def _distribute_events(self, reply_loc=None):
next_worker = 0
items_run_time = 0
try:
tasks_events = self._fetch_tasks_events_to_send()
items_run_time = ratelimit(items_run_time,
self.max_items_per_second)
next_worker = self._send_task_event(next(tasks_events), reply_loc,
next_worker)
self.sending = True
for task_event in tasks_events:
items_run_time = ratelimit(items_run_time,
self.max_items_per_second)
next_worker = self._send_task_event(task_event, reply_loc,
next_worker)
if not self.tool.running:
break
except Exception as exc:
if not isinstance(exc, StopIteration) and self.tool.running:
self.logger.error("Failed to distribute events: %s", exc)
self.tool.success = False
finally:
self.sending = False
def process(self, beanstalkd_job):
job_id = beanstalkd_job['job_id']
job_config = beanstalkd_job['job_config']
task = self.tasks.get(job_id)
if task is None:
job_type = beanstalkd_job['job_type']
task_class = JOB_TYPES[job_type].TASK_CLASS
job_params = job_config['params']
task = task_class(self.conf, job_params, logger=self.logger)
self.tasks[job_id] = task
tasks_per_second = job_config['tasks_per_second']
tasks = beanstalkd_job['tasks']
task_errors = Counter()
task_results = Counter()
tasks_run_time = 0
for task_id, task_payload in iteritems(tasks):
tasks_run_time = ratelimit(tasks_run_time, tasks_per_second)
reqid = job_id + request_id('-')
reqid = reqid[:STRLEN_REQID]
try:
task_result = task.process(task_id, task_payload, reqid=reqid)
task_results.update(task_result)
except Exception as exc:
self.logger.warn('[job_id=%s] Fail to process task %s: %s',
job_id, task_id, exc)
task_errors[type(exc).__name__] += 1
return job_id, list(tasks.keys()), task_results, task_errors, \
beanstalkd_job['beanstalkd_reply']
def rebuilder_pass(self, num, queue, **kwargs):
start_time = report_time = time.time()
while True:
item = queue.get()
begin_time = time.time()
self._rebuild_one(item, **kwargs)
end_time = time.time()
self.rebuilder_time += (end_time - begin_time)
total_time = end_time - start_time
self.waiting_time = total_time - self.rebuilder_time
self.total_items_processed += 1
queue.task_done()
if end_time - self.last_reported >= self.report_interval:
self.logger.info(
self._get_report(num, start_time, end_time, total_time,
report_time, **kwargs))
report_time = end_time
self.last_reported = end_time
self.passes = 0
self.items_run_time = ratelimit(self.items_run_time,
self.max_items_per_second)
def rebuilder_pass(self, num, queue, retry_queue=None, **kwargs):
while True:
info = None
err = None
item = queue.get()
try:
info = self._rebuild_one(item, **kwargs)
except exceptions.RetryLater as exc:
if retry_queue:
self.logger.warn("Putting an item in the retry queue: %s",
exc.args[1])
retry_queue.put(exc.args[0])
else:
err = str(exc)
except Exception as exc:
err = str(exc)
queue.task_done()
self.update_processed(item, info, error=err, **kwargs)
self.log_report(**kwargs)
self.items_run_time = ratelimit(self.items_run_time,
self.max_items_per_second)
if self.random_wait:
eventlet.sleep(random.randint(0, self.random_wait) / 1.0e6)
def _distribute_events(self, reply_loc=None):
next_worker = 0
items_run_time = 0
tasks_events = self._fetch_tasks_events_to_send()
try:
items_run_time = ratelimit(items_run_time,
self.max_items_per_second)
next_worker = self._send_task_event(next(tasks_events), reply_loc,
next_worker)
self.sending = True
except StopIteration:
self.sending = False
return
for task_event in tasks_events:
items_run_time = ratelimit(items_run_time,
self.max_items_per_second)
next_worker = self._send_task_event(task_event, reply_loc,
next_worker)
self.sending = False
def converter_pass(self, input_file=None):
def report(tag, now=None):
if now is None:
now = time.time()
total_time = now - self.start_time
self.logger.info(
'%(tag)s %(volume)s '
'started=%(start_time)s '
'passes=%(passes)d '
'errors=%(errors)d '
'chunks=%(nb_chunks)d %(c_rate).2f/s '
'total_time=%(total_time).2f '
'(converter: %(success_rate).2f%%)' % {
'tag':
tag,
'volume':
self.volume_id,
'start_time':
datetime.fromtimestamp(int(self.start_time)).isoformat(),
'passes':
self.passes,
'errors':
self.errors,
'nb_chunks':
self.total_chunks_processed,
'c_rate':
self.total_chunks_processed / total_time,
'total_time':
total_time,
'success_rate':
100 * ((self.total_chunks_processed - self.errors) /
float(self.total_chunks_processed))
})
self.passes = 0
self.last_reported = now
self.start_time = time.time()
self.errors = 0
self.passes = 0
self.backup_name = 'backup_%s_%f' % (self.volume_id, self.start_time)
paths = self.paths_gen(input_file=input_file)
for path in paths:
self.safe_convert_chunk(path)
now = time.time()
if now - self.last_reported >= self.report_interval:
report('RUN', now=now)
self.chunks_run_time = ratelimit(self.chunks_run_time,
self.max_chunks_per_second)
report('DONE')
return self.errors == 0
def index_pass(self):
start_time = report_time = time.time()
total_errors = 0
paths = paths_gen(self.volume)
for path in paths:
self.safe_update_index(path)
self.chunks_run_time = ratelimit(self.chunks_run_time,
self.max_chunks_per_second)
self.total_chunks_processed += 1
now = time.time()
if now - self.last_reported >= self.report_interval:
self.logger.info(
'started=%(start_time)s '
'passes=%(passes)d '
'errors=%(errors)d '
'chunks=%(nb_chunks)d %(c_rate).2f/s '
'total=%(total).2f ' % {
'start_time':
datetime.fromtimestamp(int(report_time)).isoformat(),
'passes':
self.passes,
'errors':
self.errors,
'nb_chunks':
self.total_chunks_processed,
'c_rate':
self.passes / (now - report_time),
'total': (now - start_time)
})
report_time = now
total_errors += self.errors
self.passes = 0
self.errors = 0
self.last_reported = now
end_time = time.time()
elapsed = (end_time - start_time) or 0.000001
self.logger.info(
'started=%(start_time)s '
'ended=%(end_time)s '
'elapsed=%(elapsed).02f '
'errors=%(errors)d '
'chunks=%(nb_chunks)d %(c_rate).2f/s ' % {
'start_time': datetime.fromtimestamp(
int(start_time)).isoformat(),
'end_time': datetime.fromtimestamp(int(end_time)).isoformat(),
'elapsed': elapsed,
'errors': total_errors + self.errors,
'nb_chunks': self.total_chunks_processed,
'c_rate': self.total_chunks_processed / elapsed
})
if elapsed < self.interval:
time.sleep(self.interval - elapsed)
def index_pass(self):
def safe_update_index(path):
chunk_id = path.rsplit('/', 1)[-1]
if len(chunk_id) != STRLEN_CHUNKID:
return
for c in chunk_id:
if c not in hexdigits:
return
try:
self.update_index(path)
self.successes += 1
self.logger.debug('Updated %s', path)
except OioNetworkException as exc:
self.errors += 1
self.logger.warn('ERROR while updating %s: %s', path, exc)
except Exception:
self.errors += 1
self.logger.exception('ERROR while updating %s', path)
def report(tag):
total = self.errors + self.successes
now = time.time()
elapsed = (now - start_time) or 0.000001
self.logger.info(
'%(tag)s=%(current_time)s '
'elapsed=%(elapsed).02f '
'pass=%(pass)d '
'errors=%(errors)d '
'chunks=%(nb_chunks)d %(c_rate).2f/s' % {
'tag': tag,
'current_time': datetime.fromtimestamp(
int(now)).isoformat(),
'pass': self.passes,
'errors': self.errors,
'nb_chunks': total,
'c_rate': total / (now - self.last_reported),
'elapsed': elapsed
})
self.last_reported = now
start_time = time.time()
self.last_reported = start_time
self.errors = 0
self.successes = 0
paths = paths_gen(self.volume)
report('started')
for path in paths:
safe_update_index(path)
self.chunks_run_time = ratelimit(self.chunks_run_time,
self.max_chunks_per_second)
now = time.time()
if now - self.last_reported >= self.report_interval:
report('running')
report('ended')
def _fill_queue(self):
"""
Fill the queue.
"""
items_run_time = 0
items_with_beanstalkd_reply = \
self.tool.fetch_items_with_beanstalkd_reply()
for item_with_beanstalkd_reply in items_with_beanstalkd_reply:
items_run_time = ratelimit(items_run_time,
self.max_items_per_second)
self.queue_workers.put(item_with_beanstalkd_reply)
def chunk_file_audit(self, chunk_file, chunk_id):
try:
meta, _ = read_chunk_metadata(chunk_file, chunk_id)
except exc.MissingAttribute as err:
raise exc.FaultyChunk(err)
size = int(meta['chunk_size'])
md5_checksum = meta['chunk_hash'].lower()
reader = ChunkReader(chunk_file,
size,
md5_checksum,
compression=meta.get("compression", ""))
with closing(reader):
for buf in reader:
buf_len = len(buf)
self.bytes_running_time = ratelimit(self.bytes_running_time,
self.max_bytes_per_second,
increment=buf_len)
self.bytes_processed += buf_len
self.total_bytes_processed += buf_len
try:
container_id = meta['container_id']
content_id = meta['content_id']
_obj_meta, data = self.container_client.content_locate(
cid=container_id, content=content_id, properties=False)
# Check chunk data
chunk_data = None
metachunks = set()
for c in data:
if c['url'].endswith(meta['chunk_id']):
metachunks.add(c['pos'].split('.', 2)[0])
chunk_data = c
if not chunk_data:
raise exc.OrphanChunk('Not found in content')
metachunk_size = meta.get('metachunk_size')
if metachunk_size is not None \
and chunk_data['size'] != int(metachunk_size):
raise exc.FaultyChunk('Invalid metachunk size found')
metachunk_hash = meta.get('metachunk_hash')
if metachunk_hash is not None \
and chunk_data['hash'] != meta['metachunk_hash']:
raise exc.FaultyChunk('Invalid metachunk hash found')
if chunk_data['pos'] != meta['chunk_pos']:
raise exc.FaultyChunk('Invalid chunk position found')
except exc.NotFound:
raise exc.OrphanChunk('Chunk not found in container')
def rebuilder_pass(self, num, queue, **kwargs):
while True:
info = None
err = None
item = queue.get()
try:
info = self._rebuild_one(item, **kwargs)
except Exception as exc:
err = str(exc)
queue.task_done()
self.update_processed(item, info, error=err, **kwargs)
self.log_report(**kwargs)
self.items_run_time = ratelimit(self.items_run_time,
self.max_items_per_second)
def chunk_audit(self, path):
with open(path) as f:
try:
meta = read_chunk_metadata(f)
except exc.MissingAttribute as e:
raise exc.FaultyChunk('Missing extended attribute %s' % e)
size = int(meta['chunk_size'])
md5_checksum = meta['chunk_hash'].lower()
reader = ChunkReader(f, size, md5_checksum)
with closing(reader):
for buf in reader:
buf_len = len(buf)
self.bytes_running_time = ratelimit(
self.bytes_running_time,
self.max_bytes_per_second,
increment=buf_len)
self.bytes_processed += buf_len
self.total_bytes_processed += buf_len
try:
container_id = meta['container_id']
content_path = meta['content_path']
content_attr, data = self.container_client.content_locate(
cid=container_id, path=content_path)
# Check chunk data
chunk_data = None
metachunks = set()
for c in data:
if c['url'].endswith(meta['chunk_id']):
metachunks.add(c['pos'].split('.', 2)[0])
chunk_data = c
if not chunk_data:
raise exc.OrphanChunk('Not found in content')
if chunk_data['size'] != int(meta['chunk_size']):
raise exc.FaultyChunk('Invalid chunk size found')
if chunk_data['hash'] != meta['chunk_hash']:
raise exc.FaultyChunk('Invalid chunk hash found')
if chunk_data['pos'] != meta['chunk_pos']:
raise exc.FaultyChunk('Invalid chunk position found')
except exc.NotFound:
raise exc.OrphanChunk('Chunk not found in container')
def index_pass(self):
start_time = time.time()
self.last_reported = start_time
self.errors = 0
self.successes = 0
paths = paths_gen(self.volume)
self.report('started', start_time)
for path in paths:
self.safe_update_index(path)
self.chunks_run_time = ratelimit(
self.chunks_run_time,
self.max_chunks_per_second
)
now = time.time()
if now - self.last_reported >= self.report_interval:
self.report('running', start_time)
self.report('ended', start_time)
def run(self):
start_time = report_time = time.time()
total_errors = 0
for (container_id, content_id) in self._list_contents():
self.safe_change_policy(container_id, content_id)
self.contents_run_time = ratelimit(
self.contents_run_time,
self.max_contents_per_second
)
self.total_contents_processed += 1
now = time.time()
if now - self.last_reported >= self.report_interval:
self.logger.info(
'%(start_time)s '
'%(passes)d '
'%(errors)d '
'%(c_rate).2f '
'%(total).2f ' % {
'start_time': time.ctime(report_time),
'passes': self.passes,
'errors': self.errors,
'c_rate': self.passes / (now - report_time),
'total': (now - start_time)
}
)
report_time = now
total_errors += self.errors
self.passes = 0
self.errors = 0
self.last_reported = now
elapsed = (time.time() - start_time) or 0.000001
self.logger.info(
'%(elapsed).02f '
'%(errors)d '
'%(content_rate).2f ' % {
'elapsed': elapsed,
'errors': total_errors + self.errors,
'content_rate': self.total_contents_processed / elapsed
}
)
def _fill_queue(self):
"""
Fill the queue.
"""
items_run_time = 0
try:
items_with_beanstalkd_reply = \
self.tool.fetch_items_with_beanstalkd_reply()
for item_with_beanstalkd_reply in items_with_beanstalkd_reply:
items_run_time = ratelimit(items_run_time,
self.max_items_per_second)
self.queue_workers.put(item_with_beanstalkd_reply)
if not self.tool.running:
break
except Exception as exc:
if self.tool.running:
self.logger.error("Failed to fill queue: %s", exc)
self.tool.success = False
def pass_volume(self):
self.start_time = self.last_report = time.time()
self.log_report('START', force=True)
paths = paths_gen(self.volume)
for path in paths:
try:
self.pass_chunk_file(path)
self.chunks_processed += 1
except Exception as exc:
self.logger.error(
'Failed to pass chunk file (chunk_file=%s): %s', path, exc)
self.chunk_errors += 1
self.log_report('RUN')
self.chunks_run_time = ratelimit(self.chunks_run_time,
self.max_chunks_per_second)
self.log_report('DONE', force=True)
return self.chunk_errors == 0 \
and all(errors == 0 for errors in self.bean_errors.values())
def crawl_volume(self):
"""
Crawl the volume assigned to this worker, and index every database.
"""
paths = paths_gen(self.volume)
self.full_scan_nb += 1
self.success_nb = 0
self.failed_nb = 0
now = time.time()
self.last_report_time = now
self.report("starting")
for db_path in paths:
# Graceful exit, hopefully
if self._stop:
break
db_id = db_path.rsplit("/")[-1].rsplit(".")
if len(db_id) != 3:
self.warn("Malformed db file name !", db_path)
continue
db_id = ".".join(db_id[:2])
self.index_meta2_database(db_id)
self.last_index_time = ratelimit(
self.last_index_time,
self.max_indexed_per_second
)
now = time.time()
if now - self.last_report_time >= self.report_interval:
self.report("running")
self.report("ended")
def audit_pass(self):
self.namespace, self.address = check_volume(self.volume)
start_time = report_time = time.time()
total_errors = 0
total_corrupted = 0
total_orphans = 0
total_faulty = 0
audit_time = 0
paths = paths_gen(self.volume)
for path in paths:
loop_time = time.time()
self.safe_chunk_audit(path)
self.chunks_run_time = ratelimit(self.chunks_run_time,
self.max_chunks_per_second)
self.total_chunks_processed += 1
now = time.time()
if now - self.last_reported >= self.report_interval:
self.logger.info(
'%(start_time)s '
'%(passes)d '
'%(corrupted)d '
'%(faulty)d '
'%(orphans)d '
'%(errors)d '
'%(c_rate).2f '
'%(b_rate).2f '
'%(total).2f '
'%(audit_time).2f'
'%(audit_rate).2f' % {
'start_time': time.ctime(report_time),
'passes': self.passes,
'corrupted': self.corrupted_chunks,
'faulty': self.faulty_chunks,
'orphans': self.orphan_chunks,
'errors': self.errors,
'c_rate': self.passes / (now - report_time),
'b_rate': self.bytes_processed / (now - report_time),
'total': (now - start_time),
'audit_time': audit_time,
'audit_rate': audit_time / (now - start_time)
})
report_time = now
total_corrupted += self.corrupted_chunks
total_orphans += self.orphan_chunks
total_faulty += self.faulty_chunks
total_errors += self.errors
self.passes = 0
self.corrupted_chunks = 0
self.orphan_chunks = 0
self.faulty_chunks = 0
self.errors = 0
self.bytes_processed = 0
self.last_reported = now
audit_time += (now - loop_time)
elapsed = (time.time() - start_time) or 0.000001
self.logger.info(
'%(elapsed).02f '
'%(corrupted)d '
'%(faulty)d '
'%(orphans)d '
'%(errors)d '
'%(chunk_rate).2f '
'%(bytes_rate).2f '
'%(audit_time).2f '
'%(audit_rate).2f' % {
'elapsed': elapsed,
'corrupted': total_corrupted + self.corrupted_chunks,
'faulty': total_faulty + self.faulty_chunks,
'orphans': total_orphans + self.orphan_chunks,
'errors': total_errors + self.errors,
'chunk_rate': self.total_chunks_processed / elapsed,
'bytes_rate': self.total_bytes_processed / elapsed,
'audit_time': audit_time,
'audit_rate': audit_time / elapsed
})
def index_pass(self):
def safe_update_index(path):
chunk_id = path.rsplit('/', 1)[-1]
if len(chunk_id) != STRLEN_CHUNKID:
self.logger.warn('WARN Not a chunk %s' % path)
return
for c in chunk_id:
if c not in hexdigits:
self.logger.warn('WARN Not a chunk %s' % path)
return
try:
self.update_index(path, chunk_id)
self.successes += 1
self.logger.debug('Updated %s', path)
except OioNetworkException as exc:
self.errors += 1
self.logger.warn('ERROR while updating %s: %s', path, exc)
except VolumeException as exc:
self.errors += 1
self.logger.error('Cannot index %s: %s', path, exc)
# All chunks of this volume are indexed in the same service,
# no need to try another chunk, it will generate the same
# error. Let the upper level retry later.
raise
except Exception:
self.errors += 1
self.logger.exception('ERROR while updating %s', path)
self.total_since_last_reported += 1
def report(tag):
total = self.errors + self.successes
now = time.time()
elapsed = (now - start_time) or 0.000001
self.logger.info(
'%(tag)s=%(current_time)s '
'elapsed=%(elapsed).02f '
'pass=%(pass)d '
'errors=%(errors)d '
'chunks=%(nb_chunks)d %(c_rate).2f/s' % {
'tag':
tag,
'current_time':
datetime.fromtimestamp(int(now)).isoformat(),
'pass':
self.passes,
'errors':
self.errors,
'nb_chunks':
total,
'c_rate':
self.total_since_last_reported /
(now - self.last_reported),
'elapsed':
elapsed
})
self.last_reported = now
self.total_since_last_reported = 0
start_time = time.time()
self.last_reported = start_time
self.errors = 0
self.successes = 0
paths = paths_gen(self.volume)
report('started')
for path in paths:
safe_update_index(path)
self.chunks_run_time = ratelimit(self.chunks_run_time,
self.max_chunks_per_second)
now = time.time()
if now - self.last_reported >= self.report_interval:
report('running')
report('ended')
def mover_pass(self):
self.namespace, self.address = check_volume(self.volume)
start_time = report_time = time.time()
total_errors = 0
mover_time = 0
paths = paths_gen(self.volume)
for path in paths:
loop_time = time.time()
now = time.time()
if now - self.last_usage_check >= self.usage_check_interval:
used, total = statfs(self.volume)
usage = (float(used) / total) * 100
if usage <= self.usage_target:
self.logger.info(
'current usage %.2f%%: target reached (%.2f%%)', usage,
self.usage_target)
self.last_usage_check = now
break
self.safe_chunk_move(path)
self.chunks_run_time = ratelimit(self.chunks_run_time,
self.max_chunks_per_second)
self.total_chunks_processed += 1
now = time.time()
if now - self.last_reported >= self.report_interval:
self.logger.info(
'%(start_time)s '
'%(passes)d '
'%(errors)d '
'%(c_rate).2f '
'%(b_rate).2f '
'%(total).2f '
'%(mover_time).2f'
'%(mover_rate).2f' % {
'start_time': time.ctime(report_time),
'passes': self.passes,
'errors': self.errors,
'c_rate': self.passes / (now - report_time),
'b_rate': self.bytes_processed / (now - report_time),
'total': (now - start_time),
'mover_time': mover_time,
'mover_rate': mover_time / (now - start_time)
})
report_time = now
total_errors += self.errors
self.passes = 0
self.bytes_processed = 0
self.last_reported = now
mover_time += (now - loop_time)
elapsed = (time.time() - start_time) or 0.000001
self.logger.info(
'%(elapsed).02f '
'%(errors)d '
'%(chunk_rate).2f '
'%(bytes_rate).2f '
'%(mover_time).2f '
'%(mover_rate).2f' % {
'elapsed': elapsed,
'errors': total_errors + self.errors,
'chunk_rate': self.total_chunks_processed / elapsed,
'bytes_rate': self.total_bytes_processed / elapsed,
'mover_time': mover_time,
'mover_rate': mover_time / elapsed
})
def dispatch_tasks(self, job_id, job_type, job_info, job):
job_config = job_info['config']
job_params = job_config['params']
tasks_per_second = job_config['tasks_per_second']
tasks_batch_size = job_config['tasks_batch_size']
last_task_id = job_info['tasks']['last_sent']
job_tasks = job.get_tasks(job_params, marker=last_task_id)
beanstalkd_workers = self.get_beanstalkd_workers()
tasks_run_time = 0
batch_per_second = tasks_per_second / float(
tasks_batch_size)
# The backend must have the tasks in order
# to know the last task sent
tasks = OrderedDict()
for task_id, task_payload in job_tasks:
if not self.running:
break
tasks[task_id] = task_payload
if len(tasks) < tasks_batch_size:
continue
tasks_run_time = ratelimit(
tasks_run_time, batch_per_second)
sent = self.dispatch_tasks_batch(
beanstalkd_workers,
job_id, job_type, job_config, tasks)
if sent:
job_status, exc = self.handle_backend_errors(
self.backend.update_tasks_sent, job_id, tasks.keys())
tasks.clear()
if exc is not None:
self.logger.warn(
'[job_id=%s] Job has not been updated '
'with the sent tasks: %s', job_id, exc)
break
if job_status == 'PAUSED':
self.logger.info('Job %s is paused', job_id)
return
if not self.running:
break
else:
sent = True
if tasks:
sent = self.dispatch_tasks_batch(
beanstalkd_workers,
job_id, job_type, job_config, tasks)
if sent:
job_status, exc = self.handle_backend_errors(
self.backend.update_tasks_sent, job_id, tasks.keys(),
all_tasks_sent=True)
if exc is None:
if job_status == 'FINISHED':
self.logger.info('Job %s is finished', job_id)
self.logger.info(
'Finished dispatching job (job_id=%s)', job_id)
return
else:
self.logger.warn(
'[job_id=%s] Job has not been updated '
'with the last sent tasks: %s', job_id, exc)
_, exc = self.handle_backend_errors(self.backend.free, job_id)
if exc is not None:
self.logger.warn(
'[job_id=%s] Job has not been freed: %s', job_id, exc)
def rebuilder_pass(self):
start_time = report_time = time.time()
rebuilder_time = 0
chunks = self._fetch_chunks()
for cid, content_id, chunk_id_or_pos, _ in chunks:
loop_time = time.time()
if self.dry_run:
self.dryrun_chunk_rebuild(cid, content_id, chunk_id_or_pos)
else:
self.safe_chunk_rebuild(cid, content_id, chunk_id_or_pos)
self.chunks_run_time = ratelimit(self.chunks_run_time,
self.max_chunks_per_second)
self.total_chunks_processed += 1
now = time.time()
if now - self.last_reported >= self.report_interval:
self.logger.info(
'RUN %(volume)s '
'started=%(start_time)s '
'passes=%(passes)d '
'errors=%(errors)d '
'chunks=%(nb_chunks)d %(c_rate).2f/s '
'bytes=%(nb_bytes)d %(b_rate).2fB/s '
'elapsed=%(total).2f '
'(rebuilder: %(success_rate).2f%%)' % {
'volume':
self.volume,
'start_time':
datetime.fromtimestamp(int(report_time)).isoformat(),
'passes':
self.passes,
'errors':
self.errors,
'nb_chunks':
self.total_chunks_processed,
'nb_bytes':
self.total_bytes_processed,
'c_rate':
self.passes / (now - report_time),
'b_rate':
self.bytes_processed / (now - report_time),
'total': (now - start_time),
'rebuilder_time':
rebuilder_time,
'success_rate':
100 * ((self.total_chunks_processed - self.errors) /
float(self.total_chunks_processed))
})
report_time = now
self.passes = 0
self.bytes_processed = 0
self.last_reported = now
rebuilder_time += (now - loop_time)
end_time = time.time()
elapsed = (end_time - start_time) or 0.000001
self.logger.info(
'DONE %(volume)s '
'started=%(start_time)s '
'ended=%(end_time)s '
'passes=%(passes)d '
'elapsed=%(elapsed).02f '
'errors=%(errors)d '
'chunks=%(nb_chunks)d %(c_rate).2f/s '
'bytes=%(nb_bytes)d %(b_rate).2fB/s '
'elapsed=%(rebuilder_time).2f '
'(rebuilder: %(success_rate).2f%%)' % {
'volume':
self.volume,
'start_time':
datetime.fromtimestamp(int(start_time)).isoformat(),
'end_time':
datetime.fromtimestamp(int(end_time)).isoformat(),
'passes':
self.passes,
'elapsed':
elapsed,
'errors':
self.errors,
'nb_chunks':
self.total_chunks_processed,
'nb_bytes':
self.total_bytes_processed,
'c_rate':
self.total_chunks_processed / elapsed,
'b_rate':
self.total_bytes_processed / elapsed,
'rebuilder_time':
rebuilder_time,
'success_rate':
100 * ((self.total_chunks_processed - self.errors) /
float(self.total_chunks_processed or 1))
})
def dispatch_tasks(self, job_id, job_type, job_info, job):
job_config = job_info['config']
job_params = job_config['params']
tasks_per_second = job_config['tasks_per_second']
tasks_batch_size = job_config['tasks_batch_size']
last_task_id = job_info['tasks']['last_sent']
job_tasks = job.get_tasks(job_params, marker=last_task_id)
beanstalkd_workers = self.get_beanstalkd_workers()
tasks_run_time = 0
batch_per_second = tasks_per_second / float(tasks_batch_size)
# The backend must have the tasks in order
# to know the last task sent
tasks = OrderedDict()
for task_id, task_payload in job_tasks:
if not self.running:
break
tasks[task_id] = task_payload
if len(tasks) < tasks_batch_size:
continue
tasks_run_time = ratelimit(tasks_run_time, batch_per_second)
# Make sure that the sent tasks will be saved
# before being processed
exc = None
sent = False
while not sent:
(job_status, old_last_sent), exc = self.handle_backend_errors(
self.backend.update_tasks_sent, job_id, tasks.keys())
if exc is not None:
self.logger.warn(
'[job_id=%s] Job could not update '
'the sent tasks: %s', job_id, exc)
break
sent = self.dispatch_tasks_batch(beanstalkd_workers, job_id,
job_type, job_config, tasks)
if not sent:
self.logger.warn(
'[job_id=%s] Job aborting the last sent tasks', job_id)
job_status, exc = self.handle_backend_errors(
self.backend.abort_tasks_sent, job_id, tasks.keys(),
old_last_sent)
if exc is not None:
self.logger.warn(
'[job_id=%s] Job could not abort '
'the last sent tasks: %s', job_id, exc)
break
if job_status == XcuteJobStatus.PAUSED:
self.logger.info('Job %s is paused', job_id)
return
if not self.running:
break
sleep(1)
if exc is not None and not self.running:
break
tasks.clear()
else:
# Make sure that the sent tasks will be saved
# before being processed
sent = False
while not sent:
(job_status, old_last_sent), exc = self.handle_backend_errors(
self.backend.update_tasks_sent,
job_id,
tasks.keys(),
all_tasks_sent=True)
if exc is not None:
self.logger.warn(
'[job_id=%s] Job could not update '
'the sent tasks: %s', job_id, exc)
break
if tasks:
sent = self.dispatch_tasks_batch(beanstalkd_workers,
job_id, job_type,
job_config, tasks)
else:
sent = True
if not sent:
self.logger.warn(
'[job_id=%s] Job aborting the last sent tasks', job_id)
job_status, exc = self.handle_backend_errors(
self.backend.abort_tasks_sent, job_id, tasks.keys(),
old_last_sent)
if exc is not None:
self.logger.warn(
'[job_id=%s] Job could not abort '
'the last sent tasks: %s', job_id, exc)
break
else:
if job_status == XcuteJobStatus.FINISHED:
self.logger.info('Job %s is finished', job_id)
self.logger.info('Finished dispatching job (job_id=%s)',
job_id)
return
if job_status == XcuteJobStatus.PAUSED:
self.logger.info('Job %s is paused', job_id)
return
if not self.running:
break
sleep(1)
self.logger.warn('[job_id=%s] Job was stopped before it was finished',
job_id)
_, exc = self.handle_backend_errors(self.backend.free, job_id)
if exc is not None:
self.logger.warn('[job_id=%s] Job has not been freed: %s', job_id,
exc)
def mover_pass(self, **kwargs):
start_time = report_time = time.time()
total_errors = 0
mover_time = 0
pool = GreenPool(self.concurrency)
paths = paths_gen(self.volume)
for path in paths:
loop_time = time.time()
now = time.time()
if now - self.last_usage_check >= self.usage_check_interval:
free_ratio = statfs(self.volume)
usage = (1 - float(free_ratio)) * 100
if usage <= self.usage_target:
self.logger.info(
'current usage %.2f%%: target reached (%.2f%%)', usage,
self.usage_target)
break
self.last_usage_check = now
# Spawn a chunk move task.
# The call will block if no green thread is available.
pool.spawn_n(self.safe_chunk_move, path)
self.chunks_run_time = ratelimit(self.chunks_run_time,
self.max_chunks_per_second)
self.total_chunks_processed += 1
now = time.time()
if now - self.last_reported >= self.report_interval:
self.logger.info(
'%(start_time)s '
'%(passes)d '
'%(errors)d '
'%(c_rate).2f '
'%(b_rate).2f '
'%(total).2f '
'%(mover_time).2f'
'%(mover_rate).2f' % {
'start_time': time.ctime(report_time),
'passes': self.passes,
'errors': self.errors,
'c_rate': self.passes / (now - report_time),
'b_rate': self.bytes_processed / (now - report_time),
'total': (now - start_time),
'mover_time': mover_time,
'mover_rate': mover_time / (now - start_time)
})
report_time = now
total_errors += self.errors
self.passes = 0
self.bytes_processed = 0
self.last_reported = now
mover_time += (now - loop_time)
if self.limit != 0 and self.total_chunks_processed >= self.limit:
break
pool.waitall()
elapsed = (time.time() - start_time) or 0.000001
self.logger.info(
'%(elapsed).02f '
'%(errors)d '
'%(chunk_rate).2f '
'%(bytes_rate).2f '
'%(mover_time).2f '
'%(mover_rate).2f' % {
'elapsed': elapsed,
'errors': total_errors + self.errors,
'chunk_rate': self.total_chunks_processed / elapsed,
'bytes_rate': self.total_bytes_processed / elapsed,
'mover_time': mover_time,
'mover_rate': mover_time / elapsed
})
