def test_exception(self):
lines = [
'+ python mr_boom.py --step-num=0 --mapper\n',
'Traceback (most recent call last):\n',
' File "mr_boom.py", line 10, in <module>\n',
' MRBoom.run()\n',
'Exception: BOOM\n',
]
self.assertEqual(
_parse_python_task_stderr(lines),
dict(error=dict(exception='Exception: BOOM',
traceback=[
'Traceback (most recent call last):',
' File "mr_boom.py", line 10, in <module>',
' MRBoom.run()',
])))
def test_exception(self):
lines = [
'+ python mr_boom.py --step-num=0 --mapper\n',
'Traceback (most recent call last):\n',
' File "mr_boom.py", line 10, in <module>\n',
' MRBoom.run()\n',
'Exception: BOOM\n',
]
self.assertEqual(
_parse_python_task_stderr(lines),
dict(error=dict(
exception='Exception: BOOM',
traceback=[
'Traceback (most recent call last):',
' File "mr_boom.py", line 10, in <module>',
' MRBoom.run()',
])))
def test_exception(self):
lines = [
"+ python mr_boom.py --step-num=0 --mapper\n",
"Traceback (most recent call last):\n",
' File "mr_boom.py", line 10, in <module>\n',
" MRBoom.run()\n",
"Exception: BOOM\n",
]
self.assertEqual(
_parse_python_task_stderr(lines),
dict(
error=dict(
exception="Exception: BOOM",
traceback=[
"Traceback (most recent call last):",
' File "mr_boom.py", line 10, in <module>',
" MRBoom.run()",
],
)
),
)
def test_empty(self):
self.assertEqual(_parse_python_task_stderr([]), dict(error=None))
def test_empty(self):
self.assertEqual(_parse_python_task_stderr([]), dict(error=None))
def _find_error_in_task_logs(fs,
log_dirs_stream,
hadoop_version,
application_id=None,
job_id=None):
"""Given a filesystem and a stream of lists of log dirs to search in,
find the last error and return details about it. *hadoop_version*
is required, as task logs have very different paths in YARN.
In YARN, you must set *application_id*, and pre-YARN, you must set
*job_id*, or we'll bail out and return None.
Returns a dictionary with the following keys ("optional" means
that something may be None):
syslog: dict with keys:
path: path of syslog we found error in
error: error details; dict with keys:
exception: Java exception (as string)
stack_trace: array of lines with Java stack trace
split: optional input split we were reading; dict with keys:
path: path of input file
start_line: first line of split (0-indexed)
num_lines: number of lines in split
stderr: optional dict with keys:
path: path of stderr corresponding to syslog
error: optional error details; dict with keys:
exception: string (Python exception)
traceback: array of lines with Python stack trace
type: always set to 'task'
"""
syslog_paths = []
yarn = uses_yarn(hadoop_version)
if ((yarn and application_id is None) or (not yarn and job_id is None)):
return None
# we assume that each set of log paths contains the same copies
# of syslogs, so stop once we find any non-empty set of log dirs
for log_dirs in log_dirs_stream:
if yarn:
syslog_paths = _ls_yarn_task_syslogs(fs,
log_dirs,
application_id=application_id)
else:
syslog_paths = _ls_pre_yarn_task_syslogs(fs,
log_dirs,
job_id=job_id)
if syslog_paths:
break
for syslog_path in syslog_paths:
log.debug('Looking for error in %s' % syslog_path)
syslog_info = _parse_task_syslog(_cat_log(fs, syslog_path))
if not syslog_info['error']:
continue
# found error! see if we can explain it
# TODO: don't bother if error wasn't due to child process
stderr_path = _stderr_for_syslog(syslog_path)
stderr_info = _parse_python_task_stderr(_cat_log(fs, stderr_path))
# output error info
syslog_info['path'] = syslog_path
stderr_info['path'] = stderr_path
return dict(type='task', syslog=syslog_info, stderr=stderr_info)
return None
def _find_error_in_task_logs(fs, log_dirs_stream, hadoop_version,
application_id=None, job_id=None):
"""Given a filesystem and a stream of lists of log dirs to search in,
find the last error and return details about it. *hadoop_version*
is required, as task logs have very different paths in YARN.
In YARN, you must set *application_id*, and pre-YARN, you must set
*job_id*, or we'll bail out and return None.
Returns a dictionary with the following keys ("optional" means
that something may be None):
syslog: dict with keys:
path: path of syslog we found error in
error: error details; dict with keys:
exception: Java exception (as string)
stack_trace: array of lines with Java stack trace
split: optional input split we were reading; dict with keys:
path: path of input file
start_line: first line of split (0-indexed)
num_lines: number of lines in split
stderr: optional dict with keys:
path: path of stderr corresponding to syslog
error: optional error details; dict with keys:
exception: string (Python exception)
traceback: array of lines with Python stack trace
type: always set to 'task'
"""
syslog_paths = []
yarn = uses_yarn(hadoop_version)
if ((yarn and application_id is None) or (not yarn and job_id is None)):
return None
# we assume that each set of log paths contains the same copies
# of syslogs, so stop once we find any non-empty set of log dirs
for log_dirs in log_dirs_stream:
if yarn:
syslog_paths = _ls_yarn_task_syslogs(fs, log_dirs,
application_id=application_id)
else:
syslog_paths = _ls_pre_yarn_task_syslogs(fs, log_dirs,
job_id=job_id)
if syslog_paths:
break
for syslog_path in syslog_paths:
log.debug('Looking for error in %s' % syslog_path)
syslog_info = _parse_task_syslog(_cat_log(fs, syslog_path))
if not syslog_info['error']:
continue
# found error! see if we can explain it
# TODO: don't bother if error wasn't due to child process
stderr_path = _stderr_for_syslog(syslog_path)
stderr_info = _parse_python_task_stderr(_cat_log(fs, stderr_path))
# output error info
syslog_info['path'] = syslog_path
stderr_info['path'] = stderr_path
return dict(type='task', syslog=syslog_info, stderr=stderr_info)
return None