def purge_files(min_size, purgable_files):
# we try to keep the cache as full as possible and defer deleting
# any files until we can fullfil an allocation request completely
# - since no request will start opportunisticly anyway. Also only
# delete enough to ensure min_size bytes.
bytes_freed = 0
bytes_freeable = sum(sf.size for sf in purgable_files)
requests_to_delete = set()
if min_size <= bytes_freeable:
for sf in purgable_files:
if bytes_freed >= min_size:
break
logger.debug('deleting %s' % sf)
# raises OSError on any error except if not existing:
if sf.staging_task:
util.unlink(sf.staging_task.path_out)
is_deleted = util.unlink(sf.path_staged())
if not is_deleted:
logger.warn('tried purging staged file %s - but it is already '
'gone' % sf)
sf.state = 'offline'
bytes_freed += sf.size
requests_to_delete.update(set(sf.requests))
return bytes_freed, requests_to_delete
def stage_file(self, file_name, target_dir, path_out):
try:
end_target = os.path.join(target_dir, file_name)
tmp_target = os.path.join(target_dir, uuid.uuid4().get_hex())
logger.info('staging %s' % end_target)
logger.debug('tmp_target: %s' % tmp_target)
mars_request = create_mars_request(verb='RETRIEVE',
file_name=file_name,
target=tmp_target)
logger.debug('mars_request: %s' % mars_request)
with open(path_out, 'w') as f:
for rc,fd,l in util.exec_proc([ 'mars' ], logger, stdin=str(mars_request)):
if fd is not None and l is not None:
if fd == 1:
logger.debug('fd=%s, l=%s' % (fd, l.strip() if l else l))
else:
logger.warning('fd=%s, l=%s' % (fd, l.strip() if l else l))
f.write(l)
f.flush()
f.flush()
os.fdatasync(f.fileno())
f.close()
if rc != 0:
logger.error('failed to stage %s, rc = %d' % (file_name, rc))
logger.error('MARS request was: %s' % mars_request)
logger.debug('removing temp file %s' % tmp_target)
util.unlink(tmp_target) # FIXME: use try...finally
raise TaskFailure('mars returned %d' % rc)
logger.debug('moving temp file %s -> %s' % (tmp_target, end_target))
os.rename(tmp_target, end_target)
logger.info('%s is staged online' % end_target)
logger.debug('=> invoking scheduler')
tasks.scheduler.join_staging_task.delay(stage_file.request.id)
except Exception, e:
logger.error('stage_file(file_name=%s, target_dir=%s, path_out=%s) '
'unexpectedly failed (task id %s): %s, retrying in '
'60 s...' %
(file_name, target_dir, path_out, stage_file.request.id,
str(e)))
raise self.retry(exc=e, countdown=60)
sf.staging_task = None
assert sf.staging_task is None, sf
logger.debug('db commit')
tasks.session.commit()
for r in dispatched_requests(sf):
logger.info('request %s failed' % r.uuid)
fail_request(r, 'Staging of %s failed: %s' % (sf, str(e)))
return # we deliberately only handle exceptions from the staging task, any other exception will propagate
logger.debug('updating file size if necessary')
update_size_if_different(sf)
logger.info('%s is online' % sf.name)
sf.state = 'online'
logger.debug('deregistering %s from %s' % (sf.staging_task, sf))
tasks.session.delete(sf.staging_task)
util.unlink(sf.staging_task.path_out)
sf.staging_task = None
assert sf.staging_task is None, sf
logger.debug('db commit')
tasks.session.commit() # verify that we really need this - but it's important that the query below includes r
for r in finishable_requests(sf):
logger.info('request %s finished' % r.uuid)
finish_request(r)
@tasks.cel.task(bind=True, acks_late=True, ignore_results=True)
def join_staging_task(self, task_id):
try:
staging_task = get_task(task_id)
logger.debug('staging task for %s is: %s' % (task_id, staging_task))
# BUG: FIXME: why is this a list and not a file, we do use uselist=False