send.py

来自「linux下基于c++的处理器仿真平台。具有处理器流水线」· Python 代码 · 共 244 行

#!/usr/bin/env python
import os, os.path, re, socket, sys
from os import environ as env, listdir
from os.path import basename, isdir, isfile, islink, join as joinpath, normpath
from filecmp import cmp as filecmp
from shutil import copy

def nfspath(dir):
    if dir.startswith('/.automount/'):
        dir = '/n/%s' % dir[12:]
    elif not dir.startswith('/n/'):
        dir = '/n/%s%s' % (socket.gethostname().split('.')[0], dir)
    return dir

def syncdir(srcdir, destdir):
    srcdir = normpath(srcdir)
    destdir = normpath(destdir)
    if not isdir(destdir):
        sys.exit('destination directory "%s" does not exist' % destdir)
        
    for root, dirs, files in os.walk(srcdir):
        root = normpath(root)
        prefix = os.path.commonprefix([root, srcdir])
        root = root[len(prefix):]
        if root.startswith('/'):
            root = root[1:]
        for rem in [ d for d in dirs if d.startswith('.') or d == 'SCCS']:
            dirs.remove(rem)

        for entry in dirs:
            newdir = joinpath(destdir, root, entry)
            if not isdir(newdir):
                os.mkdir(newdir)
                print 'mkdir', newdir

        for i,d in enumerate(dirs):
            if islink(joinpath(srcdir, root, d)):
                dirs[i] = joinpath(d, '.')

        for entry in files:
            dest = normpath(joinpath(destdir, root, entry))
            src = normpath(joinpath(srcdir, root, entry))
            if not isfile(dest) or not filecmp(src, dest):
                print 'copy %s %s' % (dest, src)
                copy(src, dest)

progpath = nfspath(sys.path[0])
progname = basename(sys.argv[0])
usage = """\
Usage:
    %(progname)s [-c] [-e] [-f] [-j <jobfile>] [-q queue] [-v] <regexp>
    -c           clean directory if job can be run
    -e           only echo pbs command info, don't actually send the job
    -f           force the job to run regardless of state
    -q <queue>   submit job to the named queue
    -j <jobfile> specify the jobfile (default is <basedir>/test.py)
    -v           be verbose

    %(progname)s [-j <jobfile>] -l [-v] <regexp>
    -j <jobfile> specify the jobfile (default is <basedir>/test.py)
    -l           list job names, don't submit
    -v           be verbose (list job parameters)

    %(progname)s -h
    -h           display this help
""" % locals()

try:
    import getopt
    opts, args = getopt.getopt(sys.argv[1:], '-CRcd:efhj:lq:v')
except getopt.GetoptError:
    sys.exit(usage)

clean = False
onlyecho = False
exprs = []
force = False
listonly = False
queue = ''
verbose = False
jfile = 'Base/test.py'
docpts = False
doruns = True
runflag = False

for opt,arg in opts:
    if opt == '-C':
        docpts = True
    if opt == '-R':
        runflag = True
    if opt == '-c':
        clean = True
    if opt == '-e':
        onlyecho = True
    if opt == '-f':
        force = True
    if opt == '-h':
        print usage
        sys.exit(0)
    if opt == '-j':
        jfile = arg
    if opt == '-l':
        listonly = True
    if opt == '-q':
        queue = arg
    if opt == '-v':
        verbose = True

if docpts:
    doruns = runflag

for arg in args:
    exprs.append(re.compile(arg))

import jobfile, pbs
from job import JobDir, date

conf = jobfile.JobFile(jfile)

if not listonly and not onlyecho and isdir(conf.linkdir):
    if verbose:
        print 'Checking for outdated files in Link directory'
    syncdir(conf.linkdir, conf.basedir)

jobnames = {}
joblist = []

if docpts and doruns:
    gen = conf.alljobs()
elif docpts:
    gen = conf.checkpoints()
elif doruns:
    gen = conf.jobs()

for job in gen:
    if job.name in jobnames:
        continue

    if exprs:
        for expr in exprs:
            if expr.match(job.name):
                joblist.append(job)
                break
    else:
        joblist.append(job)

if listonly:
    if verbose:
        for job in joblist:
            job.printinfo()
    else:
        for job in joblist:
            print job.name
    sys.exit(0)

if not onlyecho:
    newlist = []
    for job in joblist:
        jobdir = JobDir(joinpath(conf.rootdir, job.name))
        if jobdir.exists():
            if not force:
                status = jobdir.getstatus()
                if status == 'queued':
                    continue

                if status == 'running':
                    continue

                if status == 'success':
                    continue

            if not clean:
                sys.exit('job directory %s not clean!' % jobdir)

            jobdir.clean()
        newlist.append(job)
    joblist = newlist

class NameHack(object):
    def __init__(self, host='pbs.pool', port=24465):
        self.host = host
        self.port = port
        self.socket = None

    def setname(self, jobid, jobname):
        try:
            jobid = int(jobid)
        except ValueError:
            jobid = int(jobid.strip().split('.')[0])

        jobname = jobname.strip()
        # since pbs can handle jobnames of 15 characters or less,
        # don't use the raj hack.
        if len(jobname) <= 15:
            return

        if self.socket is None:
            import socket
            self.socket = socket.socket()
            # Connect to pbs.pool and send the jobid/jobname pair to port
            # 24465 (Raj didn't realize that there are only 64k ports and
            # setup inetd to point to port 90001)
            self.socket.connect((self.host, self.port))

        self.socket.send("%s %s\n" % (jobid, jobname))

namehack = NameHack()

for job in joblist:
    jobdir = JobDir(joinpath(conf.rootdir, job.name))
    
    if not onlyecho:
        jobdir.create()
        
    print 'Job name:       %s' % job.name
    print 'Job directory:  %s' % jobdir

    qsub = pbs.qsub()
    qsub.pbshost = 'simpool.eecs.umich.edu'
    qsub.stdout = jobdir.file('jobout')
    qsub.name = job.name[:15]
    qsub.join = True
    qsub.node_type = 'FAST'
    qsub.env['ROOTDIR'] = conf.rootdir
    qsub.env['JOBNAME'] = job.name
    if len(queue):
        qsub.queue = queue
    qsub.build(joinpath(progpath, 'job.py'))

    if verbose:
        print 'PBS Command:    %s' % qsub.command

    if not onlyecho:
        ec = qsub.do()
        if ec == 0:
            jobid = qsub.result
            print 'PBS Jobid:      %s' % jobid
            namehack.setname(jobid, job.name)
            queued = date()
            jobdir.echofile('.queued', queued)
            jobdir.setstatus('queued on %s' % queued)
        else:
            print 'PBS Failed'