def _do_cleanup(self):
"""
Periodically clean up the oldest images when the disk is close to full.
"""
logger.info('Image cleanup thread started.')
while not self._should_stop_cleanup():
# Start disk usage reduction loop
# Iteratively try to remove the oldest *used* images until the disk usage
# is within the limit.
try:
while True:
total_bytes, reclaimable_bytes = self._docker.get_disk_usage(
)
if total_bytes > self._max_images_bytes and len(
self._last_used) > 0 and reclaimable_bytes > 0:
logger.debug('Docker images disk usage: %s (max %s)',
size_str(total_bytes),
size_str(self._max_images_bytes))
self._remove_stalest_image()
else:
# Break out of the loop when disk usage is normal.
break
except Exception:
# Print the error then go to sleep and try again next time.
traceback.print_exc()
# Allow chance to be interrupted before going to sleep.
if self._should_stop_cleanup():
break
time.sleep(1)
def _check_and_report_resource_utilization(self, report):
new_metadata = {}
# Get wall clock time.
new_metadata['time'] = time.time() - self._start_time
if (self._resources['request_time'] and
new_metadata['time'] > self._resources['request_time']):
self.kill('Time limit %s exceeded.' % duration_str(self._resources['request_time']))
# Get memory, time_user and time_system.
new_metadata.update(self._docker.get_container_stats(self._container_id))
if 'memory' in new_metadata and new_metadata['memory'] > self._max_memory:
self._max_memory = new_metadata['memory']
new_metadata['memory_max'] = self._max_memory
if (self._resources['request_memory'] and
'memory' in new_metadata and
new_metadata['memory'] > self._resources['request_memory']):
self.kill('Memory limit %sb exceeded.' % size_str(self._resources['request_memory']))
# Get disk utilization.
with self._disk_utilization_lock:
new_metadata['data_size'] = self._disk_utilization
if (self._resources['request_disk'] and
new_metadata['data_size'] > self._resources['request_disk']):
self.kill('Disk limit %sb exceeded.' % size_str(self._resources['request_disk']))
new_metadata['last_updated'] = int(time.time())
if report:
logger.debug('Reporting resource utilization for run with UUID %s', self._uuid)
try:
self._bundle_service.update_bundle_metadata(self._worker.id, self._uuid, new_metadata)
except BundleServiceException:
traceback.print_exc()
def download_image(self, docker_image, loop_callback):
if len(docker_image.split(":")) < 2:
logger.debug(
'Missing tag/digest on request docker image "%s", defaulting to latest',
docker_image)
docker_image = ':'.join([docker_image, 'latest'])
logger.debug('Downloading Docker image %s', docker_image)
for response in self.client.api.pull(docker_image,
stream=True,
decode=True):
if 'error' in response:
raise DockerException(response['error'])
status = ''
try:
status = response['status']
except KeyError:
pass
try:
status += ' (%s / %s)' % (
size_str(response['progressDetail']['current']),
size_str(response['progressDetail']['total']))
except KeyError:
pass
loop_callback(status)
def status_str():
if total_bytes:
return 'Downloaded %s/%s (%d%%)' % (formatting.size_str(
num_bytes), formatting.size_str(total_bytes), 100.0 *
num_bytes / total_bytes)
else:
return 'Downloaded %s' % (formatting.size_str(num_bytes))
def status_str():
if total_bytes:
return 'Downloaded %s/%s (%d%%)' % (
formatting.size_str(num_bytes),
formatting.size_str(total_bytes),
100.0 * num_bytes / total_bytes,
)
else:
return 'Downloaded %s' % (formatting.size_str(num_bytes))
def _do_cleanup(self):
while not self._should_stop_cleanup():
while True:
# If the total size of all dependencies exceeds
# self._max_work_dir_size_bytes, remove the oldest unused
# dependency. Otherwise, break out of the loop.
total_size_bytes = 0
first_used_time = float('inf')
first_used_target = None
self._lock.acquire()
for target, dependency in self._dependencies.items():
if dependency.downloading:
continue
# We compute the size of dependencies here to keep the code
# that adds new bundles to the dependency manager simpler.
if dependency.size_bytes is None:
self._lock.release()
size_bytes = get_path_size(
os.path.join(self._bundles_dir, dependency.path))
self._lock.acquire()
dependency.size_bytes = size_bytes
self._save_state()
total_size_bytes += dependency.size_bytes
if (not dependency.has_children()
and dependency.last_used < first_used_time):
first_used_time = dependency.last_used
first_used_target = target
self._lock.release()
if (total_size_bytes > self._max_work_dir_size_bytes
and first_used_target is not None):
logger.info(
'used ({}) exceeds capacity ({}), removing oldest bundle from cache'
.format(size_str(total_size_bytes),
size_str(self._max_work_dir_size_bytes)))
with self._lock:
dependency = self._dependencies[first_used_target]
if dependency.has_children():
# Since we released the lock there could be new
# children.
continue
del self._dependencies[first_used_target]
self._paths.remove(dependency.path)
self._save_state()
remove_path(
os.path.join(self._bundles_dir, dependency.path))
else:
break
# Sleep for 10 seconds, allowing interruptions every second.
for _ in xrange(0, self._cleanup_sleep_secs):
time.sleep(1)
if self._should_stop_cleanup():
break
def download_image(self, docker_image, loop_callback):
if len(docker_image.split(":")) < 2:
logger.debug('Missing tag/digest on request docker image "%s", defaulting to latest', docker_image)
docker_image = ':'.join([docker_image, 'latest'])
logger.debug('Downloading Docker image %s', docker_image)
with closing(self._create_connection()) as conn:
conn.request('POST',
'/images/create?fromImage=%s' % docker_image)
create_image_response = conn.getresponse()
if create_image_response.status != 200:
raise DockerException(create_image_response.read())
# Wait for the download to finish. Docker sends a stream of JSON
# objects. Since we don't know how long each one is we read a
# character at a time until what we have so far parses as a valid
# JSON object.
while True:
response = None
line = ''
while True:
ch = create_image_response.read(1)
if not ch:
break
line += ch
try:
response = json.loads(line)
logger.debug(line.strip())
break
except ValueError:
pass
if not response:
break
if 'error' in response:
raise DockerException(response['error'])
status = ''
try:
status = response['status']
except KeyError:
pass
try:
status += ' (%s / %s)' % (
size_str(response['progressDetail']['current']),
size_str(response['progressDetail']['total']))
except KeyError:
pass
loop_callback(status)
def _check_and_report_resource_utilization(self, report):
new_metadata = {}
# Get wall clock time.
new_metadata['time'] = time.time() - self._start_time
if (self._resources['request_time'] and new_metadata['time'] > self._resources['request_time']):
self.kill('Time limit %s exceeded.' % duration_str(self._resources['request_time']))
# Get memory, time_user and time_system.
new_metadata.update(self._docker.get_container_stats(self._container_id))
if 'memory' in new_metadata and new_metadata['memory'] > self._max_memory:
self._max_memory = new_metadata['memory']
new_metadata['memory_max'] = self._max_memory
# Get disk utilization.
with self._disk_utilization_lock:
new_metadata['data_size'] = self._disk_utilization
if (self._resources['request_disk'] and
new_metadata['data_size'] > self._resources['request_disk']):
self.kill('Disk limit %sb exceeded.' % size_str(self._resources['request_disk']))
new_metadata['last_updated'] = int(time.time())
if report:
logger.debug('Reporting resource utilization for run with UUID %s', self._uuid)
try:
self._bundle_service.update_bundle_metadata(self._worker.id, self._uuid, new_metadata)
except BundleServiceException:
traceback.print_exc()
def copy(source, dest, autoflush=True, print_status=None):
"""
Read from the source file handle and write the data to the dest file handle.
"""
n = 0
while True:
buf = source.read(BUFFER_SIZE)
if not buf:
break
dest.write(buf)
n += len(buf)
if autoflush:
dest.flush()
if print_status:
print >>sys.stderr, "\r%s: %s" % (print_status, formatting.size_str(n)),
sys.stderr.flush()
if print_status:
print >>sys.stderr, "\r%s: %s [done]" % (print_status, formatting.size_str(n))
def copy(source, dest, autoflush=True, print_status=False):
'''
Read from the source file handle and write the data to the dest file handle.
'''
n = 0
while True:
buffer = source.read(BUFFER_SIZE)
if not buffer:
break
dest.write(buffer)
n += len(buffer)
if autoflush:
dest.flush()
if print_status:
print >>sys.stderr, "\rCopied %s" % formatting.size_str(n),
sys.stderr.flush()
if print_status:
print >>sys.stderr, "\rCopied %s [done]" % formatting.size_str(n)
def download_image(self, docker_image, loop_callback):
logger.debug('Downloading Docker image %s', docker_image)
with closing(self._create_connection()) as conn:
conn.request('POST',
'/images/create?fromImage=%s' % docker_image)
create_image_response = conn.getresponse()
if create_image_response.status != 200:
raise DockerException(create_image_response.read())
# Wait for the download to finish. Docker sends a stream of JSON
# objects. Since we don't know how long each one is we read a
# character at a time until what we have so far parses as a valid
# JSON object.
while True:
response = None
line = ''
while True:
ch = create_image_response.read(1)
if not ch:
break
line += ch
try:
response = json.loads(line)
logger.debug(line.strip())
break
except ValueError:
pass
if not response:
break
if 'error' in response:
raise DockerException(response['error'])
status = ''
try:
status = response['status']
except KeyError:
pass
try:
status += ' (%s / %s)' % (
size_str(response['progressDetail']['current']),
size_str(response['progressDetail']['total']))
except KeyError:
pass
loop_callback(status)
def update_status(bytes_uploaded):
updater('Uploading results: %s done (archived size)' %
size_str(bytes_uploaded))
def update_status_and_check_killed(bytes_downloaded):
updater('Downloading dependency %s/%s: %s done (archived size)' %
(uuid, path, size_str(bytes_downloaded)))
self._check_killed()
def update_status(bytes_uploaded):
updater('Uploading results: %s done (archived size)' %
size_str(bytes_uploaded))
def update_status_and_check_killed(bytes_downloaded):
updater('Downloading dependency %s: %s done (archived size)' % (
dep['child_path'], size_str(bytes_downloaded)))
check_killed()
def update_status_and_check_killed(bytes_downloaded):
updater(
'Downloading dependency %s: %s done (archived size)'
% (dep['child_path'], size_str(bytes_downloaded)))
check_killed()
