def is_atlassetup_error(job):
"""
Did AtlasSetup fail with a fatal error?
:param job: job object.
:return: Boolean. (note: True means the error was found)
"""
stdout = os.path.join(job.workdir, config.Payload.payloadstdout)
_tail = tail(stdout)
res_tmp = _tail[:2048]
return "AtlasSetup(FATAL): Fatal exception" in res_tmp
def is_installation_error(job):
"""
Did the payload fail to run? (Due to faulty/missing installation).
:param job: job object.
:return: Boolean. (note: True means the error was found)
"""
stdout = os.path.join(job.workdir, config.Payload.payloadstdout)
_tail = tail(stdout)
res_tmp = _tail[:1024]
return res_tmp[
0:
3] == "sh:" and 'setup.sh' in res_tmp and 'No such file or directory' in res_tmp
def is_atlassetup_error(job):
"""
Did AtlasSetup fail with a fatal error?
:param job: job object.
:return: Boolean. (note: True means the error was found)
"""
stdout = os.path.join(job.workdir, config.Payload.payloadstdout)
_tail = tail(stdout)
res_tmp = _tail[:2048]
if "AtlasSetup(FATAL): Fatal exception" in res_tmp:
logger.warning('AtlasSetup FATAL failure detected')
return True
else:
return False
def set_error_nousertarball(job):
"""
Set error code for NOUSERTARBALL.
:param job: job object.
:return:
"""
# get the tail of the stdout since it will contain the URL of the user log
filename = os.path.join(job.workdir, config.Payload.payloadstdout)
_tail = tail(filename)
_tail += 'http://someurl.se/path'
if _tail:
# try to extract the tarball url from the tail
tarball_url = extract_tarball_url(_tail)
job.piloterrorcodes, job.piloterrordiags = errors.add_error_code(errors.NOUSERTARBALL)
job.piloterrorcode = errors.NOUSERTARBALL
job.piloterrordiag = "User tarball %s cannot be downloaded from PanDA server" % tarball_url
def get_pilot_log_extracts(job):
"""
Get the extracts from the pilot log (warning/fatal messages, as well as tail of the log itself).
:param job: job object.
:return: tail of pilot log (string).
"""
log = get_logger(job.jobid)
extracts = ""
path = os.path.join(job.workdir, config.Pilot.pilotlog)
if os.path.exists(path):
# get the last 20 lines of the pilot log in case it contains relevant error information
_tail = tail(path, nlines=20)
if _tail != "":
if extracts != "":
extracts += "\n"
extracts += "- Log from %s -" % config.Pilot.pilotlog
extracts += _tail
# grep for fatal/critical errors in the pilot log
#errormsgs = ["FATAL", "CRITICAL", "ERROR"]
#matched_lines = grep(errormsgs, path)
#_extracts = ""
#if len(matched_lines) > 0:
# log.debug("dumping warning messages from %s:\n" % os.path.basename(path))
# for line in matched_lines:
# _extracts += line + "\n"
#if _extracts != "":
# if config.Pilot.error_log != "":
# path = os.path.join(job.workdir, config.Pilot.error_log)
# write_file(path, _extracts)
# extracts += "\n- Error messages from %s -\n" % config.Pilot.pilotlog
# extracts += _extracts
else:
log.warning('pilot log file does not exist: %s' % path)
return extracts
def get_pilot_log_extracts(job):
"""
Get the extracts from the pilot log (warning/fatal messages, as well as tail of the log itself).
:param job: job object.
:return: tail of pilot log (string).
"""
extracts = ""
path = os.path.join(job.workdir, config.Pilot.pilotlog)
if os.path.exists(path):
# get the last 20 lines of the pilot log in case it contains relevant error information
_tail = tail(path, nlines=20)
if _tail != "":
if extracts != "":
extracts += "\n"
extracts += "- Log from %s -\n" % config.Pilot.pilotlog
extracts += _tail
else:
logger.warning('pilot log file does not exist: %s' % path)
return extracts