class Capture:
def __init__(self, prefix, echo_stdout=True, echo_stderr=True,
capture_logging=False):
self.prefix = prefix
self.echo_stdout = echo_stdout
self.echo_stderr = echo_stderr
self.capture_logging = capture_logging
def start(self):
self.capture = OutputCapture(prefix=self.prefix,
echo_stdout=self.echo_stdout,
echo_stderr=self.echo_stderr)
# TODO: add whether we should just capture and not echo
self.old_emit = logging.StreamHandler.emit
# if self.capture_logging:
def my_emit(_, log_record):
msg = colorize_loglevel(log_record.levelno, log_record.msg)
# levelname = log_record.levelname
name = log_record.name
#print('%s:%s:%s' % (name, levelname, msg))
# print('%s:%s' % (name, msg))
self.capture.old_stderr.write('>%s:%s\n' % (name, msg))
logging.StreamHandler.emit = my_emit
@contextmanager
def go(self):
self.start()
try:
yield
finally:
self.stop()
def stop(self):
self.capture.deactivate()
logging.StreamHandler.emit = self.old_emit
def get_logged_stderr(self):
return self.capture.get_logged_stderr()
def get_logged_stdout(self):
return self.capture.get_logged_stdout()
def start(self):
self.capture = OutputCapture(prefix=self.prefix,
echo_stdout=self.echo_stdout,
echo_stderr=self.echo_stderr)
# TODO: add whether we should just capture and not echo
self.old_emit = logging.StreamHandler.emit
# if self.capture_logging:
def my_emit(_, log_record):
msg = colorize_loglevel(log_record.levelno, log_record.msg)
# levelname = log_record.levelname
name = log_record.name
#print('%s:%s:%s' % (name, levelname, msg))
# print('%s:%s' % (name, msg))
self.capture.old_stderr.write('>%s:%s\n' % (name, msg))
logging.StreamHandler.emit = my_emit
def make(job_id, more=False):
""" Makes a single job. Returns the user-object or raises JobFailed """
host = compmake_config.hostname #@UndefinedVariable
setproctitle(job_id)
# TODO: should we make sure we are up to date???
up, reason = up_to_date(job_id) #@UnusedVariable
cache = get_job_cache(job_id)
want_more = cache.state == Cache.MORE_REQUESTED
if up and not (more and want_more):
# print "%s is up to date" % job_id
assert is_job_userobject_available(job_id)
return get_job_userobject(job_id)
else:
# if up and (more and want_more): # XXX review the logic
# reason = 'want more'
# print "Making %s (%s)" % (job_id, reason)
computation = get_job(job_id)
assert(cache.state in [Cache.NOT_STARTED, Cache.IN_PROGRESS,
Cache.MORE_REQUESTED, Cache.DONE, Cache.FAILED])
if cache.state == Cache.NOT_STARTED:
previous_user_object = None
cache.state = Cache.IN_PROGRESS
if cache.state == Cache.FAILED:
previous_user_object = None
cache.state = Cache.IN_PROGRESS
elif cache.state == Cache.IN_PROGRESS:
if is_job_tmpobject_available(job_id):
previous_user_object = get_job_tmpobject(job_id)
else:
previous_user_object = None
elif cache.state == Cache.MORE_REQUESTED:
assert(is_job_userobject_available(job_id))
if is_job_tmpobject_available(job_id):
# resuming more computation
previous_user_object = get_job_tmpobject(job_id)
else:
# starting more computation
previous_user_object = get_job_userobject(job_id)
elif cache.state == Cache.DONE:
# If we are done, it means children have been updated
assert(not up)
previous_user_object = None
else:
assert(False)
# update state
cache.time_start = time()
cpu_start = clock()
set_job_cache(job_id, cache)
def progress_callback(stack):
publish('job-progress-plus', job_id=job_id, host=host, stack=stack)
init_progress_tracking(progress_callback)
num, total = 0, None
user_object = None
capture = OutputCapture(prefix=job_id,
echo_stdout=compmake_config.echo_stdout, #@UndefinedVariable
echo_stderr=compmake_config.echo_stderr) #@UndefinedVariable
try:
result = computation.compute(previous_user_object)
if type(result) == GeneratorType:
try:
while True:
next = result.next()
if isinstance(next, tuple):
if len(next) != 3:
raise CompmakeException('If computation yields a tuple, ' +
'should be a tuple with 3 elemnts.' +
'Got: %s' % str(next))
user_object, num, total = next
publish('job-progress', job_id=job_id, host=host,
done=None, progress=num, goal=total)
if compmake_config.save_progress: #@UndefinedVariable
set_job_tmpobject(job_id, user_object)
except StopIteration:
pass
else:
publish('job-progress', job_id=job_id, host='XXX',
done=1, progress=1, goal=1)
user_object = result
except KeyboardInterrupt:
# TODO: clear progress cache
# Save the current progress:
cache.iterations_in_progress = num
cache.iterations_goal = total
if user_object:
set_job_tmpobject(job_id, user_object)
set_job_cache(job_id, cache)
# clear progress cache
publish('job-interrupted', job_id=job_id, host=host)
raise JobInterrupted('Keyboard interrupt')
except Exception as e:
sio = StringIO()
print_exc(file=sio)
bt = sio.getvalue()
error("Job %s failed: %s" % (job_id, e))
error(bt)
mark_as_failed(job_id, e, bt)
# clear progress cache
publish('job-failed', job_id=job_id, host=host, reason=e)
raise JobFailed('Job %s failed: %s' % (job_id, e))
finally:
capture.deactivate()
# even if we send an error, let's save the output of the process
cache = get_job_cache(job_id)
cache.captured_stderr = capture.stderr_replacement.buffer.getvalue()
cache.captured_stdout = capture.stdout_replacement.buffer.getvalue()
set_job_cache(job_id, cache)
set_job_userobject(job_id, user_object)
if is_job_tmpobject_available(job_id):
# We only have one with yield
delete_job_tmpobject(job_id)
cache.state = Cache.DONE
cache.timestamp = time()
walltime = cache.timestamp - cache.time_start
cputime = clock() - cpu_start
# FIXME walltime/cputime not precise (especially for "more" computation)
cache.walltime_used = walltime
cache.cputime_used = cputime
cache.done_iterations = num # XXX not true
cache.host = compmake_config.hostname #@UndefinedVariable
set_job_cache(job_id, cache)
publish('job-succeeded', job_id=job_id, host=host)
# TODO: clear these records in other place
return user_object
