import argparse
import arvados
+import arvados.collection
import base64
import datetime
import errno
args = arg_parser.parse_args(arguments)
if len(args.paths) == 0:
- args.paths += ['/dev/stdin']
+ args.paths = ['-']
+
+ args.paths = map(lambda x: "-" if x == "/dev/stdin" else x, args.paths)
if len(args.paths) != 1 or os.path.isdir(args.paths[0]):
if args.filename:
args.progress = True
if args.paths == ['-']:
- args.paths = ['/dev/stdin']
+ args.resume = False
if not args.filename:
- args.filename = '-'
+ args.filename = 'stdin'
return args
self.cache_file.seek(0)
return json.load(self.cache_file)
+ def check_cache(self, api_client=None, num_retries=0):
+ try:
+ state = self.load()
+ locator = None
+ try:
+ if "_finished_streams" in state and len(state["_finished_streams"]) > 0:
+ locator = state["_finished_streams"][0][1][0]
+ elif "_current_stream_locators" in state and len(state["_current_stream_locators"]) > 0:
+ locator = state["_current_stream_locators"][0]
+ if locator is not None:
+ kc = arvados.keep.KeepClient(api_client=api_client)
+ kc.head(locator, num_retries=num_retries)
+ except Exception as e:
+ self.restart()
+ except (ValueError):
+ pass
+
def save(self, data):
try:
new_cache_fd, new_cache_name = tempfile.mkstemp(
replication=replication)
except (TypeError, ValueError,
arvados.errors.StaleWriterStateError) as error:
- return cls(cache, reporter, bytes_expected, num_retries=num_retries)
+ return cls(cache, reporter, bytes_expected,
+ num_retries=num_retries,
+ replication=replication)
else:
return writer
def flush_data(self):
start_buffer_len = self._data_buffer_len
- start_block_count = self.bytes_written / self.KEEP_BLOCK_SIZE
+ start_block_count = self.bytes_written / arvados.config.KEEP_BLOCK_SIZE
super(ArvPutCollectionWriter, self).flush_data()
if self._data_buffer_len < start_buffer_len: # We actually PUT data.
self.bytes_written += (start_buffer_len - self._data_buffer_len)
self.report_progress()
- if (self.bytes_written / self.KEEP_BLOCK_SIZE) > start_block_count:
+ if (self.bytes_written / arvados.config.KEEP_BLOCK_SIZE) > start_block_count:
self.cache_state()
def _record_new_input(self, input_type, source_name, dest_name):
print >>stderr, error
sys.exit(1)
- # Apply default replication, if none specified. TODO (#3410): Use
- # default replication given by discovery document.
- if args.replication <= 0:
- args.replication = 2
+ # write_copies diverges from args.replication here.
+ # args.replication is how many copies we will instruct Arvados to
+ # maintain (by passing it in collections().create()) after all
+ # data is written -- and if None was given, we'll use None there.
+ # Meanwhile, write_copies is how many copies of each data block we
+ # write to Keep, which has to be a number.
+ #
+ # If we simply changed args.replication from None to a default
+ # here, we'd end up erroneously passing the default replication
+ # level (instead of None) to collections().create().
+ write_copies = (args.replication or
+ api_client._rootDesc.get('defaultCollectionReplication', 2))
if args.progress:
reporter = progress_writer(human_progress)
if args.resume:
try:
resume_cache = ResumeCache(ResumeCache.make_path(args))
+ resume_cache.check_cache(api_client=api_client, num_retries=args.retries)
except (IOError, OSError, ValueError):
pass # Couldn't open cache directory/file. Continue without it.
except ResumeCacheConflict:
writer = ArvPutCollectionWriter(
resume_cache, reporter, bytes_expected,
num_retries=args.retries,
- replication=args.replication)
+ replication=write_copies)
else:
writer = ArvPutCollectionWriter.from_cache(
resume_cache, reporter, bytes_expected,
num_retries=args.retries,
- replication=args.replication)
+ replication=write_copies)
# Install our signal handler for each code in CAUGHT_SIGNALS, and save
# the originals.
writer.report_progress()
writer.do_queued_work() # Do work resumed from cache.
for path in args.paths: # Copy file data to Keep.
- if os.path.isdir(path):
+ if path == '-':
+ writer.start_new_stream()
+ writer.start_new_file(args.filename)
+ r = sys.stdin.read(64*1024)
+ while r:
+ # Need to bypass _queued_file check in ResumableCollectionWriter.write() to get
+ # CollectionWriter.write().
+ super(arvados.collection.ResumableCollectionWriter, writer).write(r)
+ r = sys.stdin.read(64*1024)
+ elif os.path.isdir(path):
writer.write_directory_tree(
path, max_manifest_depth=args.max_manifest_depth)
else:
if args.progress: # Print newline to split stderr from stdout for humans.
print >>stderr
+ output = None
if args.stream:
output = writer.manifest_text()
if args.normalize:
- output = CollectionReader(output).manifest_text(normalize=True)
+ output = arvados.collection.CollectionReader(output).manifest_text(normalize=True)
elif args.raw:
output = ','.join(writer.data_locators())
else:
try:
manifest_text = writer.manifest_text()
if args.normalize:
- manifest_text = CollectionReader(manifest_text).manifest_text(normalize=True)
+ manifest_text = arvados.collection.CollectionReader(manifest_text).manifest_text(normalize=True)
replication_attr = 'replication_desired'
if api_client._schema.schemas['Collection']['properties'].get(replication_attr, None) is None:
- # API calls it 'redundancy' until #3410.
+ # API called it 'redundancy' before #3410.
replication_attr = 'redundancy'
# Register the resulting collection in Arvados.
collection = api_client.collections().create(
status = 1
# Print the locator (uuid) of the new collection.
- stdout.write(output)
- if not output.endswith('\n'):
- stdout.write('\n')
+ if output is None:
+ status = status or 1
+ else:
+ stdout.write(output)
+ if not output.endswith('\n'):
+ stdout.write('\n')
for sigcode, orig_handler in orig_signal_handlers.items():
signal.signal(sigcode, orig_handler)
projects / arvados.git / blobdiff