def sig_handler(signum, frame):
# for some reason, I (rmb) was unable to handle TSTP and CONT in the same way
global manSocket
if signum == SIGINT:
raise mpdrunInterrupted, 'SIGINT'
elif signum == SIGTSTP:
raise mpdrunInterrupted, 'SIGTSTP'
elif signum == SIGCONT:
if manSocket:
msgToSend = {'cmd': 'signal', 'signo': 'SIGCONT'}
mpd_send_one_msg(manSocket, msgToSend)
elif signum == SIGALRM:
raise mpdrunInterrupted, 'SIGALRM'
def get_vals_for_attach():
global nprocs, pgm, pgmArgs, mship, rship, argsFilename, delArgsFile, \
try0Locally, lineLabels, jobAlias, mergingOutput, conSocket
global stdinGoesToWho, myExitStatus, manSocket, jobid, username, cwd, totalview
global outXmlDoc, outXmlEC, outXmlFile, linesPerRank, gdb, gdbAttachJobid
global execs, users, cwds, paths, args, envvars, limits, hosts, hostList
global singinitPID, singinitPORT, doingBNR, myHost, myIP
sjobid = gdbAttachJobid.split('@') # jobnum and originating host
msgToSend = {'cmd': 'mpdlistjobs'}
mpd_send_one_msg(conSocket, msgToSend)
msg = recv_one_msg_with_timeout(conSocket, 5)
if not msg:
mpd_raise('no msg recvd from mpd before timeout')
if msg['cmd'] != 'local_mpdid': # get full id of local mpd for filters later
mpd_raise(
'did not recv local_mpdid msg from local mpd; instead, recvd: %s' %
msg)
else:
if len(sjobid) == 1:
sjobid.append(msg['id'])
got_info = 0
while 1:
msg = mpd_recv_one_msg(conSocket)
if not msg.has_key('cmd'):
print 'mpdlistjobs: INVALID msg=:%s:' % (msg)
exit(-1)
if msg['cmd'] == 'mpdlistjobs_info':
got_info = 1
smjobid = msg['jobid'].split(
' ') # jobnum, mpdid, and alias (if present)
if sjobid[0] == smjobid[0] and sjobid[1] == smjobid[
1]: # jobnum and mpdid
rank = int(msg['rank'])
users[(rank, rank)] = msg['username']
hosts[(rank, rank)] = msg['host']
execs[(rank, rank)] = msg['pgm']
cwds[(rank, rank)] = cwd
paths[(rank, rank)] = environ['PATH']
args[(rank, rank)] = [msg['clipid']]
envvars[(rank, rank)] = {}
limits[(rank, rank)] = {}
elif msg['cmd'] == 'mpdlistjobs_trailer':
if not got_info:
print 'no info on this jobid; probably invalid'
exit(-1)
break
else:
print 'invaild msg from mpd :%s:' % (msg)
exit(-1)
nprocs = len(execs.keys()) # all dicts are the same len here
elif argv[i] == '-g':
single_or_group = 'g'
else:
print '** unrecognized arg: %s' % (argv[i])
usage()
i += 1
msgToSend = {
'cmd': 'mpdsigjob',
'sigtype': sigtype,
'jobnum': jobnum,
'mpdid': mpdid,
'jobalias': jobalias,
's_or_g': single_or_group,
'username': username
}
mpd_send_one_msg(conSocket, msgToSend)
msg = recv_one_msg_with_timeout(conSocket, 5)
if not msg:
mpd_raise('no msg recvd from mpd before timeout')
if msg['cmd'] != 'mpdsigjob_ack':
if msg['cmd'] == 'already_have_a_console':
print 'mpd already has a console (e.g. for long ringtest); try later'
else:
print 'unexpected message from mpd: %s' % (msg)
exit(-1)
if not msg['handled']:
print 'job not found'
exit(-1)
conSocket.close()
def mpdboot():
global myHost, fullDirName, topMPDBoot, user
mpd_set_my_id('mpdboot_rank_notset')
fullDirName = path.abspath(path.split(argv[0])[0])
rshCmd = 'ssh'
user = mpd_get_my_username()
mpdCmd = path.join(fullDirName, 'mpd.py')
mpdbootCmd = path.join(fullDirName, 'mpdboot.py')
hostsFilename = 'mpd.hosts'
totalNum = 1 # may get chgd below
debug = 0
verbosity = 0
localConsoleArg = ''
remoteConsoleArg = ''
myConsoleVal = ''
oneMPDPerHost = 1
entryHost = ''
entryPort = ''
topMPDBoot = 1
myHost = gethostname()
myNcpus = 1
myIfhn = ''
try:
shell = path.split(environ['SHELL'])[-1]
except:
shell = 'csh'
argidx = 1 # skip arg 0
while argidx < len(argv):
if argv[argidx] == '-h' or argv[argidx] == '--help':
usage()
elif argv[argidx] == '-zentry': # entry host and port
if ':' not in argv[argidx + 1]:
print 'invalid pair of entry host and entry port for -zentry option'
usage()
(entryHost, entryPort) = argv[argidx + 1].split(':')
try:
ip = gethostbyname_ex(entryHost)[2] # may fail if invalid host
except:
print 'invalid entry host ', entryHost
stdout.flush()
usage()
if not entryPort.isdigit():
print 'invalid (nonumeric) entry port ', entryPort
stdout.flush()
usage()
entryHost = entryHost
entryPort = entryPort
argidx += 2
elif argv[argidx] == '-zrank':
topMPDBoot = 0
myBootRank = int(argv[argidx + 1])
argidx += 2
elif argv[argidx] == '-zhosts':
zhosts = argv[argidx + 1]
zhosts = zhosts.split(',')
hostsAndInfo = []
for zhost in zhosts:
(host, ncpus, ifhn) = zhost.split(':')
hostsAndInfo.append({
'host': host,
'ncpus': ncpus,
'ifhn': ifhn
})
argidx += 2
elif argv[argidx] == '-r': # or --rsh=
rshCmd = argv[argidx + 1]
argidx += 2
elif argv[argidx].startswith('--rsh'):
splitArg = argv[argidx].split('=')
try:
rshCmd = splitArg[1]
except:
print 'mpdboot: invalid argument:', argv[argidx]
usage()
argidx += 1
elif argv[argidx] == '-u': # or --user=
user = argv[argidx + 1]
argidx += 2
elif argv[argidx].startswith('--user'):
splitArg = argv[argidx].split('=')
try:
user = splitArg[1]
except:
print 'mpdboot: invalid argument:', argv[argidx]
usage()
argidx += 1
elif argv[argidx] == '-m': # or --mpd=
mpdCmd = argv[argidx + 1]
argidx += 2
elif argv[argidx].startswith('--mpd'):
splitArg = argv[argidx].split('=')
try:
mpdCmd = splitArg[1]
except:
print 'mpdboot: invalid argument:', argv[argidx]
usage()
argidx += 1
elif argv[argidx] == '-f': # or --file=
hostsFilename = argv[argidx + 1]
argidx += 2
elif argv[argidx].startswith('--file'):
splitArg = argv[argidx].split('=')
try:
hostsFilename = splitArg[1]
except:
print 'mpdboot: invalid argument:', argv[argidx]
usage()
argidx += 1
elif argv[argidx].startswith('--ncpus'):
splitArg = argv[argidx].split('=')
try:
myNcpus = splitArg[1]
except:
print 'mpdboot: invalid argument:', argv[argidx]
usage()
argidx += 1
elif argv[argidx].startswith('--ifhn'):
splitArg = argv[argidx].split('=')
myIfhn = splitArg[1]
myHost = splitArg[1]
argidx += 1
elif argv[argidx] == '-n': # or --totalnum=
totalNum = int(argv[argidx + 1])
argidx += 2
elif argv[argidx].startswith('--totalnum'):
splitArg = argv[argidx].split('=')
try:
totalNum = int(splitArg[1])
except:
print 'mpdboot: invalid argument:', argv[argidx]
usage()
argidx += 1
elif argv[argidx] == '-d' or argv[argidx] == '--debug':
debug = 1
argidx += 1
elif argv[argidx] == '-s' or argv[argidx] == '--shell':
shell = 'bourne'
argidx += 1
elif argv[argidx] == '-v' or argv[argidx] == '--verbose':
verbosity = 1
argidx += 1
elif argv[argidx] == '-1':
oneMPDPerHost = 0
argidx += 1
elif argv[argidx] == '--loccons':
localConsoleArg = '--loccons'
argidx += 1
elif argv[argidx] == '--remcons':
remoteConsoleArg = '--remcons'
argidx += 1
else:
print 'mpdboot: unrecognized argument:', argv[argidx]
usage()
if topMPDBoot:
lines = []
if totalNum > 1:
try:
f = open(hostsFilename, 'r')
for line in f:
lines.append(line)
except:
print 'unable to open (or read) hostsfile %s' % (hostsFilename)
exit(-1)
hostsAndInfo = [{'host': myHost, 'ncpus': myNcpus, 'ifhn': myIfhn}]
for line in lines:
line = line.strip()
if not line or line[0] == '#':
continue
splitLine = re.split(r'\s+', line)
host = splitLine[0]
ncpus = 1 # default
if ':' in host:
(host, ncpus) = host.split(':', 1)
ncpus = int(ncpus)
ifhn = '' # default
for kv in splitLine[1:]:
(k, v) = kv.split('=', 1)
if k == 'ifhn':
ifhn = v
hostsAndInfo.append({'host': host, 'ncpus': ncpus, 'ifhn': ifhn})
if oneMPDPerHost and totalNum > 1:
oldHosts = hostsAndInfo[:]
hostsAndInfo = []
for x in oldHosts:
keep = 1
for y in hostsAndInfo:
if mpd_same_ips(x['host'], y['host']):
keep = 0
break
if keep:
hostsAndInfo.append(x)
if len(hostsAndInfo) < totalNum: # one is local
print 'totalNum=%d num hosts=%d' % (totalNum, len(hostsAndInfo))
print 'there are not enough hosts on which to start all processes'
exit(-1)
myBootRank = 0
if localConsoleArg:
myConsoleVal = '-n'
else:
if remoteConsoleArg:
myConsoleVal = '-n'
anMPDalreadyHere = 0
for i in range(myBootRank):
if mpd_same_ips(hostsAndInfo[i]['host'],
myHost): # if one before me on this host
myConsoleVal = '-n'
anMPDalreadyHere = 1
break
if not anMPDalreadyHere:
try:
system('%s/mpdallexit.py > /dev/null' %
(fullDirName)) # stop any current mpds
except:
pass
mpd_set_my_id('mpdboot_%s_%d' % (myHost, myBootRank))
if debug:
mpd_print(1, 'starting')
(parent, lchild,
rchild) = mpd_get_ranks_in_binary_tree(myBootRank, totalNum)
if debug:
mpd_print(1, 'p=%d l=%d r=%d' % (parent, lchild, rchild))
if myIfhn:
ifhnVal = '--if %s' % (myIfhn)
elif hostsAndInfo[myBootRank]['ifhn']:
ifhnVal = '--if %s' % (hostsAndInfo[myBootRank]['ifhn'])
else:
ifhnVal = ''
if entryHost:
cmd = '%s %s -h %s -p %s -d -e --ncpus %s %s' % \
(mpdCmd,myConsoleVal,entryHost,entryPort,myNcpus,ifhnVal)
else:
cmd = '%s %s -d -e --ncpus %s %s' % \
(mpdCmd,myConsoleVal,myNcpus,ifhnVal)
if verbosity:
mpd_print(1, 'starting local mpd on %s' % (myHost))
if debug:
mpd_print(1, 'cmd to run local mpd = :%s:' % (cmd))
if not access(mpdCmd, X_OK):
err_exit('cannot access mpd cmd :%s:' % (mpdCmd))
locMPD = Popen4(cmd, 0)
locMPDFD = locMPD.fromchild
locMPDPort = locMPDFD.readline().strip()
if locMPDPort.isdigit():
# can't do this until he's already in his ring
locMPDSocket = mpd_get_inet_socket_and_connect(myHost, int(locMPDPort))
if locMPDSocket:
msgToSend = {
'cmd': 'ping',
'host': 'ping',
'port': 0
} # dummy host & port
mpd_send_one_msg(locMPDSocket, {
'cmd': 'ping',
'host': myHost,
'port': 0
})
msg = mpd_recv_one_msg(locMPDSocket) # RMB: WITH TIMEOUT ??
if not msg or not msg.has_key('cmd') or msg['cmd'] != 'ping_ack':
err_exit(
'%d: unable to ping local mpd; invalid msg from mpd :%s:' %
(myBootRank, msg))
locMPDSocket.close()
else:
err_exit('failed to connect to mpd')
else:
err_exit('%d: invalid port from mpd %s' %
(myBootRank, str(locMPDPort)))
if not entryHost:
entryHost = myHost
entryPort = locMPDPort
if rshCmd == 'ssh':
xOpt = '-x'
else:
xOpt = ''
lfd = 0
rfd = 0
fdsToSelect = []
if debug:
debugArg = '-d'
else:
debugArg = ''
if verbosity:
verboseArg = '-v'
else:
verboseArg = ''
if lchild >= 0:
zhosts = [
"%s:%s:%s" % (h['host'], h['ncpus'], h['ifhn'])
for h in hostsAndInfo
]
if hostsAndInfo[lchild]['ifhn']:
ifhnVal = '--ifhn=%s' % (hostsAndInfo[lchild]['ifhn'])
else:
ifhnVal = ''
cmd = "%s %s %s -n '%s --ncpus=%s %s -r %s -m %s -n %d %s %s %s -zentry %s:%s -zrank %s -zhosts %s </dev/null ' " % \
(rshCmd, xOpt, hostsAndInfo[lchild]['host'], mpdbootCmd,
hostsAndInfo[lchild]['ncpus'],ifhnVal,
rshCmd, mpdCmd, totalNum, debugArg, verboseArg, remoteConsoleArg, entryHost,
entryPort, lchild,
','.join(zhosts) )
if verbosity:
mpd_print(1, 'starting remote mpd on %s' % (hostsAndInfo[lchild]))
if debug:
mpd_print(1, 'cmd to run lchild boot = :%s:' % (cmd))
lchildMPDBoot = Popen4(cmd, 0)
lfd = lchildMPDBoot.fromchild
fdsToSelect.append(lfd)
if rchild >= 0:
zhosts = [
"%s:%s:%s" % (h['host'], h['ncpus'], h['ifhn'])
for h in hostsAndInfo
]
if hostsAndInfo[rchild]['ifhn']:
ifhnVal = '--ifhn=%s' % (hostsAndInfo[rchild]['ifhn'])
else:
ifhnVal = ''
cmd = "%s %s %s -n '%s --ncpus=%s %s -r %s -m %s -n %d %s %s %s -zentry %s:%s -zrank %s -zhosts %s </dev/null ' " % \
(rshCmd, xOpt, hostsAndInfo[rchild]['host'], mpdbootCmd,
hostsAndInfo[rchild]['ncpus'],ifhnVal,
rshCmd, mpdCmd, totalNum, debugArg, verboseArg, remoteConsoleArg, entryHost,
entryPort, rchild,
','.join(zhosts) )
if verbosity:
mpd_print(1, 'starting remote mpd on %s' % (hostsAndInfo[rchild]))
if debug:
mpd_print(1, 'cmd to run rchild boot = :%s:' % (cmd))
rchildMPDBoot = Popen4(cmd, 0)
rfd = rchildMPDBoot.fromchild
fdsToSelect.append(rfd)
lfd_first_line = 1
rfd_first_line = 1
while fdsToSelect:
try:
(readyFDs, unused1, unused2) = select(fdsToSelect, [], [], 0.1)
except error, errmsg:
mpd_raise('mpdboot: select failed: errmsg=:%s:' % (errmsg))
if lfd and lfd in readyFDs:
line = lfd.readline()
if line:
if line.find('RC=MPDBOOT_ERREXIT') >= 0:
err_exit('RC=MPDBOOT_ERREXIT')
else:
if not verbosity and lfd_first_line:
lfd_first_line = 0
mpd_print(
1,
"error trying to start mpd(boot) at %d %s; output:"
% (lchild, hostsAndInfo[lchild]))
print ' ', line,
stdout.flush()
else:
lfd.close()
fdsToSelect.remove(lfd)
if rfd and rfd in readyFDs:
line = rfd.readline()
if line:
if line.find('RC=MPDBOOT_ERREXIT') >= 0:
err_exit('RC=MPDBOOT_ERREXIT')
else:
if not verbosity and rfd_first_line:
rfd_first_line = 0
mpd_print(
1,
"error trying to start mpd(boot) at %d %s; output:"
% (rchild, hostsAndInfo[rchild]))
print ' ', line,
stdout.flush()
else:
rfd.close()
fdsToSelect.remove(rfd)
mpd_send_one_line(conSocket, msgToSend)
mpdid = ''
if argv[1] == '-a':
jobalias = argv[2]
jobnum = '0'
else:
jobalias = ''
jobid = argv[1]
sjobid = jobid.split('@')
jobnum = sjobid[0]
if len(sjobid) > 1:
mpdid = sjobid[1]
mpd_send_one_msg(
conSocket, {
'cmd': 'mpdkilljob',
'jobnum': jobnum,
'mpdid': mpdid,
'jobalias': jobalias,
'username': username
})
msg = recv_one_msg_with_timeout(conSocket, 5)
if not msg:
mpd_raise('no msg recvd from mpd before timeout')
if msg['cmd'] != 'mpdkilljob_ack':
if msg['cmd'] == 'already_have_a_console':
print 'mpd already has a console (e.g. for long ringtest); try later'
else:
print 'unexpected message from mpd: %s' % (msg)
exit(-1)
if not msg['handled']:
print 'job not found'
exit(-1)
def get_args_from_file():
global nprocs, pgm, pgmArgs, mship, rship, argsFilename, delArgsFile, \
try0Locally, lineLabels, jobAlias, mergingOutput, conSocket
global stdinGoesToWho, myExitStatus, manSocket, jobid, username, cwd, totalview
global outXmlDoc, outXmlEC, outXmlFile, linesPerRank, gdb, gdbAttachJobid
global execs, users, cwds, paths, args, envvars, limits, hosts, hostList
global singinitPID, singinitPORT, doingBNR, myHost, myIP
try:
argsFile = open(argsFilename, 'r')
except:
print 'could not open job specification file %s' % (argsFilename)
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
file_contents = argsFile.read()
if delArgsFile:
unlink(argsFilename)
try:
from xml.dom.minidom import parseString #import only if needed
except:
print 'need xml parser like xml.dom.minidom'
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
parsedArgs = parseString(file_contents)
if parsedArgs.documentElement.tagName != 'create-process-group':
print 'expecting create-process-group; got unrecognized doctype: %s' % \
(parsedArgs.documentElement.tagName)
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
createReq = parsedArgs.getElementsByTagName('create-process-group')[0]
if createReq.hasAttribute('totalprocs'):
nprocs = int(createReq.getAttribute('totalprocs'))
else:
print '** totalprocs not specified in %s' % argsFilename
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
if createReq.hasAttribute('dont_try_0_locally'):
try0Locally = 0
if createReq.hasAttribute('output') and \
createReq.getAttribute('output') == 'label':
lineLabels = 1
if createReq.hasAttribute('net_interface'):
myHost = createReq.getAttribute('net_interface')
myIfhn = myHost
if createReq.hasAttribute('pgid'): # our jobalias
jobAlias = createReq.getAttribute('pgid')
if createReq.hasAttribute('stdin_goes_to_who'):
stdinGoesToWho = createReq.getAttribute('stdin_goes_to_who')
if createReq.hasAttribute('doing_bnr'):
doingBNR = int(createReq.getAttribute('doing_bnr'))
if createReq.hasAttribute('gdb'):
gdb = int(createReq.getAttribute('gdb'))
if gdb:
mergingOutput = 1 # implied
lineLabels = 1 # implied
stdinGoesToWho = 'all' # chgd to 0 - nprocs-1 when nprocs avail
if createReq.hasAttribute('tv'):
totalview = int(createReq.getAttribute('tv'))
nextHost = 0
hostSpec = createReq.getElementsByTagName('host-spec')
if hostSpec:
for node in hostSpec[0].childNodes:
node = node.data.strip()
hostnames = findall(r'\S+', node)
for hostname in hostnames:
if hostname: # some may be the empty string
try:
ipaddr = gethostbyname_ex(hostname)[2][0]
except:
print 'unable to determine IP info for host %s' % (
hostname)
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
if ipaddr.startswith('127.0.0'):
hostList.append(myHost)
else:
hostList.append(ipaddr)
if hostSpec and hostSpec[0].hasAttribute('check'):
hostSpecMode = hostSpec[0].getAttribute('check')
if hostSpecMode == 'yes':
msgToSend = {'cmd': 'verify_hosts_in_ring', 'host_list': hostList}
mpd_send_one_msg(conSocket, msgToSend)
msg = recv_one_msg_with_timeout(conSocket, 5)
if not msg:
mpd_raise('no msg recvd from mpd mpd during chk hosts up')
elif msg['cmd'] != 'verify_hosts_in_ring_response':
mpd_raise('unexpected msg from mpd :%s:' % (msg))
if msg['host_list']:
print 'These hosts are not in the mpd ring:'
for host in msg['host_list']:
if host[0].isdigit():
print ' %s' % (host),
try:
print ' (%s)' % (gethostbyaddr(host)[0])
except:
print ''
else:
print ' %s' % (host)
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
covered = [0] * nprocs
procSpec = createReq.getElementsByTagName('process-spec')
if not procSpec:
print 'No process-spec specified'
usage()
for p in procSpec:
if p.hasAttribute('range'):
therange = p.getAttribute('range')
splitRange = therange.split('-')
if len(splitRange) == 1:
loRange = int(splitRange[0])
hiRange = loRange
else:
(loRange, hiRange) = (int(splitRange[0]), int(splitRange[1]))
else:
(loRange, hiRange) = (0, nprocs - 1)
for i in xrange(loRange, hiRange + 1):
if i >= nprocs:
print '*** exiting; rank %d is greater than nprocs for args' % (
i)
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
if covered[i]:
print '*** exiting; rank %d is doubly used in proc specs' % (i)
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
covered[i] = 1
if p.hasAttribute('exec'):
execs[(loRange, hiRange)] = p.getAttribute('exec')
else:
print '*** exiting; range %d-%d has no exec' % (loRange, hiRange)
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
if p.hasAttribute('user'):
tempuser = p.getAttribute('user')
try:
pwent = getpwnam(tempuser)
except:
pwent = None
if not pwent:
print tempuser, 'is an invalid username'
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
if tempuser == username or getuid() == 0:
users[(loRange, hiRange)] = p.getAttribute('user')
else:
print tempuser, 'username does not match yours and you are not root'
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
else:
users[(loRange, hiRange)] = username
if p.hasAttribute('cwd'):
cwds[(loRange, hiRange)] = p.getAttribute('cwd')
else:
cwds[(loRange, hiRange)] = cwd
if p.hasAttribute('path'):
paths[(loRange, hiRange)] = p.getAttribute('path')
else:
paths[(loRange, hiRange)] = environ['PATH']
if p.hasAttribute('host'):
host = p.getAttribute('host')
if host.startswith('_any_'):
hosts[(loRange, hiRange)] = host
else:
try:
hosts[(loRange, hiRange)] = gethostbyname_ex(host)[2][0]
except:
print 'unable to do find info for host %s' % (host)
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
else:
if hostList:
hosts[(loRange, hiRange)] = '_any_from_pool_'
else:
hosts[(loRange, hiRange)] = '_any_'
argDict = {}
argList = p.getElementsByTagName('arg')
for argElem in argList:
argDict[int(
argElem.getAttribute('idx'))] = argElem.getAttribute('value')
argVals = [0] * len(argList)
for i in argDict.keys():
argVals[i - 1] = unquote(argDict[i])
args[(loRange, hiRange)] = argVals
limitDict = {}
limitList = p.getElementsByTagName('limit')
for limitElem in limitList:
type = limitElem.getAttribute('type')
if type in known_rlimit_types:
limitDict[type] = limitElem.getAttribute('value')
else:
print 'mpdrun: invalid type in limit: %s' % (type)
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
limits[(loRange, hiRange)] = limitDict
envVals = {}
envVarList = p.getElementsByTagName('env')
for envVarElem in envVarList:
envkey = envVarElem.getAttribute('name')
envval = envVarElem.getAttribute('value')
envVals[envkey] = envval
envvars[(loRange, hiRange)] = envVals
def mpdrun():
global nprocs, pgm, pgmArgs, mship, rship, argsFilename, delArgsFile, \
try0Locally, lineLabels, jobAlias, mergingOutput, conSocket
global stdinGoesToWho, myExitStatus, manSocket, jobid, username, cwd, totalview
global outXmlDoc, outXmlEC, outXmlFile, linesPerRank, gdb, gdbAttachJobid
global execs, users, cwds, paths, args, envvars, limits, hosts, hostList
global singinitPID, singinitPORT, doingBNR, myHost, myIP, myIfhn
mpd_set_my_id('mpdrun_' + ` getpid() `)
pgm = ''
mship = ''
rship = ''
nprocs = 0
jobAlias = ''
argsFilename = ''
outExitCodesFilename = ''
outXmlFile = ''
outXmlDoc = ''
outXmlEC = ''
delArgsFile = 0
try0Locally = 1
lineLabels = 0
stdinGoesToWho = '0'
mergingOutput = 0
hostList = []
gdb = 0
gdbAttachJobid = ''
singinitPID = 0
singinitPORT = 0
doingBNR = 0
totalview = 0
myHost = gethostname() # default; may be chgd by -if arg
myIfhn = ''
known_rlimit_types = [
'core', 'cpu', 'fsize', 'data', 'stack', 'rss', 'nproc', 'nofile',
'ofile', 'memlock', 'as', 'vmem'
]
username = mpd_get_my_username()
cwd = path.abspath(getcwd())
recvTimeout = 20
execs = {}
users = {}
cwds = {}
paths = {}
args = {}
envvars = {}
limits = {}
hosts = {}
get_args_from_cmdline(
) # verify args as much as possible before connecting to mpd
(listenSocket, listenPort) = mpd_get_inet_listen_socket('', 0)
signal(SIGALRM, sig_handler)
if environ.has_key('MPDRUN_TIMEOUT'):
jobTimeout = int(environ['MPDRUN_TIMEOUT'])
elif environ.has_key('MPIEXEC_TIMEOUT'):
jobTimeout = int(environ['MPIEXEC_TIMEOUT'])
else:
jobTimeout = 0
if environ.has_key('MPIEXEC_BNR'):
doingBNR = 1
if environ.has_key('UNIX_SOCKET'):
conFD = int(environ['UNIX_SOCKET'])
conSocket = fromfd(conFD, AF_UNIX, SOCK_STREAM)
close(conFD)
else:
if environ.has_key('MPD_CON_EXT'):
conExt = '_' + environ['MPD_CON_EXT']
else:
conExt = ''
consoleName = '/tmp/mpd2.console_' + username + conExt
conSocket = socket(AF_UNIX, SOCK_STREAM) # note: UNIX socket
try:
conSocket.connect(consoleName)
except Exception, errmsg:
print 'cannot connect to local mpd (%s); possible causes:' % consoleName
print ' 1. no mpd running on this host'
print ' 2. mpd is running but was started without a "console" (-n option)'
print 'you can start an mpd with the "mpd" command; to get help, run:'
print ' mpd -h'
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
# mpd_raise('cannot connect to local mpd; errmsg: %s' % (str(errmsg)) )
msgToSend = 'realusername=%s\n' % username
mpd_send_one_line(conSocket, msgToSend)
msgToSend = {'cmd': 'get_mpd_version'}
mpd_send_one_msg(conSocket, msgToSend)
msg = recv_one_msg_with_timeout(conSocket, recvTimeout)
if not msg:
mpd_raise('no msg recvd from mpd during version check')
elif msg['cmd'] != 'mpd_version_response':
mpd_raise('unexpected msg from mpd :%s:' % (msg))
if msg['mpd_version'] != mpd_version:
mpd_raise('mpd version %s does not match mine %s' %
(msg['mpd_version'], mpd_version))
}
if try0Locally:
msgToSend['try_0_locally'] = 1
if lineLabels:
msgToSend['line_labels'] = 1
if rship:
msgToSend['rship'] = rship
msgToSend['mship_host'] = gethostname()
msgToSend['mship_port'] = mshipPort
if doingBNR:
msgToSend['doing_bnr'] = 1
if stdinGoesToWho == 'all':
stdinGoesToWho = '0-%d' % (nprocs - 1)
msgToSend['stdin_goes_to_who'] = stdinGoesToWho
mpd_send_one_msg(conSocket, msgToSend)
msg = recv_one_msg_with_timeout(conSocket, recvTimeout)
if not msg:
mpd_raise('no msg recvd from mpd when expecting ack of request')
elif msg['cmd'] == 'mpdrun_ack':
currRingSize = msg['ringsize']
currRingNCPUs = msg['ring_ncpus']
else:
if msg['cmd'] == 'already_have_a_console':
print 'mpd already has a console (e.g. for long ringtest); try later'
myExitStatus = -1 # used in main
exit(myExitStatus) # really forces jump back into main
elif msg['cmd'] == 'job_failed':
if msg['reason'] == 'some_procs_not_started':
print 'mpdrun: unable to start all procs; may have invalid machine names'
print ' remaining specified hosts:'
conExt = ''
consoleName = '/tmp/mpd2.console_' + username + conExt
conSocket = socket(AF_UNIX, SOCK_STREAM) # note: UNIX socket
try:
conSocket.connect(consoleName)
except Exception, errmsg:
print 'mpdallexit: cannot connect to local mpd (%s); possible causes:' % consoleName
print ' 1. no mpd running on this host'
print ' 2. mpd is running but was started without a "console" (-n option)'
print 'you can start an mpd with the "mpd" command; to get help, run:'
print ' mpd -h'
exit(-1)
# mpd_raise('cannot connect to local mpd; errmsg: %s' % (str(errmsg)) )
msgToSend = 'realusername=%s\n' % username
mpd_send_one_line(conSocket,msgToSend)
mpd_send_one_msg(conSocket, {'cmd':'mpdallexit'})
msg = recv_one_msg_with_timeout(conSocket,5)
if not msg:
mpd_raise('no msg recvd from mpd before timeout')
if msg['cmd'] != 'mpdallexit_ack':
if msg['cmd'] == 'already_have_a_console':
print 'mpd already has a console (e.g. for long ringtest); try later'
exit(-1)
elif msg['cmd'] == 'invalid_username_to_make_this_request':
print 'you can not stop this mpd; it must have been started by root'
exit(-1)
else:
print 'mpdallexit failed: unexpected message from mpd: %s' % (msg)
exit(-1)
conSocket.close()