type*    'http)             ;; override with [server] transport http|rpc|nmsg
(define *runremote*         #f)                ;; if set up for server communication this will hold <host port>
;; (define *max-cache-size*    0)
(define *logged-in-clients* (make-hash-table))
(define *server-id*         #f)
(define *server-info*       #f)  ;; good candidate for easily convert to non-global
(define *time-to-exit*      #f)
(define *server-run*        #t)
(define *run-id*            #f)
(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 *common:this-exe-fullpath* (common:get-this-exe-fullpath))
(define *common:this-exe-dir*      (pathname-directory *common:this-exe-fullpath*))
(define *common:this-exe-name*     (pathname-strip-directory *common:this-exe-fullpath*))

(define (common:get-sync-lock-filepath)
  (let* ((tmp-area     (common:get-db-tmp-area))
         (lockfile     (conc tmp-area "/megatest.db.lock")))
    lockfile))

(define *common:logpro-exit-code->status-sym-alist*
  '( ( 0 . pass )
     ( 1 . fail )
     ( 2 . warn )

  (- megatest-version (common:get-last-run-version-number)))

(define (common:version-changed?)
  (not (equal? (common:get-last-run-version)
               (common:version-signature))))


;; From 1.70 to 1.80, db's are compatible.



(define (common:api-changed?)
  (let* (
    (megatest-major-version (substring (->string megatest-version) 0 4))
    (run-major-version (substring (conc (common:get-last-run-version)) 0 4))
   )
   (and (not (equal? megatest-major-version "1.80"))
     (not (equal? megatest-major-version megatest-run-version)))
  )
)

;;======================================================================
;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma
;;
(define (common:cleanup-db dbstruct #!key (full #f))
  (case (rmt:transport-mode)
    ((http)
     (apply db:multi-db-sync 
	    dbstruct
	    'schema
	    'killservers
	    'adj-target
	    'new2old
	    '(dejunk)
	    ))
    ((tcp nfs)
     (debug:print 0 *default-log-port* "WARNING: cleanup-db NOT implemented yet for tcp and nfs.")
     #;(apply db:multi-db-sync 
	    dbstruct
	    'schema
	    'killservers
	    'adj-target
	    'new2old
	    '(dejunk)
	    )))

(define (common:exit-on-version-changed)
  (if (and *toppath*              ;; do nothing if *toppath* not yet provided
	   (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") ".mtdb/main.db"))
                 (read-only (not (file-write-access? dbfile)))
                 (dbstruct (db:setup #t))) ;; (db:setup-db *dbstruct-dbs* *toppath* #f))) ;;  #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)
             ((and (common:file-exists? mtconf) (common:file-exists? dbfile) (not read-only)

	    (if (equal? thepath "/")
		(begin
		  (debug:print-error 0 *default-log-port* "Unable to find megatest home directory.")
		  #f)
		(loop (pathname-directory thepath)))))
      ))


(define (common:db-tmp-area-path)
  (conc "/tmp/" 
         (current-user-name)
	 "/megatest_localdb/"
	 (common:get-testsuite-name)
         "/"
	 (string-translate *toppath* "/" ".")
  )
)


;;======================================================================
;; redefine for future cleanup (converge on area-name, the more generic
;;
(define common:get-area-name common:get-testsuite-name)

(define (common:get-db-tmp-area . junk)

		     (tsname (common:get-testsuite-name))
		     (dbpath (common:get-create-writeable-dir
			      (list (conc "/tmp/" (current-user-name)
					  "/megatest_localdb/"
					  tsname "/"
					  (string-translate toppath "/" "."))
				    (conc "/tmp/" (current-process-id) ;; just in case we have an issue with the dir by own user name
					  "/megatest_localdb/"
					  tsname
					  (string-translate toppath "/" "."))
				    ))))
		(set! *db-cache-path* dbpath)
		;; ensure megatest area has .mtdb
		(let ((dbarea (conc *toppath* "/.mtdb")))
		  (if (not (file-exists? dbarea))

;;======================================================================
;; lazy-safe get file mod time. on any error (file not existing etc.) return 0
;;
(define (common:lazy-modification-time fpath)
  (handle-exceptions
      exn
    (begin
      (debug:print 0 *default-log-port* "Failed to get modifcation time for " fpath ", treating it as zero. exn=" exn)
      0)
    (if (file-exists? fpath)
	(file-modification-time fpath)
	0)))

;;======================================================================
;; find timestamp of newest file associated with a sqlite db file

;;
;; (define (common:print-delay-table)
;;   (let loop ((x 0))
;;     (print x "," (common:get-delay x 1))
;;     (if (< x 2)
;; 	(loop (+ x 0.1)))))

(define (get-cpu-load #!key (remote-host #f))
  (car (common:get-cpu-load remote-host)))

;;======================================================================
;;   (let* ((load-res (process:cmd-run->list "uptime"))
;; 	 (load-rx  (regexp "load average:\\s+(\\d+)"))
;; 	 (cpu-load #f))
;;     (for-each (lambda (l)
;; 		(let ((match (string-search load-rx l)))
;; 		  (if match
;; 		      (let ((newval (string->number (cadr match))))
;; 			(if (number? newval)
;; 			    (set! cpu-load newval))))))
;; 	      (car load-res))
;;     cpu-load))

;;======================================================================
;; get values from cached info from dropping file in logs dir
;;  e.g. key is host and dtype is normalized-load
;;
(define (common:get-cached-info key dtype #!key (age 10))
  (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))
	    (handle-exceptions
		exn
	      (begin

	    exn
	  (begin
	    (debug:print 0 *default-log-port* "failed to write file " fullpath ", exn=" exn)
	    #f)
	  (with-output-to-file fullpath (lambda ()(pp dat)))))
      #f))

(define (common:raw-get-remote-host-load-orig remote-host)
  (handle-exceptions
      exn
    (begin
      (debug:print 0 *default-log-port* "failed to ssh to " remote-host " and get loadavg. exn=" exn)
      #f) ;; more specific handling of errors needed
    (with-input-from-pipe 
     (conc "ssh " remote-host " cat /proc/loadavg")
     (lambda ()(list (read)(read)(read))))))

(define (common:raw-get-remote-host-load remote-host)
  (let* ((inp #f))
    (handle-exceptions
	exn
      (begin
	(close-input-pipe inp)
	(debug:print 0 *default-log-port* "failed to ssh to " remote-host " and get loadavg. exn=" exn)

  (handle-exceptions
      exn
    (begin
      (debug:print 0 *default-log-port* "failed to ssh or read loadavg from host " remote-host ", exn=" exn)
      '(-99 -99 -99))
    (let* ((actual-hostname (or remote-host (get-host-name) "localhost")))
      (or (common:get-cached-info actual-hostname "cpu-load")
	  (let ((result (if remote-host

			    (map (lambda (res)
				   (if (eof-object? res) 9e99 res))
			         (common:raw-get-remote-host-load remote-host))
			    (with-input-from-file "/proc/loadavg" 
			      (lambda ()(list (read)(read)(read)))))))
	    (match
		result

						   (if (> numcpu 0)
						       numcpu
						       #f) ;; if zero return #f so caller knows that things are not working
						   (loop (if (string-match "^processor\\s+:\\s+\\d+$" inl)
							     (+ numcpu 1)
							     numcpu)
							 (read-line))))))
				   (result (if remote-host

					       (common:generic-ssh
						(conc "ssh " remote-host " cat /proc/cpuinfo")
						proc -1)
					       (with-input-from-file "/proc/cpuinfo" proc))))
			      (if (and (number? result)
				       (> result 0))
				  (common:write-cached-info actual-host "num-cpus" result))

;; returns: ok/not dbspace required-space
;;
(define (common:check-db-dir-space)
  (let* ((required (string->number 
                    ;; default is 1GB (or actually a billion bytes) This is the number of 1 kB blocks.
		    (or (configf:lookup *configdat* "setup" "dbdir-space-required")
			"1000000")))
	 (dbdir    (common:get-db-tmp-area)) ;; (db:get-dbdir))
	 (tdbspace (common:check-space-in-dir dbdir required))
	 (mdbspace (common:check-space-in-dir *toppath* required)))
    (sort (list tdbspace mdbspace) (lambda (a b)
				     (< (cadr a)(cadr b))))))

;;======================================================================
;; check available space in dbdir, exit if insufficient