async def set_finished(self, job_id, succeeded):
"""Mark an started job as finished."""
now = format_utc(utcnow())
async with self.lock:
if job_id not in self._state['started_jobs']:
return False
job = self._state['started_jobs'][job_id]
job['finished'] = now
job['duration'] = (
diff_sec(parse_utc(now), parse_utc(job['started'])))
job['pid'] = None
job['succeeded'] = succeeded
if succeeded:
await self.history.update(job)
self._state['finished_jobs'][job_id] = job
del self._state['started_jobs'][job_id]
if succeeded:
self.logger.info(f'Finished job {job_id} [succeeded]')
else:
self.logger.warning(f'Finished job {job_id} [FAILED]')
self.lock.notify_all()
self._check_empty()
self._schedule_dump()
return True
async def _handle_du(self):
"""summarize parameters with time information"""
params = await self._execute_chain('params')
param_id_max_len = self._get_param_id_max_len(params)
for param in params:
meta = param['_']
line = f'{get_param_id(param):{param_id_max_len}} '
line += f'{get_hash_id(param)} '
if meta['finished'] is not None:
finished = meta['finished']
if self.args.local:
finished = format_local(parse_utc(finished))
line += f"[{finished.partition('.')[0]:>19}] "
else:
line += ' ' * (19 + 3)
if meta['duration'] is not None:
line += f"[{format_sec_short(meta['duration']):>7}] "
else:
line += ' ' * (7 + 3)
if meta['succeeded'] is None:
line += f' '
elif meta['succeeded']:
line += f'succeeded '
else:
line += f'FAILED '
line += get_name(param)
print(line)
async def estimate_remaining_time(self, state, oneshot, use_similar):
"""Estimate the remaining time using the queue state."""
now = utcnow()
epsilon = 0.001 # Potential underestimation until the progress reaches 0.1%
# Future parallelism cannot be higher than the remaining job count
concurrency = max(
1.,
min(state['concurrency'],
len(state['started_jobs']) + len(state['queued_jobs'])))
hash_ids = await self.history.hash_ids()
history_list = await self.history.history_list(hash_ids)
history_map = dict(zip(hash_ids, history_list))
if use_similar:
fid_mapping, fvec_map = await self._make_fid_mapping(
history_map, hash_ids)
# Estimate average per-job duration
known_hash_ids = set()
known_duration = 0.
known_count = 0
# Consider recent jobs first (in case some jobs have duplicate hash_id)
for job in reversed(state['started_jobs']):
hash_id = get_hash_id(job['param'])
if hash_id in known_hash_ids:
continue
known_hash_ids.add(hash_id)
if job['started'] is None:
started = now
else:
started = parse_utc(job['started'])
if job.get('status',
None) and job['status'].get('progress') >= epsilon:
known_duration += diff_sec(now,
started) / job['status']['progress']
known_count += 1
for hash_id, history in history_map.items():
if hash_id in known_hash_ids:
continue
known_hash_ids.add(hash_id)
if history['duration'] is not None:
known_duration += history['duration']
known_count += 1
avg_duration = known_duration / max(known_count, 1)
remaining_time_map = {}
for job in state['finished_jobs']:
remaining_time_map[job['job_id']] = 0.
# Calculate started jobs' remaining time
remaining_duration = 0.
for job in state['started_jobs']:
hash_id = get_hash_id(job['param'])
history = history_map.get(hash_id, None)
if job['started'] is None:
started = now
else:
started = parse_utc(job['started'])
if job.get('status',
None) and job['status'].get('progress') >= epsilon:
exp_duration = diff_sec(now,
started) / job['status']['progress']
remaining_duration += max(
exp_duration - diff_sec(now, started), 0.)
elif history and history['duration'] is not None:
remaining_duration += max(
history['duration'] - diff_sec(now, started), 0.)
else:
if use_similar:
exp_duration = (await self._find_closest_duration(
history_map, fid_mapping, fvec_map, job['param']))
if exp_duration is None:
exp_duration = avg_duration
else:
exp_duration = avg_duration
remaining_duration += max(
exp_duration - diff_sec(now, started), 0.)
# Take into account concurrency
remaining_time_map[
job['job_id']] = remaining_duration / concurrency
# Calculate queued jobs' remaining time
if not oneshot:
for job in state['queued_jobs']:
hash_id = get_hash_id(job['param'])
history = history_map.get(hash_id, None)
if history and history['duration'] is not None:
remaining_duration += history['duration']
else:
if use_similar:
exp_duration = (await self._find_closest_duration(
history_map, fid_mapping, fvec_map, job['param']))
if exp_duration is None:
exp_duration = avg_duration
else:
exp_duration = avg_duration
remaining_duration += exp_duration
# Take into account concurrency
remaining_time_map[
job['job_id']] = remaining_duration / concurrency
else:
for job in state['queued_jobs']:
remaining_time_map[
job['job_id']] = remaining_duration / concurrency
# Take into account concurrency
remaining_time = remaining_duration / concurrency
return remaining_time, remaining_time_map
def test_diff_sec():
t1 = parse_utc('2000-01-02 03:04:05.678901')
t2 = parse_utc('2000-01-02 02:03:04.678900')
assert diff_sec(t1, t2) == 3600 + 60 + 1 + 0.000001
def test_parse_utc_format_utc_short():
assert format_utc_short(
parse_utc('2000-01-02 03:04:05.678901')) == '2000-01-02 03:04:05'
def test_as_utc_as_local_as_utc():
t = parse_utc('2000-01-02 03:04:05.678901')
assert t == as_utc(as_local(t))
t = parse_local('2000-01-02 03:04:05.678901')
assert t == as_local(as_utc(t))
def test_as_utc():
t = parse_utc('2000-01-02 03:04:05.678901')
assert t == as_utc(t)
def test_job_elapsed_time_started_job(mock_utcnow):
mock_utcnow.return_value = parse_utc('2000-01-02 03:04:15.678901')
job = {'finished': None, 'started': '2000-01-02 03:04:05.678901'}
assert job_elapsed_time(job) == 10.
