Megatest

;;======================================================================
;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.
;;======================================================================

(use srfi-1 posix regex-case base64 format dot-locking csv-xml z3 nanomsg sql-de-lite hostinfo)
(require-extension regex posix)

(require-extension (srfi 18) extras tcp rpc)

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

  (conc megatest-version "-" megatest-fossil-hash))

(define (common:version-signature)
  (conc megatest-version "-" (substring megatest-fossil-hash 0 4)))

;; from metadat lookup MEGATEST_VERSION
;;
(define (common:get-last-run-version)
  (rmt:get-var "MEGATEST_VERSION"))





(define (common:set-last-run-version)
  (rmt:set-var "MEGATEST_VERSION" (common:version-signature)))

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

;; Move me elsewhere ...

;;
(define (common:cleanup-db)
  (db:multi-db-sync 
   #f ;; do all run-ids
   ;; 'new2old
   'killservers
   'dejunk
   ;; 'adj-testids
   ;; 'old2new
   'new2old)
  (if (common:version-changed?)
      (common:set-last-run-version)))



(define (common:exit-on-version-changed)
  (if (common:version-changed?)
      (let ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config")))
        (debug:print 0 *default-log-port*
		     "ERROR: Version mismatch!\n"
		     "   expected: " (common:version-signature) "\n"
		     "   got:      " (common:get-last-run-version))
	(if (and (file-exists? mtconf)
		 (eq? (current-user-id)(file-owner mtconf))) ;; safe to run -cleanup-db
	    (begin
	      (debug:print 0 *default-log-port* "   I see you are the owner of megatest.config, attempting to cleanup and reset to new version")
	      (handle-exceptions

		    (if (> curr-rownum rownum) curr-rownum rownum)
		    (if (> curr-colnum colnum) curr-colnum colnum)
		    ))))))

;;======================================================================
;; S Y S T E M   S T U F F
;;======================================================================









;; return a nice clean pathname made absolute
(define (common:nice-path dir)
  (let ((match (string-match "^(~[^\\/]*)(\\/.*|)$" dir)))
    (if match ;; using ~ for home?
	(common:nice-path (conc (common:read-link-f (cadr match)) "/" (caddr match)))
	(normalize-pathname (if (absolute-pathname? dir)

	(debug:print-error 0 *default-log-port* "command \"/bin/readlink -f " path "\" failed.")
	path) ;; just give up
    (with-input-from-pipe
	(conc "/bin/readlink -f " path)
      (lambda ()
	(read-line)))))

(define (get-cpu-load)
  (car (common:get-cpu-load)))
;;   (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 cpu load by reading from /proc/loadavg, return all three values
;;
(define (common:get-cpu-load)






  (with-input-from-file "/proc/loadavg" 
    (lambda ()(list (read)(read)(read)))))

(define (common:wait-for-cpuload maxload numcpus waitdelay #!key (count 1000) (msg #f))
  (let* ((loadavg (common:get-cpu-load))
	 (first   (car loadavg))
	 (next    (cadr loadavg))
	 (adjload (* maxload numcpus))
	 (loadjmp (- first next)))
    (cond
     ((and (> first adjload)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load " first " exceeding max of " adjload (if msg msg ""))
      (thread-sleep! waitdelay)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1)))
     ((and (> loadjmp numcpus)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load jump " loadjmp " > numcpus " numcpus (if msg msg ""))
      (thread-sleep! waitdelay)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1))))))

(define (common:get-num-cpus)
  (with-input-from-file "/proc/cpuinfo"
    (lambda ()
      (let loop ((numcpu 0)
		 (inl    (read-line)))
	(if (eof-object? inl)
	    numcpu
	    (loop (if (string-match "^processor\\s+:\\s+\\d+$" inl)
		      (+ numcpu 1)
		      numcpu)
		  (read-line)))))))






;; wait for normalized cpu load to drop below maxload
;;
(define (common:wait-for-normalized-load maxload #!key (msg #f))
  (let ((num-cpus (common:get-num-cpus)))
    (common:wait-for-cpuload maxload num-cpus 15 msg: msg)))

(define (get-uname . params)
  (let* ((uname-res (process:cmd-run->list (conc "uname " (if (null? params) "-a" (car params)))))
	 (uname #f))
    (if (null? (car uname-res))
	"unknown"

     ;; NOTE: This can fail when there is no mention of the host in /etc/hosts. FIXME
     (vector->list (hostinfo-addresses (hostname->hostinfo hostname))))
    (string-intersperse 
     (map number->string
	  (u8vector->list
	   (if res res (hostname->ip hostname)))) ".")))

(define (common:open-nm-req addr)
  (let* ((req (nn-socket 'req))
	 (res (nn-connect req addr)))
    req))

;; (with-output-to-string (lambda ()(serialize obj)))
(define (common:nm-send-receive soc msg)
  (nn-send soc msg)
  (nn-recv soc))

(define (common:close-nm-req soc)
  (nn-close soc))

(define (common:send-dboard-main-changed)
  (let* ((dashboard-ips (mddb:get-dashboards)))
    (for-each
     (lambda (ipadr)
       (let* ((soc (common:open-nm-req (conc "tcp://" ipadr)))
	      (msg (conc "main " *toppath*))
	      (res (common:nm-send-receive-timeout soc msg)))
	 (if (not res) ;; couldn't reach that dashboard - remove it from db
	     (print "ERROR: couldn't reach dashboard " ipadr))
	 res))
     dashboard-ips)))
    
(define (common:nm-send-receive-timeout req msg)
  (let* ((key     "ping")
	 (success #f)
	 (keepwaiting #t)
	 (result  #f)
	 (sendrec (make-thread
		   (lambda ()
		     (nn-send req msg)
		     (set! result (nn-recv req))
		     (set! success #t))
		   "send-receive"))
	 (timeout (make-thread (lambda ()
				 (let loop ((count 0))
				   (thread-sleep! 1)
				   (print "still waiting after count seconds...")
				   (if (and keepwaiting (< count 10))
				       (loop (+ count 1))))
				 (if keepwaiting
				     (begin
				       (print "timeout waiting for reply")
				       (thread-terminate! sendrec))))
			       "timeout")))
    (handle-exceptions
     exn
     (begin
       (print-call-chain)
       (print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
       (print "exn=" (condition->list exn)))
     (thread-start! timeout)
     (thread-start! sendrec)
     (thread-join!  sendrec)
     (if success (thread-terminate! timeout)))
    result))
    
(define (common:ping-nm req)
  ;; send a random number and check that we get it back
  (let* ((key     "ping")
	 (success #f)
	 (keepwaiting #t)
	 (ping    (make-thread
		   (lambda ()
		     (print "ping: sending string \"" key "\", expecting " (current-process-id))
		     (nn-send req key)
		     (let ((result  (nn-recv req)))
		       (if (equal? (conc (current-process-id)) result)
			   (begin
			     (print "ping, success: received \"" result "\"")
			     (set! success #t))
			   (begin
			     (print "ping, failed: received key \"" result "\"")
			     (set! keepwaiting #f)
			     (set! success #f)))))
		   "ping"))
	 (timeout (make-thread (lambda ()
				 (let loop ((count 0))
				   (thread-sleep! 1)
				   (print "still waiting after count seconds...")
				   (if (and keepwaiting (< count 10))
				       (loop (+ count 1))))
				 (if keepwaiting
				     (begin
				       (print "timeout waiting for ping")
				       (thread-terminate! ping))))
			       "timeout")))
    (handle-exceptions
     exn
     (begin
       (print-call-chain)
       (print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
       (print "exn=" (condition->list exn))
       (print "ping failed to connect to tcp://" hostport))
     (thread-start! timeout)
     (thread-start! ping)
     (thread-join! ping)
     (if success (thread-terminate! timeout)))
    (if return-socket
	(if success req #f)
	(begin
	  (nn-close req)
	  success))))

;;======================================================================
;; D A S H B O A R D   D B 
;;======================================================================

(define (mddb:open-db)
  (let* ((db (open-database (conc (get-environment-variable "HOME") "/.dashboard.db"))))
    (set-busy-handler! db (busy-timeout 10000))