-#!/usr/bin/env python
-
-# TODO:
-# --md5sum - display md5 of each file as read from disk
-
+from __future__ import division
+from future.utils import listitems, listvalues
+from builtins import str
+from builtins import object
import argparse
import arvados
import arvados.collection
import tempfile
import threading
import time
+import traceback
+
from apiclient import errors as apiclient_errors
from arvados._version import __version__
block. Default is to use the server-provided default (if any) or 2.
""")
+upload_opts.add_argument('--threads', type=int, metavar='N', default=None,
+ help="""
+Set the number of upload threads to be used. Take into account that
+using lots of threads will increase the RAM requirements. Default is
+to use 2 threads.
+On high latency installations, using a greater number will improve
+overall throughput.
+""")
+
run_opts = argparse.ArgumentParser(add_help=False)
run_opts.add_argument('--project-uuid', metavar='UUID', help="""
if len(args.paths) == 0:
args.paths = ['-']
- args.paths = map(lambda x: "-" if x == "/dev/stdin" else x, args.paths)
+ args.paths = ["-" if x == "/dev/stdin" else x for x in args.paths]
if len(args.paths) != 1 or os.path.isdir(args.paths[0]):
if args.filename:
@classmethod
def make_path(cls, args):
md5 = hashlib.md5()
- md5.update(arvados.config.get('ARVADOS_API_HOST', '!nohost'))
+ md5.update(arvados.config.get('ARVADOS_API_HOST', '!nohost').encode())
realpaths = sorted(os.path.realpath(path) for path in args.paths)
- md5.update('\0'.join(realpaths))
+ md5.update(b'\0'.join([p.encode() for p in realpaths]))
if any(os.path.isdir(path) for path in realpaths):
- md5.update("-1")
+ md5.update(b'-1')
elif args.filename:
- md5.update(args.filename)
+ md5.update(args.filename.encode())
return os.path.join(
arv_cmd.make_home_conf_dir(cls.CACHE_DIR, 0o700, 'raise'),
md5.hexdigest())
def __init__(self, paths, resume=True, use_cache=True, reporter=None,
bytes_expected=None, name=None, owner_uuid=None,
- ensure_unique_name=False, num_retries=None, replication_desired=None,
- filename=None, update_time=20.0, update_collection=None,
+ ensure_unique_name=False, num_retries=None,
+ put_threads=None, replication_desired=None,
+ filename=None, update_time=60.0, update_collection=None,
logger=logging.getLogger('arvados.arv_put'), dry_run=False):
self.paths = paths
self.resume = resume
self.ensure_unique_name = ensure_unique_name
self.num_retries = num_retries
self.replication_desired = replication_desired
+ self.put_threads = put_threads
self.filename = filename
self._state_lock = threading.Lock()
self._state = None # Previous run state (file list & manifest)
self._collection_lock = threading.Lock()
self._remote_collection = None # Collection being updated (if asked)
self._local_collection = None # Collection from previous run manifest
- self._file_paths = [] # Files to be updated in remote collection
+ self._file_paths = set() # Files to be updated in remote collection
self._stop_checkpointer = threading.Event()
self._checkpointer = threading.Thread(target=self._update_task)
self._checkpointer.daemon = True
self._update_task_time = update_time # How many seconds wait between update runs
self._files_to_upload = FileUploadList(dry_run=dry_run)
+ self._upload_started = False
self.logger = logger
self.dry_run = dry_run
+ self._checkpoint_before_quit = True
if not self.use_cache and self.resume:
raise ArvPutArgumentConflict('resume cannot be True when use_cache is False')
# report initial progress.
self._update()
# Actual file upload
+ self._upload_started = True # Used by the update thread to start checkpointing
self._upload_files()
+ except (SystemExit, Exception) as e:
+ self._checkpoint_before_quit = False
+ # Log stack trace only when Ctrl-C isn't pressed (SIGINT)
+ # Note: We're expecting SystemExit instead of KeyboardInterrupt because
+ # we have a custom signal handler in place that raises SystemExit with
+ # the catched signal's code.
+ if not isinstance(e, SystemExit) or e.code != -2:
+ self.logger.warning("Abnormal termination:\n{}".format(
+ traceback.format_exc()))
+ raise
finally:
if not self.dry_run:
# Stop the thread before doing anything else
self._stop_checkpointer.set()
self._checkpointer.join()
- # Commit all pending blocks & one last _update()
- self._local_collection.manifest_text()
- self._update(final=True)
- if save_collection:
- self.save_collection()
+ if self._checkpoint_before_quit:
+ # Commit all pending blocks & one last _update()
+ self._local_collection.manifest_text()
+ self._update(final=True)
+ if save_collection:
+ self.save_collection()
if self.use_cache:
self._cache_file.close()
Recursively get the total size of the collection
"""
size = 0
- for item in collection.values():
+ for item in listvalues(collection):
if isinstance(item, arvados.collection.Collection) or isinstance(item, arvados.collection.Subcollection):
size += self._collection_size(item)
else:
Periodically called support task. File uploading is
asynchronous so we poll status from the collection.
"""
- while not self._stop_checkpointer.wait(self._update_task_time):
+ while not self._stop_checkpointer.wait(1 if not self._upload_started else self._update_task_time):
self._update()
def _update(self, final=False):
"""
Update cached manifest text and report progress.
"""
- with self._collection_lock:
- self.bytes_written = self._collection_size(self._local_collection)
- if self.use_cache:
- # Update cache
- with self._state_lock:
+ if self._upload_started:
+ with self._collection_lock:
+ self.bytes_written = self._collection_size(self._local_collection)
+ if self.use_cache:
if final:
- self._state['manifest'] = self._local_collection.manifest_text()
+ manifest = self._local_collection.manifest_text()
else:
# Get the manifest text without comitting pending blocks
- self._state['manifest'] = self._local_collection._get_manifest_text(".", strip=False, normalize=False, only_committed=True)
- self._save_state()
+ manifest = self._local_collection.manifest_text(strip=False,
+ normalize=False,
+ only_committed=True)
+ # Update cache
+ with self._state_lock:
+ self._state['manifest'] = manifest
+ if self.use_cache:
+ try:
+ self._save_state()
+ except Exception as e:
+ self.logger.error("Unexpected error trying to save cache file: {}".format(e))
+ else:
+ self.bytes_written = self.bytes_skipped
# Call the reporter, if any
self.report_progress()
self.reporter(self.bytes_written, self.bytes_expected)
def _write_stdin(self, filename):
- output = self._local_collection.open(filename, 'w')
+ output = self._local_collection.open(filename, 'wb')
self._write(sys.stdin, output)
output.close()
should_upload = False
new_file_in_cache = False
# Record file path for updating the remote collection before exiting
- self._file_paths.append(filename)
+ self._file_paths.add(filename)
with self._state_lock:
# If no previous cached data on this file, store it for an eventual
def _upload_files(self):
for source, resume_offset, filename in self._files_to_upload:
- with open(source, 'r') as source_fd:
+ with open(source, 'rb') as source_fd:
with self._state_lock:
self._state['files'][source]['mtime'] = os.path.getmtime(source)
self._state['files'][source]['size'] = os.path.getsize(source)
if resume_offset > 0:
# Start upload where we left off
- output = self._local_collection.open(filename, 'a')
+ output = self._local_collection.open(filename, 'ab')
source_fd.seek(resume_offset)
else:
# Start from scratch
- output = self._local_collection.open(filename, 'w')
+ output = self._local_collection.open(filename, 'wb')
self._write(source_fd, output)
output.close(flush=False)
if self.use_cache:
# Set up cache file name from input paths.
md5 = hashlib.md5()
- md5.update(arvados.config.get('ARVADOS_API_HOST', '!nohost'))
+ md5.update(arvados.config.get('ARVADOS_API_HOST', '!nohost').encode())
realpaths = sorted(os.path.realpath(path) for path in self.paths)
- md5.update('\0'.join(realpaths))
+ md5.update(b'\0'.join([p.encode() for p in realpaths]))
if self.filename:
- md5.update(self.filename)
+ md5.update(self.filename.encode())
cache_filename = md5.hexdigest()
cache_filepath = os.path.join(
arv_cmd.make_home_conf_dir(self.CACHE_DIR, 0o700, 'raise'),
cache_filename)
- if self.resume:
+ if self.resume and os.path.exists(cache_filepath):
+ self.logger.info("Resuming upload from cache file {}".format(cache_filepath))
self._cache_file = open(cache_filepath, 'a+')
else:
# --no-resume means start with a empty cache file.
+ self.logger.info("Creating new cache file at {}".format(cache_filepath))
self._cache_file = open(cache_filepath, 'w+')
self._cache_filename = self._cache_file.name
self._lock_file(self._cache_file)
# Cache file empty, set up new cache
self._state = copy.deepcopy(self.EMPTY_STATE)
else:
+ self.logger.info("No cache usage requested for this run.")
# No cache file, set empty state
self._state = copy.deepcopy(self.EMPTY_STATE)
# Load the previous manifest so we can check if files were modified remotely.
- self._local_collection = arvados.collection.Collection(self._state['manifest'], replication_desired=self.replication_desired)
+ self._local_collection = arvados.collection.Collection(self._state['manifest'], replication_desired=self.replication_desired, put_threads=self.put_threads)
def collection_file_paths(self, col, path_prefix='.'):
"""Return a list of file paths by recursively go through the entire collection `col`"""
file_paths = []
- for name, item in col.items():
+ for name, item in listitems(col):
if isinstance(item, arvados.arvfile.ArvadosFile):
file_paths.append(os.path.join(path_prefix, name))
elif isinstance(item, arvados.collection.Subcollection):
"""
Atomically save current state into cache.
"""
+ with self._state_lock:
+ # We're not using copy.deepcopy() here because it's a lot slower
+ # than json.dumps(), and we're already needing JSON format to be
+ # saved on disk.
+ state = json.dumps(self._state)
try:
- with self._state_lock:
- state = copy.deepcopy(self._state)
- new_cache_fd, new_cache_name = tempfile.mkstemp(
- dir=os.path.dirname(self._cache_filename))
- self._lock_file(new_cache_fd)
- new_cache = os.fdopen(new_cache_fd, 'r+')
- json.dump(state, new_cache)
+ new_cache = tempfile.NamedTemporaryFile(
+ mode='w+',
+ dir=os.path.dirname(self._cache_filename), delete=False)
+ self._lock_file(new_cache)
+ new_cache.write(state)
new_cache.flush()
os.fsync(new_cache)
- os.rename(new_cache_name, self._cache_filename)
+ os.rename(new_cache.name, self._cache_filename)
except (IOError, OSError, ResumeCacheConflict) as error:
self.logger.error("There was a problem while saving the cache file: {}".format(error))
try:
return self._my_collection().manifest_locator()
def portable_data_hash(self):
- return self._my_collection().portable_data_hash()
+ pdh = self._my_collection().portable_data_hash()
+ m = self._my_collection().stripped_manifest().encode()
+ local_pdh = '{}+{}'.format(hashlib.md5(m).hexdigest(), len(m))
+ if pdh != local_pdh:
+ logger.warning("\n".join([
+ "arv-put: API server provided PDH differs from local manifest.",
+ " This should not happen; showing API server version."]))
+ return pdh
def manifest_text(self, stream_name=".", strip=False, normalize=False):
return self._my_collection().manifest_text(stream_name, strip, normalize)
locators.append(loc)
return locators
elif isinstance(item, arvados.collection.Collection):
- l = [self._datablocks_on_item(x) for x in item.values()]
+ l = [self._datablocks_on_item(x) for x in listvalues(item)]
# Fast list flattener method taken from:
# http://stackoverflow.com/questions/952914/making-a-flat-list-out-of-list-of-lists-in-python
return [loc for sublist in l for loc in sublist]
global api_client
logger = logging.getLogger('arvados.arv_put')
+ logger.setLevel(logging.INFO)
args = parse_arguments(arguments)
status = 0
if api_client is None:
else:
reporter = None
+ # If this is used by a human, and there's at least one directory to be
+ # uploaded, the expected bytes calculation can take a moment.
+ if args.progress and any([os.path.isdir(f) for f in args.paths]):
+ logger.info("Calculating upload size, this could take some time...")
bytes_expected = expected_bytes_for(args.paths)
try:
bytes_expected = bytes_expected,
num_retries = args.retries,
replication_desired = args.replication,
+ put_threads = args.threads,
name = collection_name,
owner_uuid = project_uuid,
ensure_unique_name = True,
if not output.endswith('\n'):
stdout.write('\n')
- for sigcode, orig_handler in orig_signal_handlers.items():
+ for sigcode, orig_handler in listitems(orig_signal_handlers):
signal.signal(sigcode, orig_handler)
if status != 0: