view commons/core/sql/TableJobAdaptator.py @ 6:769e306b7933

Change the repository level.

# Copyright INRA (Institut National de la Recherche Agronomique)
# http://www.inra.fr
# http://urgi.versailles.inra.fr
#
# This software is governed by the CeCILL license under French law and
# abiding by the rules of distribution of free software.  You can  use, 
# modify and/ or redistribute the software under the terms of the CeCILL
# license as circulated by CEA, CNRS and INRIA at the following URL
# "http://www.cecill.info". 
#
# As a counterpart to the access to the source code and  rights to copy,
# modify and redistribute granted by the license, users are provided only
# with a limited warranty  and the software's author,  the holder of the
# economic rights,  and the successive licensors  have only  limited
# liability. 
#
# In this respect, the user's attention is drawn to the risks associated
# with loading,  using,  modifying and/or developing or reproducing the
# software by the user in light of its specific status of free software,
# that may mean  that it is complicated to manipulate,  and  that  also
# therefore means  that it is reserved for developers  and  experienced
# professionals having in-depth computer knowledge. Users are therefore
# encouraged to load and test the software's suitability as regards their
# requirements in conditions enabling the security of their systems and/or 
# data to be ensured and,  more generally, to use and operate it in the 
# same conditions as regards security. 
#
# The fact that you are presently reading this means that you have had
# knowledge of the CeCILL license and that you accept its terms.


import os
import time
import datetime
import sys
from commons.core.sql.Job import Job 
from commons.core.sql.TableAdaptator import TableAdaptator

## Methods for Job persistence 
#
class TableJobAdaptator(TableAdaptator):
        
    ## Record a job
    #
    # @param job Job instance with the job informations
    #
    def recordJob(self, job):
        self.removeJob(job)
        sqlCmd = "INSERT INTO %s" % self._table
        sqlCmd += " VALUES ("
        sqlCmd += " \"%s\"," % job.jobid
        sqlCmd += " \"%s\"," % job.jobname
        sqlCmd += " \"%s\"," % job.groupid
        sqlCmd += " \"%s\"," % job.launcher
        sqlCmd += " \"%s\"," % job.queue
        sqlCmd += " \"%s\"," % job.lResources
        sqlCmd += " \"waiting\","
        sqlCmd += " \"%s\"," % time.strftime("%Y-%m-%d %H:%M:%S")
        sqlCmd += " \"?\" );"
        self._iDb.execute(sqlCmd)
        
       
    ## Remove a job from the job table
    #
    #  @param job: job instance to remove
    #
    def removeJob(self, job):
        qry = "DELETE FROM %s" % self._table
        qry += " WHERE groupid='%s'" % job.groupid
        qry += " AND jobname='%s'" % job.jobname
        qry += " AND launcher='%s';" % job.launcher
        self._iDb.execute(qry)
            
            
    ## Set the jobid of a job with the id of SGE
    #
    # @param job job instance
    # @param jobid integer
    #
    def updateJobIdInDB(self, job, jobid):
        #TODO: check if only one job will be updated
        qry = "UPDATE %s" % self._table
        qry += " SET jobid='%i'" % int(jobid)
        qry += " WHERE jobname='%s'" % job.jobname
        qry += " AND groupid='%s'" % job.groupid
        qry += " AND launcher='%s';" % job.launcher
        self._iDb.execute(qry)
        
        
    ## Get a job status
    #
    # @param job: a Job instance with the job informations
    #
    def getJobStatus(self, job):
        if job.jobid != 0 and job.jobname == "":
            job.jobname = job.jobid
            job.jobid = 0
        qry = "SELECT status FROM %s" % self._table
        qry += " WHERE groupid='%s'" % job.groupid
        qry += " AND jobname='%s'" % job.jobname
        qry += " AND launcher='%s';" % job.launcher
        self._iDb.execute(qry)
        res = self._iDb.fetchall()
        if len(res) > 1:
            sys.stderr.write("ERROR while getting job status: non-unique jobs\n")
            sys.stderr.flush()
            sys.exit(1)
        if res == None or len(res) == 0:
            return "unknown"
        return res[0][0]
    
    
    ## Change a job status
    #
    # @param job: a Job instance with the job informations
    # @param status: the new status (waiting,finished,error)
    #
    def changeJobStatus(self, job, status):
        sqlCmd = "UPDATE %s" % self._table
        sqlCmd += " SET status='%s'" % status
        sqlCmd += ", node='%s'" % job.node
        sqlCmd += " WHERE groupid='%s'" % job.groupid
        sqlCmd += " AND jobname='%s'" % job.jobname
        sqlCmd += " AND launcher='%s';" % job.launcher
        self._iDb.execute(sqlCmd)
        
        
    ## Get the number of jobs belonging to the desired groupid with the desired status.
    #
    # @param groupid string a group identifier to record related job series 
    # @param status string job status (waiting, running, finished, error)
    # @return int
    #
    def getCountStatus(self, groupid, status):
        qry = "SELECT count(jobname) FROM %s" % self._table
        qry += " WHERE groupid='%s'" % groupid
        qry += " AND status='%s';" % status
        self._iDb.execute(qry)
        res = self._iDb.fetchall()
        return int(res[0][0])
        
        
    ## Clean all job from a job group
    #
    # @param groupid: a group identifier to record related job series
    #
    def cleanJobGroup(self, groupid):
        qry = "DELETE FROM %s WHERE groupid='%s';" % (self._table, groupid)
        self._iDb.execute(qry)
            
            
    ## Check if there is unfinished job from a job group.
    #
    # @param groupid string a group identifier to record related job series 
    #        
    def hasUnfinishedJob(self, groupid):
        qry = "SELECT * FROM %s" % self._table
        qry += " WHERE groupid='%s'" % groupid
        qry += " and status!='finished';" 
        self._iDb.execute(qry)
        res = self._iDb.fetchall()
        if len(res) == 0:
            return False
        return True
    

    ## Wait job finished status from a job group.
    #  Job are re-launched if error (max. 3 times)
    #
    # @param groupid string a group identifier to record related job series
    # @param checkInterval integer time laps in seconds between two checks (default = 5)
    # @param maxRelaunch integer max nb of times a job in error is relaunch before exiting (default = 3)
    # @param exitIfTooManyErrors boolean exit if a job is still in error above maxRelaunch (default = True)
    # @param timeOutPerJob integer max nb of seconds after which one tests if a job is still in SGE or not (default = 60*60=1h)
    #
    def waitJobGroup(self, groupid, checkInterval=5, maxRelaunch=3, exitIfTooManyErrors=True, timeOutPerJob=60*60):
        dJob2Err = {}
        
        # retrieve the total number of jobs belonging to the desired groupid
        qry = "SELECT count(jobname) FROM %s WHERE groupid='%s';" % (self._table, groupid)
        self._iDb.execute(qry)
        totalNbJobs = int(self._iDb.fetchall()[0][0])
        
        nbTimeOuts = 0
        
        while True:
            time.sleep(checkInterval)
            # retrieve the finished jobs and stop if all jobs are finished
            nbFinishedJobs = self.getCountStatus(groupid, "finished")
            if nbFinishedJobs == totalNbJobs:
                break

            # retrieve the jobs in error and relaunch them if they are in error (max. 'maxRelaunch' times)
            qry = "SELECT * FROM %s" % self._table
            qry += " WHERE groupid='%s'" % groupid
            qry += " AND status ='error';"
            self._iDb.execute(qry)
            lJobsInError = self._iDb.fetchall()
            for job in lJobsInError:
                jobName = job[1]
                if not dJob2Err.has_key(jobName):
                    dJob2Err[jobName] = 1
                if dJob2Err[jobName] < maxRelaunch:
                    print "job '%s' in error, re-submitting (%i)" % (job[1], dJob2Err[job[1]])
                    sys.stdout.flush()
                    lResources = job[5].replace("[", "").replace("]", "").replace("'", "").split(", ")
                    newJob = Job(jobname=jobName, groupid=job[2], launcherFile=job[3], queue=job[4], lResources=lResources)
                    self.submitJob(newJob)
                    dJob2Err[jobName] += 1
                else:
                    dJob2Err[jobName] += 1
                    cmd = "job '%s' in permanent error (>%i)" % (jobName, maxRelaunch)
                    cmd += "\ngroupid = %s" % groupid
                    cmd += "\nnb of jobs = %i" % totalNbJobs
                    cmd += "\nnb of finished jobs = %i" % self.getCountStatus(groupid, "finished")
                    cmd += "\nnb of waiting jobs = %i" % self.getCountStatus(groupid, "waiting")
                    cmd += "\nnb of running jobs = %i" % self.getCountStatus(groupid, "running")
                    cmd += "\nnb of jobs in error = %i" % self.getCountStatus(groupid, "error")
                    sys.stdout.flush()
                    if exitIfTooManyErrors:
                        self.cleanJobGroup(groupid)
                        sys.exit(1)
                    else:
                        checkInterval = 60
            nbTimeOuts = self._checkIfJobsTableAndJobsManagerInfoAreConsistent(nbTimeOuts, timeOutPerJob, groupid)
    
    
    ## Submit a job to a queue and record it in job table.
    #
    # @param job a job instance
    # @param maxNbWaitingJobs integer max nb of waiting jobs before submitting a new one (default = 10000)
    # @param checkInterval integer time laps in seconds between two checks (default = 30)
    # @param verbose integer (default = 0)
    #               
    def submitJob(self, job, verbose=0, maxNbWaitingJobs=10000, checkInterval=30):
        if self.getJobStatus(job) in ["waiting", "running", "finished"]:
            sys.stderr.write( "WARNING: job '%s' was already submitted\n" % job.jobname)
            sys.stderr.flush()
            self.cleanJobGroup(job.groupid)
            sys.exit(1)
            
        while self.getCountStatus(job.groupid, "waiting") > maxNbWaitingJobs:
            time.sleep(checkInterval)

        self.recordJob(job)
        cmd = self._getQsubCommand(job)
        returnStatus = os.system(cmd)

        if returnStatus == 0:
            fileName = "jobid.stdout"
            jobidFileHandler = open(fileName, "r")
            jobid = self._getJobidFromJobManager(jobidFileHandler)
            if verbose > 0:
                print "job '%i %s' submitted" % (jobid, job.jobname)
                sys.stdout.flush()
            job.jobid = jobid
            jobidFileHandler.close()
            self.updateJobIdInDB(job, jobid)
            os.remove(fileName)
        return returnStatus


    ## Get the list of nodes where jobs of one group were executed
    #
    # @param groupid string a group identifier of job series 
    # @return lNodes list of nodes names without redundancy
    #
    def getNodesListByGroupId(self, groupId):
        qry = "SELECT DISTINCT node FROM %s" % self._table
        qry += " WHERE groupid='%s'" % groupId
        self._iDb.execute(qry)
        res = self._iDb.fetchall()
        lNodes = []
        for resTuple in res:
            lNodes.append(resTuple[0])
        return lNodes
    
    def checkJobTable(self):
        if not self._iDb.doesTableExist(self._table):
            self._iDb.createTable(self._table, "jobs")
        else:
            lExpFields = sorted(["jobid", "jobname", "groupid", "launcher", "queue", "resources", "status", "time", "node"])
            lObsFields = sorted(self._iDb.getFieldList(self._table))
            if lExpFields != lObsFields:
                self._iDb.createTable(self._table, "jobs", overwrite = True)
    
    def close(self):
        self._iDb.close() 
    
    def _getJobidAndNbJob(self, jobid) :
        tab = jobid.split(".")
        jobid = tab[0]
        tab = tab[1].split(":")
        nbJob = tab[0]
        return jobid, nbJob
    
class TableJobAdaptatorSGE(TableJobAdaptator):
        
    def _checkIfJobsTableAndJobsManagerInfoAreConsistent(self, nbTimeOuts, timeOutPerJob, groupid):
        # retrieve the date and time at which the oldest, still-running job was submitted
        sql = "SELECT jobid,jobname,time FROM %s WHERE groupid='%s' AND status='running' ORDER BY time DESC LIMIT 1" % (self._table, groupid)
        self._iDb.execute( sql )
        res = self._iDb.fetchall()
        if len(res) > 0:
            jobid = res[0][0]
            jobname = res[0][1]
            dateTimeOldestJob = res[0][2]
            dateTimeCurrent = datetime.datetime.now()
            # delta is time between (i) first job launched of the given groupid and still in running state and (ii) current time 
            delta = dateTimeCurrent - dateTimeOldestJob
            # check if delta is in an interval:  0 <= delta < 1h | 1h <= delta < 2h | 2h <= delta < 3h (timeOutPerJob = 1h)  
            if delta.seconds >= nbTimeOuts * timeOutPerJob and delta.seconds < (nbTimeOuts+1) * timeOutPerJob:
                return nbTimeOuts
            # delta outside the interval: go to next interval (time out) 
            if delta.seconds >= (nbTimeOuts+1) * timeOutPerJob:
                nbTimeOuts += 1
                # Job with 'running' status should be in qstat. Because status in DB is set at 'running' by the job launched.
                if not self.isJobStillHandledBySge(jobid, jobname):
                    # But if not, let time for the status update (in DB), if the job finished between the query execution and now.
                    time.sleep( 5 )
                # If no update at 'finished', exit
                #TODO: check status in DB
                if not self.isJobStillHandledBySge(jobid, jobname):
                    msg = "ERROR: job '%s', supposedly still running, is not handled by SGE anymore" % ( jobid )
                    msg += "\nit was launched the %s (> %.2f hours ago)" % ( dateTimeOldestJob, timeOutPerJob/3600.0 )
                    msg += "\nthis problem can be due to:"
                    msg += "\n* memory shortage, in that case, decrease the size of your jobs;"
                    msg += "\n* timeout, in that case, decrease the size of your jobs;"
                    msg += "\n* node failure or database error, in that case, launch the program again or ask your system administrator."
                    sys.stderr.write("%s\n" % msg)
                    sys.stderr.flush()
                    self.cleanJobGroup(groupid)
                    sys.exit(1)
        return nbTimeOuts
                        
    ## Check if a job is still handled by SGE
    #
    # @param jobid string job identifier
    # @param jobname string job name
    #  
    def isJobStillHandledBySge(self, jobid, jobname):
        isJobInQstat = False
        qstatFile = "qstat_stdout"
        cmd = "qstat > %s" % qstatFile
        returnStatus = os.system(cmd)
        if returnStatus != 0:
            msg = "ERROR while launching 'qstat'"
            sys.stderr.write( "%s\n" % msg )
            sys.exit(1)
        qstatFileHandler = open(qstatFile, "r")
        lLines = qstatFileHandler.readlines()
        for line in lLines:
            tokens = line.split()
            if len(tokens) > 3 and tokens[0] == str(jobid) and tokens[2] == jobname[0:len(tokens[2])]:
                isJobInQstat = True
                break
        qstatFileHandler.close()
        os.remove(qstatFile)
        return isJobInQstat
    
    def _getQsubCommand(self, job):    
        cmd = "echo '%s' | " % job.launcher
        prg = "qsub"
        cmd += prg
        cmd += " -V"
        cmd += " -N %s" % job.jobname
        if job.queue != "":
            cmd += " -q %s" % job.queue
        cmd += " -cwd"
        if job.lResources != []:
            cmd += " -l \""
            cmd += " ".join(job.lResources)
            cmd += "\""
        if job.parallelEnvironment != "":
            cmd += " -pe " + job.parallelEnvironment
        cmd += " > jobid.stdout"
        return cmd
    
    def _getJobidFromJobManager(self, jobidFileHandler):
        return int(jobidFileHandler.readline().split(" ")[2])
    

class TableJobAdaptatorTorque(TableJobAdaptator):  
                        
    def _checkIfJobsTableAndJobsManagerInfoAreConsistent(self, nbTimeOuts, timeOutPerJob, groupid):
        return nbTimeOuts
        
    def _getQsubCommand(self, job):    
        cmd = "echo '%s' | " % job.launcher
        prg = "qsub"
        cmd += prg
        cmd += " -V"
        cmd += " -d %s" % os.getcwd()
        cmd += " -N %s" % job.jobname
        if job.queue != "":
            cmd += " -q %s" % job.queue
        if job.lResources != []:
            cmd += " -l \""
            cmd += " ".join(job.lResources).replace("mem_free","mem")
            cmd += "\""
        cmd += " > jobid.stdout"
        return cmd

    def _getJobidFromJobManager(self, jobidFileHandler):
        return int(jobidFileHandler.readline().split(".")[0])