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-18)
(use srfi-18 posix hostinfo)

(declare (unit dbfile))
(declare (uses debugprint))
(declare (uses commonmod))

(module dbfile
	*
(import scheme)
	
  (import scheme
	  chicken
	  	
(cond-expand
 (chicken-4
  
  (import chicken
	  data-structures
	  extras
	  matchable
  
	  (prefix sqlite3 sqlite3:)
	  posix typed-records
	  posix posix-extras typed-records

	  srfi-18
	  srfi-1
	  srfi-69
	  stack
	  files
	  ports
	  hostinfo
	  
	  commonmod
	  debugprint
	  )
  )
 (chicken-5
  (import (prefix sqlite3 sqlite3:)
	  ;; data-structures
	  ;; extras
	  ;; files
	  ;; posix
	  ;; posix-extras
	  chicken.base
	  chicken.condition
	  chicken.file
	  chicken.file.posix
	  chicken.format
	  chicken.io
	  chicken.pathname
	  chicken.port
	  chicken.process
	  chicken.process-context
	  chicken.process-context.posix
	  chicken.sort
	  chicken.string
	  chicken.time
	  chicken.time.posix
	  
	  matchable
	  md5
	  message-digest
	  pathname-expand
	  regex
	  regex-case
	  srfi-1
	  srfi-18
	  srfi-69
	  typed-records
	  stack
	  system-information

	  commonmod
	  debugprint
	  )
  (define file-write-access? file-writable?)
  (define file-move move-file)
  ))

  
;; parameters
;;
(define dbfile:testsuite-name (make-parameter #f))

(define keep-age-param        (make-parameter 10))      ;; qif file age, if over move to attic
(define num-run-dbs           (make-parameter 10))      ;; number of db's in .mtdb
(define dbfile:sync-method    (make-parameter 'attach)) ;; 'attach or 'original

;; used in simple-get-runs (thanks Brandon!)
(define-record simple-run target id runname state status owner event_time)
(define-record-printer (simple-run x out)
  (fprintf out "#,(simple-run ~S ~S ~S ~S)"
	   (simple-run-target x) (simple-run-id x) (simple-run-runname x) (time->string (seconds->local-time (simple-run-event_time x) ))))

;; args is hash table of string to value
;;
(define (get-purpose args)
  (let* ((get-arg    (lambda (key)
		       (hash-table-ref/default args key #f)))
	 (get-switch (lambda keys
		       (fold
			(lambda (key res)
			  (if (hash-table-ref/default args key #f)
			      (or key res)
			      res))
			#f
			keys)))
	 (action     (get-switch "-server" "-execute" "-run" "-rerun")))
    (cond
     (action
      (substring action 1 (string-length action)))
     (else
      "nopurpose"))))

;; megatest process tracking

(defstruct procinf
  (start      (current-seconds))
  (end        -1)
  (host       (get-host-name)) ;; why is this not being recognised?
  (pid        (current-process-id))
  (port       -1)
  (cwd        (current-directory))
  (load       #f)
  (purpose    #f) ;; get-purpose needed to be run in megatest.scm
  (dbname     #f)
  (mtbin      (car (argv)))
  (mtversion  #f)
  (status     "running")
  
  
  )

(define *procinf* (make-procinf))
(define *dbstruct-dbs* #f)
(define *db-open-mutex* (make-mutex))
(define *db-access-mutex* (make-mutex)) ;; used in common.scm
(define *no-sync-db*   #f)
(define *db-sync-in-progress* #f)
(define *db-with-db-mutex*    (make-mutex))
(define *max-api-process-requests* 0)

           #t
          )
          #f
  )
)

(define (dbfile:make-tmpdir-name areapath tmpadj)
  (let* ((area (pathname-file areapath))
  (let* ((dname (conc "/tmp/"(current-user-name)"/" (string-translate areapath "/" ".") tmpadj)))
         (dname (conc "/tmp/"(current-user-name)"/megatest_localdb/" area "/" (string-translate areapath "/" ".") tmpadj "/.mtdb")))
    (unless (directory-exists? dname)
      (create-directory dname #t))
    dname))

(define (dbfile:run-id->path apath run-id)
  (conc apath"/"(dbfile:run-id->dbname run-id)))

;; just the filename
(define (dbfile:run-id->dbfname run-id)
  (conc (dbfile:run-id->dbnum run-id)".db"))

;; the path in MTRAH with the filename
(define (dbfile:run-id->dbname run-id)
  (conc ".mtdb/"(dbfile:run-id->dbfname run-id)))
  (conc (dbfile:run-id->dbfname run-id)))


;; Make the dbstruct, setup up auxillary db's and call for main db at least once
;;
;; called in http-transport and replicated in rmt.scm for *local* access. 
;;
(define (dbfile:setup do-sync areapath tmppath)
(define (dbfile:setup areapath tmppath)
  (cond
   (*dbstruct-dbs*
    (dbfile:print-err "WARNING: dbfile:setup called when *dbstruct-dbs* is already initialized")
    *dbstruct-dbs*) ;; TODO: when multiple areas are supported, this optimization will be a hazard
   (else
    (let* ((dbstruct (make-dbr:dbstruct areapath: areapath tmppath: tmppath)))
      (set! *dbstruct-dbs* dbstruct)

	  (dbfile:open-db dbstruct run-id init-proc))
	(let* ((dbdat (dbfile:get-dbdat dbstruct run-id)))
	  (if dbdat
	      dbdat
	      (let* ((tmppath   (dbr:dbstruct-tmppath  dbstruct))
		     (tmpdbpath (dbfile:run-id->path tmppath run-id))
		     (dbdat     (dbfile:open-sqlite3-db tmpdbpath init-proc sync-mode: 0 journal-mode: "WAL")))
		
		;; the following line short-circuits the "one db handle per thread" model
               ;; the following line short-circuits the "one db handle per thread" model
		;; 
		;; (dbfile:add-dbdat dbstruct run-id dbdat)
		;;
		dbdat))))))
    
;; COMBINE dbfile:open-sqlite-db and dbfile:lock-create-open
;;

    
(define (dbfile:print-err . params)
  (with-output-to-port
      (current-error-port)
    (lambda ()
      (apply print params))))

(define (dbfile:cautious-open-database fname init-proc sync-mode journal-mode #!optional (tries-left 500))
(define (dbfile:cautious-open-database fname init-proc sync-mode journal-mode
				       #!key (tries-left 500)(force-init #f))
  (let* ((busy-file  (conc fname "-journal"))
	 (delay-time (* (- 51 tries-left) 1.1))
      	 (write-access (file-write-access? fname))
         (dir-access (file-write-access? (pathname-directory fname)))
         (retry      (lambda ()
		       (thread-sleep! delay-time)
		       (if (> tries-left 0)
			   (dbfile:cautious-open-database fname init-proc
							  sync-mode journal-mode
							  (- tries-left 1))))))
							  tries-left: (- tries-left 1))))))
    (assert (>= tries-left 0) (conc "FATAL: too many attempts in dbfile:cautious-open-database of "fname", giving up."))
    
    (if (and (file-write-access? fname)
	     (file-exists? busy-file))
	(begin
	  (if (common:low-noise-print 120 busy-file)
	      (dbfile:print-err "INFO: dbfile:cautious-open-database: journal file "
				busy-file" exists, trying again in few seconds."))
	  (thread-sleep! 1)
	  (if (eq? tries-left 2)
	      (begin
	  	(dbfile:print-err "INFO: forcing journal rollup "busy-file)
	  	(dbfile:brute-force-salvage-db fname)))
	  (dbfile:cautious-open-database fname init-proc sync-mode journal-mode (- tries-left 1)))
	  (dbfile:cautious-open-database fname init-proc sync-mode journal-mode tries-left: (- tries-left 1)))
	
	(let* ((result (condition-case
		        (if dir-access
			    (dbfile:with-simple-file-lock
			     (conc fname ".lock")
			     (lambda ()
			       (let* ((db-exists (file-exists? fname))
				      (db        (sqlite3:open-database fname))) ;; creates an empty db if it did not already exist.
                                 (sqlite3:set-busy-handler! db (sqlite3:make-busy-timeout 30000))
				 (if sync-mode
				     (sqlite3:execute db (conc "PRAGMA synchronous = "sync-mode";")))
				 (if journal-mode
				     (sqlite3:execute db (conc "PRAGMA journal_mode = "journal-mode";")))
				 (if (and init-proc (not db-exists))
				 (if (and init-proc (or force-init
							(not db-exists)))
				     (init-proc db))
				 db)))
				 db))
			     expire-time: 30)
                            (begin
			      (if (file-exists? fname )
                                  (let ((db (sqlite3:open-database fname)))
				    ;; pragmas synchronous not needed because this db is used read-only
				    ;; (sqlite3:execute db (conc "PRAGMA synchronous = "mode";")
				    (sqlite3:set-busy-handler! db (sqlite3:make-busy-timeout 30000)) ;; read-only but still need timeout
				    db )
                                  (print "file doesn't exist: " fname))))
                                  (print "cautious-open-database: file doesn't exist: " fname))))
			(exn (io-error)
			     (dbfile:print-err exn "ERROR: i/o error with " fname ". Check permissions, disk space etc. and try again.")
			     (retry))
			(exn (corrupt)
			     (dbfile:print-err exn "ERROR: database " fname " is corrupt. Repair it to proceed.")
			     (retry))
			(exn (busy)

;; opens and returns handle and nothing else
;;
;; NOTE: this is already protected by mutex *no-sync-db-mutex*
;;
(define (dbfile:raw-open-no-sync-db dbpath)
  (if (not (file-exists? dbpath))
      (create-directory dbpath #t))
  (debug:print-info 0 *default-log-port* "Opening "dbpath"/no-sync.db")
  (debug:print-info 2 *default-log-port* "(dbfile:raw-open-no-sync-db: Opening "dbpath"/no-sync.db")
  (let* ((dbname    (conc dbpath "/no-sync.db"))
	 (db-exists (file-exists? dbname))
	 (init-proc (lambda (db)
		      (sqlite3:with-transaction
		       db
		       (lambda ()
			 ;; I have been having trouble with init of no-sync.db so
			 ;; doing the init in a transaction every time (no gating
			 ;; on file existance.
			  (for-each
			   (lambda (stmt)
			     (sqlite3:execute db stmt))
			   (list
			    "CREATE TABLE IF NOT EXISTS no_sync_metadat
                                (var TEXT,
                                 val TEXT,
                                   CONSTRAINT no_sync_metadat_constraint UNIQUE (var));"
			    "CREATE TABLE IF NOT EXISTS no_sync_locks 
                                (key TEXT,
                                 val TEXT,
                                   CONSTRAINT no_sync_metadat_constraint UNIQUE (key));"))))))
                                   CONSTRAINT no_sync_metadat_constraint UNIQUE (key));"
			    "CREATE TABLE IF NOT EXISTS processes
                                (id INTEGER PRIMARY KEY,
                                 host TEXT,
                                 port INTEGER,
                                 pid INTEGER,
                                 starttime INTEGER,
                                 endtime INTEGER,
                                 status TEXT,
                                 purpose TEXT,
                                 dbname TEXT,
                                 mtversion TEXT,
                                 reason TEXT DEFAULT 'none',
                                   CONSTRAINT no_sync_processes UNIQUE (host,pid));"
			    ))))))
	 (on-tmp      (equal? (car (string-split dbpath "/")) "tmp"))
	 (db        (if on-tmp
			(dbfile:cautious-open-database dbname init-proc 0 "WAL")
			(dbfile:cautious-open-database dbname init-proc 0 #f)
			(dbfile:cautious-open-database dbname init-proc 1 "WAL" force-init: #t) ;; WAL MODE should use syncronous=1
			;; (dbfile:cautious-open-database dbname init-proc 0 #f    force-init: #t)
			(dbfile:cautious-open-database dbname init-proc 0 "MEMORY" force-init: #t) ;; Journal mode = memory is fastest?
			;; (sqlite3:open-database dbname)
			)))
    (if on-tmp	      ;; done in cautious-open-database
	(begin
	  (sqlite3:execute db "PRAGMA synchronous = 0;")
	  (sqlite3:set-busy-handler! db (sqlite3:make-busy-timeout 136000))))
    ;; (if on-tmp	      ;; done in cautious-open-database
    ;;    (begin
    ;; (sqlite3:execute db "PRAGMA synchronous = 0;") ;; why was this here when is is handled by cautious-open-database?
    (sqlite3:set-busy-handler! db (sqlite3:make-busy-timeout 136000)) ;; ))
    db))

;; mtest processes registry calls

(define (dbfile:insert-or-update-process nsdb dat)
  (let* ((host      (procinf-host dat))
	 (pid       (procinf-pid  dat))
	 (curr-info (dbfile:get-process-info nsdb host pid)))
    (if curr-info ;; record exists, do update
	(match curr-info
	  ((host port pid starttime endtime status purpose dbname mtversion)
	   (sqlite3:execute
	    nsdb
	    "UPDATE processes SET port=?,starttime=?,endtime=?,status=?,
                                  purpose=?,dbname=?,mtversion=?
              WHERE host=? AND pid=?;"
	    (or (procinf-port      dat) port)
	    (or (procinf-start     dat) starttime)
	    (or (procinf-end       dat) endtime)
	    (or (procinf-status    dat) status)
	    (or (procinf-purpose   dat) purpose)
	    (or (procinf-dbname    dat) dbname)
	    (or (procinf-mtversion dat) mtversion)
	    host pid))
	  (else
	   #f ;; what to do?
	   ))
	(dbfile:register-process
	 nsdb
	 (procinf-host      dat)
	 (procinf-port      dat)
	 (procinf-pid       dat)
	 (procinf-start     dat)
	 (procinf-end       dat)
	 (procinf-status    dat)
	 (procinf-purpose   dat)
	 (procinf-dbname    dat)
	 (procinf-mtversion dat)))))
	  

(define (dbfile:register-process nsdb host port pid starttime endtime status purpose dbname mtversion)
  (sqlite3:execute nsdb "INSERT INTO processes (host,port,pid,starttime,endtime,status,purpose,dbname,mtversion) VALUES (?,?,?,?,?,?,?,?,?);"
		   host port pid starttime endtime status purpose dbname mtversion))

(define (dbfile:set-process-status nsdb host pid newstatus)
  (sqlite3:execute nsdb "UPDATE processes SET status=? WHERE host=? AND pid=?;" newstatus host pid))

;; as sorted should be stable. can use to choose "winner"
;;
(define (dbfile:get-process-options nsdb purpose dbname)
  (sqlite3:fold-row
   ;; host port pid starttime status mtversion
   (lambda (res . row)
     (cons row res))
   '()
   nsdb
   "SELECT host,port,pid,starttime,endtime,status,mtversion FROM processes WHERE purpose=? AND dbname LIKE ? AND status IN ('running','alive') ORDER BY starttime ASC,host,port;"
   purpose dbname))

(define (dbfile:get-process-info nsdb host pid)
  (let ((res (sqlite3:fold-row
	      ;; host port pid starttime status mtversion
	      (lambda (res . row)
		(cons row res))
	      '()
	      nsdb
	      "SELECT host,port,pid,starttime,endtime,status,purpose,dbname,mtversion FROM processes WHERE host=? AND pid=?;"
	      host pid)))
    (if (null? res)
	#f
	(car res))))

(define (dbfile:row->procinf row)
  (match row
   ((host port pid starttime endtime status mtversion)
    (make-procinf host: host port: port pid: pid starttime: starttime endtime: endtime status: status mtversion: mtversion))
   (else
    (debug:print 0 *default-log-port* "ERROR: row "row" did not match host,port,pid,starttime,endtime,status,mtversion")
    #f)))

(define (dbfile:set-process-done nsdb host pid reason)
  (sqlite3:execute nsdb "UPDATE processes SET status='done',endtime=?,reason=? WHERE host=? AND pid=?;" (current-seconds) reason host pid)
  (dbfile:cleanup-old-entries nsdb))

(define (dbfile:cleanup-old-entries nsdb)
  (sqlite3:execute nsdb "DELETE FROM process WHERE status='done' AND endtime<?;" (- (current-seconds) (* 3600 48))))

;; other no-sync functions

(define (dbfile:with-no-sync-db dbpath proc)
  (mutex-lock! *no-sync-db-mutex*)
  (let* ((already-open *no-sync-db*)
	 (db  (or already-open (dbfile:raw-open-no-sync-db dbpath)))
	 (res (proc db)))
    (if (not already-open)

;; transaction protected lock aquisition
;; either:
;;    fails    returns  (#f lock-creation-time identifier)
;;    succeeds (returns (#t lock-creation-time identifier)
;; use (db:no-sync-del! db keyname) to release the lock
;;
(define (db:no-sync-get-lock-with-id db keyname identifier)
  (debug:print 2 *default-log-port* "db:no-sync-get-lock-with-id: db: " db " keyname: " keyname " identifier: " identifier)
  (sqlite3:with-transaction
   db
   (lambda ()
     (condition-case
      (let* ((curr-val (db:no-sync-get/default db keyname #f)))
        (debug:print 2 *default-log-port* "db:no-sync-get-lock-with-id: curr-val: " curr-val)
	(if curr-val
	    (match (db:extract-time-identifier curr-val) ;; result->timestamp, identifier
	       ((timestamp . ident)
		(cons (equal? ident identifier) timestamp))
	       (else 
                (debug:print 2 *default-log-port* "db:no-sync-get-lock-with-id: malformed lock")
	       (else (cons #f 'malformed-lock)))  ;; lock malformed
                (cons #f 'malformed-lock)
                )
            )  ;; lock malformed
	    (let ((curr-sec (current-seconds))
		  (lock-value (if identifier
				  (conc (current-seconds)"+"identifier)
				  (current-seconds))))
	      (sqlite3:execute db "INSERT OR REPLACE INTO no_sync_metadat (var,val) VALUES(?,?);" keyname lock-value)
	      (cons #t curr-sec))))
      (exn (io-error)  (dbfile:print-err "ERROR: i/o error with no-sync db. Check permissions, disk space etc. and try again."))

	  (if (file-exists? fname)
	      (handle-exceptions exn
                #f 
                (with-input-from-file fname
	  	  (lambda ()
		    (equal? key-string (read-line)))))
              (begin
                 (dbfile:print-err "dbfile:simple-file-lock created " fname " but it was gone 3 seconds later")
                (dbfile:print-err "dbfile:simple-file-lock created " fname " but it was gone 0.25 seconds later")
	         #f
              )
          )
       )
    )
  )
)

	 (result      (or stmth
			  (let* ((newstmth (sqlite3:prepare db stmt)))
			    ;; (db:hoh-set! stmt-cache db stmt newstmth)
			    (hash-table-set! stmt-cache stmt newstmth)
			    newstmth))))
    (mutex-unlock! *get-cache-stmth-mutex*)
    result))

;; (define *mutex-stmth-call* (make-mutex))
;; 
;; (define (db:with-mutex-for-stmth proc)
;;   (mutex-lock! *mutex-stmth-call*)
;;   (let* ((res (proc)))
;;     (mutex-unlock! *mutex-stmth-call*)
;;     res))
)

)