1 # Copyright (C) The Arvados Authors. All rights reserved.
3 # SPDX-License-Identifier: Apache-2.0
5 from __future__ import division
6 from future.utils import listitems, listvalues
7 from builtins import str
8 from builtins import object
11 import arvados.collection
33 from apiclient import errors as apiclient_errors
34 from arvados._version import __version__
35 from arvados.util import keep_locator_pattern
37 import arvados.commands._util as arv_cmd
41 upload_opts = argparse.ArgumentParser(add_help=False)
43 upload_opts.add_argument('--version', action='version',
44 version="%s %s" % (sys.argv[0], __version__),
45 help='Print version and exit.')
46 upload_opts.add_argument('paths', metavar='path', type=str, nargs='*',
48 Local file or directory. If path is a directory reference with a trailing
49 slash, then just upload the directory's contents; otherwise upload the
50 directory itself. Default: read from standard input.
53 _group = upload_opts.add_mutually_exclusive_group()
55 _group.add_argument('--max-manifest-depth', type=int, metavar='N',
56 default=-1, help=argparse.SUPPRESS)
58 _group.add_argument('--normalize', action='store_true',
60 Normalize the manifest by re-ordering files and streams after writing
64 _group.add_argument('--dry-run', action='store_true', default=False,
66 Don't actually upload files, but only check if any file should be
67 uploaded. Exit with code=2 when files are pending for upload.
70 _group = upload_opts.add_mutually_exclusive_group()
72 _group.add_argument('--as-stream', action='store_true', dest='stream',
77 _group.add_argument('--stream', action='store_true',
79 Store the file content and display the resulting manifest on
80 stdout. Do not save a Collection object in Arvados.
83 _group.add_argument('--as-manifest', action='store_true', dest='manifest',
85 Synonym for --manifest.
88 _group.add_argument('--in-manifest', action='store_true', dest='manifest',
90 Synonym for --manifest.
93 _group.add_argument('--manifest', action='store_true',
95 Store the file data and resulting manifest in Keep, save a Collection
96 object in Arvados, and display the manifest locator (Collection uuid)
97 on stdout. This is the default behavior.
100 _group.add_argument('--as-raw', action='store_true', dest='raw',
105 _group.add_argument('--raw', action='store_true',
107 Store the file content and display the data block locators on stdout,
108 separated by commas, with a trailing newline. Do not store a
112 upload_opts.add_argument('--update-collection', type=str, default=None,
113 dest='update_collection', metavar="UUID", help="""
114 Update an existing collection identified by the given Arvados collection
115 UUID. All new local files will be uploaded.
118 upload_opts.add_argument('--use-filename', type=str, default=None,
119 dest='filename', help="""
120 Synonym for --filename.
123 upload_opts.add_argument('--filename', type=str, default=None,
125 Use the given filename in the manifest, instead of the name of the
126 local file. This is useful when "-" or "/dev/stdin" is given as an
127 input file. It can be used only if there is exactly one path given and
128 it is not a directory. Implies --manifest.
131 upload_opts.add_argument('--portable-data-hash', action='store_true',
133 Print the portable data hash instead of the Arvados UUID for the collection
134 created by the upload.
137 upload_opts.add_argument('--replication', type=int, metavar='N', default=None,
139 Set the replication level for the new collection: how many different
140 physical storage devices (e.g., disks) should have a copy of each data
141 block. Default is to use the server-provided default (if any) or 2.
144 upload_opts.add_argument('--storage-classes', help="""
145 Specify comma separated list of storage classes to be used when saving data to Keep.
148 upload_opts.add_argument('--threads', type=int, metavar='N', default=None,
150 Set the number of upload threads to be used. Take into account that
151 using lots of threads will increase the RAM requirements. Default is
153 On high latency installations, using a greater number will improve
157 upload_opts.add_argument('--exclude', metavar='PATTERN', default=[],
158 action='append', help="""
159 Exclude files and directories whose names match the given glob pattern. When
160 using a path-like pattern like 'subdir/*.txt', all text files inside 'subdir'
161 directory, relative to the provided input dirs will be excluded.
162 When using a filename pattern like '*.txt', any text file will be excluded
163 no matter where it is placed.
164 For the special case of needing to exclude only files or dirs directly below
165 the given input directory, you can use a pattern like './exclude_this.gif'.
166 You can specify multiple patterns by using this argument more than once.
169 _group = upload_opts.add_mutually_exclusive_group()
170 _group.add_argument('--follow-links', action='store_true', default=True,
171 dest='follow_links', help="""
172 Follow file and directory symlinks (default).
174 _group.add_argument('--no-follow-links', action='store_false', dest='follow_links',
176 Ignore file and directory symlinks. Even paths given explicitly on the
177 command line will be skipped if they are symlinks.
181 run_opts = argparse.ArgumentParser(add_help=False)
183 run_opts.add_argument('--project-uuid', metavar='UUID', help="""
184 Store the collection in the specified project, instead of your Home
188 run_opts.add_argument('--name', help="""
189 Save the collection with the specified name.
192 _group = run_opts.add_mutually_exclusive_group()
193 _group.add_argument('--progress', action='store_true',
195 Display human-readable progress on stderr (bytes and, if possible,
196 percentage of total data size). This is the default behavior when
200 _group.add_argument('--no-progress', action='store_true',
202 Do not display human-readable progress on stderr, even if stderr is a
206 _group.add_argument('--batch-progress', action='store_true',
208 Display machine-readable progress on stderr (bytes and, if known,
212 run_opts.add_argument('--silent', action='store_true',
214 Do not print any debug messages to console. (Any error messages will
218 run_opts.add_argument('--batch', action='store_true', default=False,
220 Retries with '--no-resume --no-cache' if cached state contains invalid/expired
224 _group = run_opts.add_mutually_exclusive_group()
225 _group.add_argument('--resume', action='store_true', default=True,
227 Continue interrupted uploads from cached state (default).
229 _group.add_argument('--no-resume', action='store_false', dest='resume',
231 Do not continue interrupted uploads from cached state.
234 _group = run_opts.add_mutually_exclusive_group()
235 _group.add_argument('--cache', action='store_true', dest='use_cache', default=True,
237 Save upload state in a cache file for resuming (default).
239 _group.add_argument('--no-cache', action='store_false', dest='use_cache',
241 Do not save upload state in a cache file for resuming.
244 _group = upload_opts.add_mutually_exclusive_group()
245 _group.add_argument('--trash-at', metavar='YYYY-MM-DDTHH:MM', default=None,
247 Set the trash date of the resulting collection to an absolute date in the future.
248 The accepted format is defined by the ISO 8601 standard. Examples: 20090103, 2009-01-03, 20090103T181505, 2009-01-03T18:15:05.\n
249 Timezone information can be added. If not, the provided date/time is assumed as being in the local system's timezone.
251 _group.add_argument('--trash-after', type=int, metavar='DAYS', default=None,
253 Set the trash date of the resulting collection to an amount of days from the
254 date/time that the upload process finishes.
257 arg_parser = argparse.ArgumentParser(
258 description='Copy data from the local filesystem to Keep.',
259 parents=[upload_opts, run_opts, arv_cmd.retry_opt])
261 def parse_arguments(arguments):
262 args = arg_parser.parse_args(arguments)
264 if len(args.paths) == 0:
267 args.paths = ["-" if x == "/dev/stdin" else x for x in args.paths]
269 if args.filename and (len(args.paths) != 1 or os.path.isdir(args.paths[0])):
271 --filename argument cannot be used when storing a directory or
275 # Turn on --progress by default if stderr is a tty.
276 if (not (args.batch_progress or args.no_progress or args.silent)
277 and os.isatty(sys.stderr.fileno())):
280 # Turn off --resume (default) if --no-cache is used.
281 if not args.use_cache:
284 if args.paths == ['-']:
285 if args.update_collection:
287 --update-collection cannot be used when reading from stdin.
290 args.use_cache = False
291 if not args.filename:
292 args.filename = 'stdin'
294 # Remove possible duplicated patterns
295 if len(args.exclude) > 0:
296 args.exclude = list(set(args.exclude))
301 class PathDoesNotExistError(Exception):
305 class CollectionUpdateError(Exception):
309 class ResumeCacheConflict(Exception):
313 class ResumeCacheInvalidError(Exception):
316 class ArvPutArgumentConflict(Exception):
320 class ArvPutUploadIsPending(Exception):
324 class ArvPutUploadNotPending(Exception):
328 class FileUploadList(list):
329 def __init__(self, dry_run=False):
331 self.dry_run = dry_run
333 def append(self, other):
335 raise ArvPutUploadIsPending()
336 super(FileUploadList, self).append(other)
339 # Appends the X-Request-Id to the log message when log level is ERROR or DEBUG
340 class ArvPutLogFormatter(logging.Formatter):
341 std_fmtr = logging.Formatter(arvados.log_format, arvados.log_date_format)
343 request_id_informed = False
345 def __init__(self, request_id):
346 self.err_fmtr = logging.Formatter(
347 arvados.log_format+' (X-Request-Id: {})'.format(request_id),
348 arvados.log_date_format)
350 def format(self, record):
351 if (not self.request_id_informed) and (record.levelno in (logging.DEBUG, logging.ERROR)):
352 self.request_id_informed = True
353 return self.err_fmtr.format(record)
354 return self.std_fmtr.format(record)
357 class ResumeCache(object):
358 CACHE_DIR = '.cache/arvados/arv-put'
360 def __init__(self, file_spec):
361 self.cache_file = open(file_spec, 'a+')
362 self._lock_file(self.cache_file)
363 self.filename = self.cache_file.name
366 def make_path(cls, args):
368 md5.update(arvados.config.get('ARVADOS_API_HOST', '!nohost').encode())
369 realpaths = sorted(os.path.realpath(path) for path in args.paths)
370 md5.update(b'\0'.join([p.encode() for p in realpaths]))
371 if any(os.path.isdir(path) for path in realpaths):
374 md5.update(args.filename.encode())
376 arv_cmd.make_home_conf_dir(cls.CACHE_DIR, 0o700, 'raise'),
379 def _lock_file(self, fileobj):
381 fcntl.flock(fileobj, fcntl.LOCK_EX | fcntl.LOCK_NB)
383 raise ResumeCacheConflict(u"{} locked".format(fileobj.name))
386 self.cache_file.seek(0)
387 return json.load(self.cache_file)
389 def check_cache(self, api_client=None, num_retries=0):
394 if "_finished_streams" in state and len(state["_finished_streams"]) > 0:
395 locator = state["_finished_streams"][0][1][0]
396 elif "_current_stream_locators" in state and len(state["_current_stream_locators"]) > 0:
397 locator = state["_current_stream_locators"][0]
398 if locator is not None:
399 kc = arvados.keep.KeepClient(api_client=api_client)
400 kc.head(locator, num_retries=num_retries)
401 except Exception as e:
406 def save(self, data):
408 new_cache_fd, new_cache_name = tempfile.mkstemp(
409 dir=os.path.dirname(self.filename))
410 self._lock_file(new_cache_fd)
411 new_cache = os.fdopen(new_cache_fd, 'r+')
412 json.dump(data, new_cache)
413 os.rename(new_cache_name, self.filename)
414 except (IOError, OSError, ResumeCacheConflict):
416 os.unlink(new_cache_name)
417 except NameError: # mkstemp failed.
420 self.cache_file.close()
421 self.cache_file = new_cache
424 self.cache_file.close()
428 os.unlink(self.filename)
429 except OSError as error:
430 if error.errno != errno.ENOENT: # That's what we wanted anyway.
436 self.__init__(self.filename)
439 class ArvPutUploadJob(object):
440 CACHE_DIR = '.cache/arvados/arv-put'
442 'manifest' : None, # Last saved manifest checkpoint
443 'files' : {} # Previous run file list: {path : {size, mtime}}
446 def __init__(self, paths, resume=True, use_cache=True, reporter=None,
447 name=None, owner_uuid=None, api_client=None,
448 ensure_unique_name=False, num_retries=None,
449 put_threads=None, replication_desired=None, filename=None,
450 update_time=60.0, update_collection=None, storage_classes=None,
451 logger=logging.getLogger('arvados.arv_put'), dry_run=False,
452 follow_links=True, exclude_paths=[], exclude_names=None,
456 self.use_cache = use_cache
458 self.reporter = reporter
459 # This will set to 0 before start counting, if no special files are going
461 self.bytes_expected = None
462 self.bytes_written = 0
463 self.bytes_skipped = 0
465 self.owner_uuid = owner_uuid
466 self.ensure_unique_name = ensure_unique_name
467 self.num_retries = num_retries
468 self.replication_desired = replication_desired
469 self.put_threads = put_threads
470 self.filename = filename
471 self.storage_classes = storage_classes
472 self._api_client = api_client
473 self._state_lock = threading.Lock()
474 self._state = None # Previous run state (file list & manifest)
475 self._current_files = [] # Current run file list
476 self._cache_file = None
477 self._collection_lock = threading.Lock()
478 self._remote_collection = None # Collection being updated (if asked)
479 self._local_collection = None # Collection from previous run manifest
480 self._file_paths = set() # Files to be updated in remote collection
481 self._stop_checkpointer = threading.Event()
482 self._checkpointer = threading.Thread(target=self._update_task)
483 self._checkpointer.daemon = True
484 self._update_task_time = update_time # How many seconds wait between update runs
485 self._files_to_upload = FileUploadList(dry_run=dry_run)
486 self._upload_started = False
488 self.dry_run = dry_run
489 self._checkpoint_before_quit = True
490 self.follow_links = follow_links
491 self.exclude_paths = exclude_paths
492 self.exclude_names = exclude_names
493 self._trash_at = trash_at
495 if self._trash_at is not None:
496 if type(self._trash_at) not in [datetime.datetime, datetime.timedelta]:
497 raise TypeError('trash_at should be None, timezone-naive datetime or timedelta')
498 if type(self._trash_at) == datetime.datetime and self._trash_at.tzinfo is not None:
499 raise TypeError('provided trash_at datetime should be timezone-naive')
501 if not self.use_cache and self.resume:
502 raise ArvPutArgumentConflict('resume cannot be True when use_cache is False')
504 # Check for obvious dry-run responses
505 if self.dry_run and (not self.use_cache or not self.resume):
506 raise ArvPutUploadIsPending()
508 # Load cached data if any and if needed
509 self._setup_state(update_collection)
511 # Build the upload file list, excluding requested files and counting the
512 # bytes expected to be uploaded.
513 self._build_upload_list()
515 def _build_upload_list(self):
517 Scan the requested paths to count file sizes, excluding requested files
518 and dirs and building the upload file list.
520 # If there aren't special files to be read, reset total bytes count to zero
522 if not any([p for p in self.paths
523 if not (os.path.isfile(p) or os.path.isdir(p))]):
524 self.bytes_expected = 0
526 for path in self.paths:
527 # Test for stdin first, in case some file named '-' exist
530 raise ArvPutUploadIsPending()
531 self._write_stdin(self.filename or 'stdin')
532 elif not os.path.exists(path):
533 raise PathDoesNotExistError(u"file or directory '{}' does not exist.".format(path))
534 elif (not self.follow_links) and os.path.islink(path):
535 self.logger.warning("Skipping symlink '{}'".format(path))
537 elif os.path.isdir(path):
538 # Use absolute paths on cache index so CWD doesn't interfere
539 # with the caching logic.
541 path = os.path.abspath(path)
542 if orig_path[-1:] == os.sep:
543 # When passing a directory reference with a trailing slash,
544 # its contents should be uploaded directly to the
548 # When passing a directory reference with no trailing slash,
549 # upload the directory to the collection's root.
550 prefixdir = os.path.dirname(path)
552 for root, dirs, files in os.walk(path,
553 followlinks=self.follow_links):
554 root_relpath = os.path.relpath(root, path)
555 if root_relpath == '.':
557 # Exclude files/dirs by full path matching pattern
558 if self.exclude_paths:
559 dirs[:] = [d for d in dirs
560 if not any(pathname_match(
561 os.path.join(root_relpath, d), pat)
562 for pat in self.exclude_paths)]
563 files = [f for f in files
564 if not any(pathname_match(
565 os.path.join(root_relpath, f), pat)
566 for pat in self.exclude_paths)]
567 # Exclude files/dirs by name matching pattern
568 if self.exclude_names is not None:
569 dirs[:] = [d for d in dirs
570 if not self.exclude_names.match(d)]
571 files = [f for f in files
572 if not self.exclude_names.match(f)]
573 # Make os.walk()'s dir traversing order deterministic
577 filepath = os.path.join(root, f)
578 # Add its size to the total bytes count (if applicable)
579 if self.follow_links or (not os.path.islink(filepath)):
580 if self.bytes_expected is not None:
581 self.bytes_expected += os.path.getsize(filepath)
582 self._check_file(filepath,
583 os.path.join(root[len(prefixdir):], f))
585 filepath = os.path.abspath(path)
586 # Add its size to the total bytes count (if applicable)
587 if self.follow_links or (not os.path.islink(filepath)):
588 if self.bytes_expected is not None:
589 self.bytes_expected += os.path.getsize(filepath)
590 self._check_file(filepath,
591 self.filename or os.path.basename(path))
592 # If dry-mode is on, and got up to this point, then we should notify that
593 # there aren't any file to upload.
595 raise ArvPutUploadNotPending()
596 # Remove local_collection's files that don't exist locally anymore, so the
597 # bytes_written count is correct.
598 for f in self.collection_file_paths(self._local_collection,
600 if f != 'stdin' and f != self.filename and not f in self._file_paths:
601 self._local_collection.remove(f)
603 def start(self, save_collection):
605 Start supporting thread & file uploading
607 self._checkpointer.start()
609 # Update bytes_written from current local collection and
610 # report initial progress.
613 self._upload_started = True # Used by the update thread to start checkpointing
615 except (SystemExit, Exception) as e:
616 self._checkpoint_before_quit = False
617 # Log stack trace only when Ctrl-C isn't pressed (SIGINT)
618 # Note: We're expecting SystemExit instead of
619 # KeyboardInterrupt because we have a custom signal
620 # handler in place that raises SystemExit with the catched
622 if isinstance(e, PathDoesNotExistError):
623 # We aren't interested in the traceback for this case
625 elif not isinstance(e, SystemExit) or e.code != -2:
626 self.logger.warning("Abnormal termination:\n{}".format(
627 traceback.format_exc()))
631 # Stop the thread before doing anything else
632 self._stop_checkpointer.set()
633 self._checkpointer.join()
634 if self._checkpoint_before_quit:
635 # Commit all pending blocks & one last _update()
636 self._local_collection.manifest_text()
637 self._update(final=True)
639 self.save_collection()
641 self._cache_file.close()
643 def _collection_trash_at(self):
645 Returns the trash date that the collection should use at save time.
646 Takes into account absolute/relative trash_at values requested
649 if type(self._trash_at) == datetime.timedelta:
650 # Get an absolute datetime for trash_at
651 return datetime.datetime.utcnow() + self._trash_at
652 return self._trash_at
654 def save_collection(self):
656 # Check if files should be updated on the remote collection.
657 for fp in self._file_paths:
658 remote_file = self._remote_collection.find(fp)
660 # File don't exist on remote collection, copy it.
661 self._remote_collection.copy(fp, fp, self._local_collection)
662 elif remote_file != self._local_collection.find(fp):
663 # A different file exist on remote collection, overwrite it.
664 self._remote_collection.copy(fp, fp, self._local_collection, overwrite=True)
666 # The file already exist on remote collection, skip it.
668 self._remote_collection.save(num_retries=self.num_retries,
669 trash_at=self._collection_trash_at())
671 if len(self._local_collection) == 0:
672 self.logger.warning("No files were uploaded, skipping collection creation.")
674 self._local_collection.save_new(
675 name=self.name, owner_uuid=self.owner_uuid,
676 ensure_unique_name=self.ensure_unique_name,
677 num_retries=self.num_retries,
678 trash_at=self._collection_trash_at())
680 def destroy_cache(self):
683 os.unlink(self._cache_filename)
684 except OSError as error:
685 # That's what we wanted anyway.
686 if error.errno != errno.ENOENT:
688 self._cache_file.close()
690 def _collection_size(self, collection):
692 Recursively get the total size of the collection
695 for item in listvalues(collection):
696 if isinstance(item, arvados.collection.Collection) or isinstance(item, arvados.collection.Subcollection):
697 size += self._collection_size(item)
702 def _update_task(self):
704 Periodically called support task. File uploading is
705 asynchronous so we poll status from the collection.
707 while not self._stop_checkpointer.wait(1 if not self._upload_started else self._update_task_time):
710 def _update(self, final=False):
712 Update cached manifest text and report progress.
714 if self._upload_started:
715 with self._collection_lock:
716 self.bytes_written = self._collection_size(self._local_collection)
719 manifest = self._local_collection.manifest_text()
721 # Get the manifest text without comitting pending blocks
722 manifest = self._local_collection.manifest_text(strip=False,
726 with self._state_lock:
727 self._state['manifest'] = manifest
731 except Exception as e:
732 self.logger.error("Unexpected error trying to save cache file: {}".format(e))
733 # Keep remote collection's trash_at attribute synced when using relative expire dates
734 if self._remote_collection is not None and type(self._trash_at) == datetime.timedelta:
736 self._api_client.collections().update(
737 uuid=self._remote_collection.manifest_locator(),
738 body={'trash_at': self._collection_trash_at().strftime("%Y-%m-%dT%H:%M:%S.%fZ")}
739 ).execute(num_retries=self.num_retries)
740 except Exception as e:
741 self.logger.error("Unexpected error trying to update remote collection's expire date: {}".format(e))
743 self.bytes_written = self.bytes_skipped
744 # Call the reporter, if any
745 self.report_progress()
747 def report_progress(self):
748 if self.reporter is not None:
749 self.reporter(self.bytes_written, self.bytes_expected)
751 def _write_stdin(self, filename):
752 output = self._local_collection.open(filename, 'wb')
753 self._write(sys.stdin.buffer, output)
756 def _check_file(self, source, filename):
758 Check if this file needs to be uploaded
760 # Ignore symlinks when requested
761 if (not self.follow_links) and os.path.islink(source):
764 should_upload = False
765 new_file_in_cache = False
766 # Record file path for updating the remote collection before exiting
767 self._file_paths.add(filename)
769 with self._state_lock:
770 # If no previous cached data on this file, store it for an eventual
772 if source not in self._state['files']:
773 self._state['files'][source] = {
774 'mtime': os.path.getmtime(source),
775 'size' : os.path.getsize(source)
777 new_file_in_cache = True
778 cached_file_data = self._state['files'][source]
780 # Check if file was already uploaded (at least partially)
781 file_in_local_collection = self._local_collection.find(filename)
783 # If not resuming, upload the full file.
786 # New file detected from last run, upload it.
787 elif new_file_in_cache:
789 # Local file didn't change from last run.
790 elif cached_file_data['mtime'] == os.path.getmtime(source) and cached_file_data['size'] == os.path.getsize(source):
791 if not file_in_local_collection:
792 # File not uploaded yet, upload it completely
794 elif file_in_local_collection.permission_expired():
795 # Permission token expired, re-upload file. This will change whenever
796 # we have a API for refreshing tokens.
797 self.logger.warning(u"Uploaded file '{}' access token expired, will re-upload it from scratch".format(filename))
799 self._local_collection.remove(filename)
800 elif cached_file_data['size'] == file_in_local_collection.size():
801 # File already there, skip it.
802 self.bytes_skipped += cached_file_data['size']
803 elif cached_file_data['size'] > file_in_local_collection.size():
804 # File partially uploaded, resume!
805 resume_offset = file_in_local_collection.size()
806 self.bytes_skipped += resume_offset
809 # Inconsistent cache, re-upload the file
811 self._local_collection.remove(filename)
812 self.logger.warning(u"Uploaded version of file '{}' is bigger than local version, will re-upload it from scratch.".format(source))
813 # Local file differs from cached data, re-upload it.
815 if file_in_local_collection:
816 self._local_collection.remove(filename)
821 self._files_to_upload.append((source, resume_offset, filename))
822 except ArvPutUploadIsPending:
823 # This could happen when running on dry-mode, close cache file to
824 # avoid locking issues.
825 self._cache_file.close()
828 def _upload_files(self):
829 for source, resume_offset, filename in self._files_to_upload:
830 with open(source, 'rb') as source_fd:
831 with self._state_lock:
832 self._state['files'][source]['mtime'] = os.path.getmtime(source)
833 self._state['files'][source]['size'] = os.path.getsize(source)
834 if resume_offset > 0:
835 # Start upload where we left off
836 output = self._local_collection.open(filename, 'ab')
837 source_fd.seek(resume_offset)
840 output = self._local_collection.open(filename, 'wb')
841 self._write(source_fd, output)
842 output.close(flush=False)
844 def _write(self, source_fd, output):
846 data = source_fd.read(arvados.config.KEEP_BLOCK_SIZE)
851 def _my_collection(self):
852 return self._remote_collection if self.update else self._local_collection
854 def _get_cache_filepath(self):
855 # Set up cache file name from input paths.
857 md5.update(arvados.config.get('ARVADOS_API_HOST', '!nohost').encode())
858 realpaths = sorted(os.path.realpath(path) for path in self.paths)
859 md5.update(b'\0'.join([p.encode() for p in realpaths]))
861 md5.update(self.filename.encode())
862 cache_filename = md5.hexdigest()
863 cache_filepath = os.path.join(
864 arv_cmd.make_home_conf_dir(self.CACHE_DIR, 0o700, 'raise'),
866 return cache_filepath
868 def _setup_state(self, update_collection):
870 Create a new cache file or load a previously existing one.
872 # Load an already existing collection for update
873 if update_collection and re.match(arvados.util.collection_uuid_pattern,
876 self._remote_collection = arvados.collection.Collection(
878 api_client=self._api_client,
879 storage_classes_desired=self.storage_classes,
880 num_retries=self.num_retries)
881 except arvados.errors.ApiError as error:
882 raise CollectionUpdateError("Cannot read collection {} ({})".format(update_collection, error))
885 elif update_collection:
886 # Collection locator provided, but unknown format
887 raise CollectionUpdateError("Collection locator unknown: '{}'".format(update_collection))
890 cache_filepath = self._get_cache_filepath()
891 if self.resume and os.path.exists(cache_filepath):
892 self.logger.info(u"Resuming upload from cache file {}".format(cache_filepath))
893 self._cache_file = open(cache_filepath, 'a+')
895 # --no-resume means start with a empty cache file.
896 self.logger.info(u"Creating new cache file at {}".format(cache_filepath))
897 self._cache_file = open(cache_filepath, 'w+')
898 self._cache_filename = self._cache_file.name
899 self._lock_file(self._cache_file)
900 self._cache_file.seek(0)
902 with self._state_lock:
905 self._state = json.load(self._cache_file)
906 if not set(['manifest', 'files']).issubset(set(self._state.keys())):
907 # Cache at least partially incomplete, set up new cache
908 self._state = copy.deepcopy(self.EMPTY_STATE)
910 # Cache file empty, set up new cache
911 self._state = copy.deepcopy(self.EMPTY_STATE)
913 self.logger.info("No cache usage requested for this run.")
914 # No cache file, set empty state
915 self._state = copy.deepcopy(self.EMPTY_STATE)
916 if not self._cached_manifest_valid():
917 raise ResumeCacheInvalidError()
918 # Load the previous manifest so we can check if files were modified remotely.
919 self._local_collection = arvados.collection.Collection(
920 self._state['manifest'],
921 replication_desired=self.replication_desired,
922 storage_classes_desired=(self.storage_classes or ['default']),
923 put_threads=self.put_threads,
924 api_client=self._api_client,
925 num_retries=self.num_retries)
927 def _cached_manifest_valid(self):
929 Validate the oldest non-expired block signature to check if cached manifest
930 is usable: checking if the cached manifest was not created with a different
933 if self._state.get('manifest', None) is None:
934 # No cached manifest yet, all good.
936 now = datetime.datetime.utcnow()
940 for m in keep_locator_pattern.finditer(self._state['manifest']):
943 exp = datetime.datetime.utcfromtimestamp(int(loc.split('@')[1], 16))
945 # Locator without signature
948 if exp > now and (oldest_exp is None or exp < oldest_exp):
952 # No block signatures found => no invalid block signatures.
954 if oldest_loc is None:
955 # Locator signatures found, but all have expired.
956 # Reset the cache and move on.
957 self.logger.info('Cache expired, starting from scratch.')
958 self._state['manifest'] = ''
960 kc = arvados.KeepClient(api_client=self._api_client,
961 num_retries=self.num_retries)
964 except arvados.errors.KeepRequestError:
965 # Something is wrong, cached manifest is not valid.
969 def collection_file_paths(self, col, path_prefix='.'):
970 """Return a list of file paths by recursively go through the entire collection `col`"""
972 for name, item in listitems(col):
973 if isinstance(item, arvados.arvfile.ArvadosFile):
974 file_paths.append(os.path.join(path_prefix, name))
975 elif isinstance(item, arvados.collection.Subcollection):
976 new_prefix = os.path.join(path_prefix, name)
977 file_paths += self.collection_file_paths(item, path_prefix=new_prefix)
980 def _lock_file(self, fileobj):
982 fcntl.flock(fileobj, fcntl.LOCK_EX | fcntl.LOCK_NB)
984 raise ResumeCacheConflict(u"{} locked".format(fileobj.name))
986 def _save_state(self):
988 Atomically save current state into cache.
990 with self._state_lock:
991 # We're not using copy.deepcopy() here because it's a lot slower
992 # than json.dumps(), and we're already needing JSON format to be
994 state = json.dumps(self._state)
996 new_cache = tempfile.NamedTemporaryFile(
998 dir=os.path.dirname(self._cache_filename), delete=False)
999 self._lock_file(new_cache)
1000 new_cache.write(state)
1003 os.rename(new_cache.name, self._cache_filename)
1004 except (IOError, OSError, ResumeCacheConflict) as error:
1005 self.logger.error("There was a problem while saving the cache file: {}".format(error))
1007 os.unlink(new_cache_name)
1008 except NameError: # mkstemp failed.
1011 self._cache_file.close()
1012 self._cache_file = new_cache
1014 def collection_name(self):
1015 return self._my_collection().api_response()['name'] if self._my_collection().api_response() else None
1017 def collection_trash_at(self):
1018 return self._my_collection().get_trash_at()
1020 def manifest_locator(self):
1021 return self._my_collection().manifest_locator()
1023 def portable_data_hash(self):
1024 pdh = self._my_collection().portable_data_hash()
1025 m = self._my_collection().stripped_manifest().encode()
1026 local_pdh = '{}+{}'.format(hashlib.md5(m).hexdigest(), len(m))
1027 if pdh != local_pdh:
1028 self.logger.warning("\n".join([
1029 "arv-put: API server provided PDH differs from local manifest.",
1030 " This should not happen; showing API server version."]))
1033 def manifest_text(self, stream_name=".", strip=False, normalize=False):
1034 return self._my_collection().manifest_text(stream_name, strip, normalize)
1036 def _datablocks_on_item(self, item):
1038 Return a list of datablock locators, recursively navigating
1039 through subcollections
1041 if isinstance(item, arvados.arvfile.ArvadosFile):
1042 if item.size() == 0:
1043 # Empty file locator
1044 return ["d41d8cd98f00b204e9800998ecf8427e+0"]
1047 for segment in item.segments():
1048 loc = segment.locator
1049 locators.append(loc)
1051 elif isinstance(item, arvados.collection.Collection):
1052 l = [self._datablocks_on_item(x) for x in listvalues(item)]
1053 # Fast list flattener method taken from:
1054 # http://stackoverflow.com/questions/952914/making-a-flat-list-out-of-list-of-lists-in-python
1055 return [loc for sublist in l for loc in sublist]
1059 def data_locators(self):
1060 with self._collection_lock:
1061 # Make sure all datablocks are flushed before getting the locators
1062 self._my_collection().manifest_text()
1063 datablocks = self._datablocks_on_item(self._my_collection())
1066 _machine_format = "{} {}: {{}} written {{}} total\n".format(sys.argv[0],
1069 # Simulate glob.glob() matching behavior without the need to scan the filesystem
1070 # Note: fnmatch() doesn't work correctly when used with pathnames. For example the
1071 # pattern 'tests/*.py' will match 'tests/run_test.py' and also 'tests/subdir/run_test.py',
1072 # so instead we're using it on every path component.
1073 def pathname_match(pathname, pattern):
1074 name = pathname.split(os.sep)
1075 # Fix patterns like 'some/subdir/' or 'some//subdir'
1076 pat = [x for x in pattern.split(os.sep) if x != '' and x != '.']
1077 if len(name) != len(pat):
1079 for i in range(len(name)):
1080 if not fnmatch.fnmatch(name[i], pat[i]):
1084 def machine_progress(bytes_written, bytes_expected):
1085 return _machine_format.format(
1086 bytes_written, -1 if (bytes_expected is None) else bytes_expected)
1088 def human_progress(bytes_written, bytes_expected):
1090 return "\r{}M / {}M {:.1%} ".format(
1091 bytes_written >> 20, bytes_expected >> 20,
1092 float(bytes_written) / bytes_expected)
1094 return "\r{} ".format(bytes_written)
1096 def progress_writer(progress_func, outfile=sys.stderr):
1097 def write_progress(bytes_written, bytes_expected):
1098 outfile.write(progress_func(bytes_written, bytes_expected))
1099 return write_progress
1101 def desired_project_uuid(api_client, project_uuid, num_retries):
1102 if not project_uuid:
1103 query = api_client.users().current()
1104 elif arvados.util.user_uuid_pattern.match(project_uuid):
1105 query = api_client.users().get(uuid=project_uuid)
1106 elif arvados.util.group_uuid_pattern.match(project_uuid):
1107 query = api_client.groups().get(uuid=project_uuid)
1109 raise ValueError("Not a valid project UUID: {}".format(project_uuid))
1110 return query.execute(num_retries=num_retries)['uuid']
1112 def main(arguments=None, stdout=sys.stdout, stderr=sys.stderr,
1113 install_sig_handlers=True):
1116 args = parse_arguments(arguments)
1117 logger = logging.getLogger('arvados.arv_put')
1119 logger.setLevel(logging.WARNING)
1121 logger.setLevel(logging.INFO)
1124 request_id = arvados.util.new_request_id()
1126 formatter = ArvPutLogFormatter(request_id)
1127 logging.getLogger('arvados').handlers[0].setFormatter(formatter)
1129 if api_client is None:
1130 api_client = arvados.api('v1', request_id=request_id)
1132 if install_sig_handlers:
1133 arv_cmd.install_signal_handlers()
1135 # Trash arguments validation
1137 if args.trash_at is not None:
1138 # ciso8601 considers YYYYMM as invalid but YYYY-MM as valid, so here we
1139 # make sure the user provides a complete YYYY-MM-DD date.
1140 if not re.match(r'^\d{4}(?P<dash>-?)\d{2}?(?P=dash)\d{2}', args.trash_at):
1141 logger.error("--trash-at argument format invalid, use --help to see examples.")
1143 # Check if no time information was provided. In that case, assume end-of-day.
1144 if re.match(r'^\d{4}(?P<dash>-?)\d{2}?(?P=dash)\d{2}$', args.trash_at):
1145 args.trash_at += 'T23:59:59'
1147 trash_at = ciso8601.parse_datetime(args.trash_at)
1149 logger.error("--trash-at argument format invalid, use --help to see examples.")
1152 if trash_at.tzinfo is not None:
1153 # Timezone aware datetime provided.
1154 utcoffset = -trash_at.utcoffset()
1156 # Timezone naive datetime provided. Assume is local.
1158 utcoffset = datetime.timedelta(seconds=time.altzone)
1160 utcoffset = datetime.timedelta(seconds=time.timezone)
1161 # Convert to UTC timezone naive datetime.
1162 trash_at = trash_at.replace(tzinfo=None) + utcoffset
1164 if trash_at <= datetime.datetime.utcnow():
1165 logger.error("--trash-at argument must be set in the future")
1167 if args.trash_after is not None:
1168 if args.trash_after < 1:
1169 logger.error("--trash-after argument must be >= 1")
1171 trash_at = datetime.timedelta(seconds=(args.trash_after * 24 * 60 * 60))
1173 # Determine the name to use
1175 if args.stream or args.raw:
1176 logger.error("Cannot use --name with --stream or --raw")
1178 elif args.update_collection:
1179 logger.error("Cannot use --name with --update-collection")
1181 collection_name = args.name
1183 collection_name = "Saved at {} by {}@{}".format(
1184 datetime.datetime.utcnow().strftime("%Y-%m-%d %H:%M:%S UTC"),
1185 pwd.getpwuid(os.getuid()).pw_name,
1186 socket.gethostname())
1188 if args.project_uuid and (args.stream or args.raw):
1189 logger.error("Cannot use --project-uuid with --stream or --raw")
1192 # Determine the parent project
1194 project_uuid = desired_project_uuid(api_client, args.project_uuid,
1196 except (apiclient_errors.Error, ValueError) as error:
1201 reporter = progress_writer(human_progress)
1202 elif args.batch_progress:
1203 reporter = progress_writer(machine_progress)
1207 # Split storage-classes argument
1208 storage_classes = None
1209 if args.storage_classes:
1210 storage_classes = args.storage_classes.strip().replace(' ', '').split(',')
1212 # Setup exclude regex from all the --exclude arguments provided
1215 exclude_names = None
1216 if len(args.exclude) > 0:
1217 # We're supporting 2 kinds of exclusion patterns:
1218 # 1) --exclude '*.jpg' (file/dir name patterns, will only match
1219 # the name, wherever the file is on the tree)
1220 # 2.1) --exclude 'foo/bar' (file/dir path patterns, will match the
1221 # entire path, and should be relative to
1222 # any input dir argument)
1223 # 2.2) --exclude './*.jpg' (Special case for excluding files/dirs
1224 # placed directly underneath the input dir)
1225 for p in args.exclude:
1226 # Only relative paths patterns allowed
1227 if p.startswith(os.sep):
1228 logger.error("Cannot use absolute paths with --exclude")
1230 if os.path.dirname(p):
1231 # We don't support of path patterns with '..'
1232 p_parts = p.split(os.sep)
1235 "Cannot use path patterns that include or '..'")
1237 # Path search pattern
1238 exclude_paths.append(p)
1240 # Name-only search pattern
1241 name_patterns.append(p)
1242 # For name only matching, we can combine all patterns into a single
1243 # regexp, for better performance.
1244 exclude_names = re.compile('|'.join(
1245 [fnmatch.translate(p) for p in name_patterns]
1246 )) if len(name_patterns) > 0 else None
1247 # Show the user the patterns to be used, just in case they weren't
1248 # specified inside quotes and got changed by the shell expansion.
1249 logger.info("Exclude patterns: {}".format(args.exclude))
1251 # If this is used by a human, and there's at least one directory to be
1252 # uploaded, the expected bytes calculation can take a moment.
1253 if args.progress and any([os.path.isdir(f) for f in args.paths]):
1254 logger.info("Calculating upload size, this could take some time...")
1256 writer = ArvPutUploadJob(paths = args.paths,
1257 resume = args.resume,
1258 use_cache = args.use_cache,
1259 filename = args.filename,
1260 reporter = reporter,
1261 api_client = api_client,
1262 num_retries = args.retries,
1263 replication_desired = args.replication,
1264 put_threads = args.threads,
1265 name = collection_name,
1266 owner_uuid = project_uuid,
1267 ensure_unique_name = True,
1268 update_collection = args.update_collection,
1269 storage_classes=storage_classes,
1271 dry_run=args.dry_run,
1272 follow_links=args.follow_links,
1273 exclude_paths=exclude_paths,
1274 exclude_names=exclude_names,
1276 except ResumeCacheConflict:
1277 logger.error("\n".join([
1278 "arv-put: Another process is already uploading this data.",
1279 " Use --no-cache if this is really what you want."]))
1281 except ResumeCacheInvalidError:
1282 logger.error("\n".join([
1283 "arv-put: Resume cache contains invalid signature: it may have expired",
1284 " or been created with another Arvados user's credentials.",
1285 " Switch user or use one of the following options to restart upload:",
1286 " --no-resume to start a new resume cache.",
1287 " --no-cache to disable resume cache."]))
1289 except (CollectionUpdateError, PathDoesNotExistError) as error:
1290 logger.error("\n".join([
1291 "arv-put: %s" % str(error)]))
1293 except ArvPutUploadIsPending:
1294 # Dry run check successful, return proper exit code.
1296 except ArvPutUploadNotPending:
1297 # No files pending for upload
1300 if not args.dry_run and not args.update_collection and args.resume and writer.bytes_written > 0:
1301 logger.warning("\n".join([
1302 "arv-put: Resuming previous upload from last checkpoint.",
1303 " Use the --no-resume option to start over."]))
1305 if not args.dry_run:
1306 writer.report_progress()
1309 writer.start(save_collection=not(args.stream or args.raw))
1310 except (arvados.errors.ApiError, arvados.errors.KeepWriteError) as error:
1311 logger.error("\n".join([
1312 "arv-put: %s" % str(error)]))
1315 if args.progress: # Print newline to split stderr from stdout for humans.
1320 output = writer.manifest_text(normalize=True)
1322 output = writer.manifest_text()
1324 output = ','.join(writer.data_locators())
1325 elif writer.manifest_locator() is not None:
1327 expiration_notice = ""
1328 if writer.collection_trash_at() is not None:
1329 # Get the local timezone-naive version, and log it with timezone information.
1331 local_trash_at = writer.collection_trash_at().replace(tzinfo=None) - datetime.timedelta(seconds=time.altzone)
1333 local_trash_at = writer.collection_trash_at().replace(tzinfo=None) - datetime.timedelta(seconds=time.timezone)
1334 expiration_notice = ". It will expire on {} {}.".format(
1335 local_trash_at.strftime("%Y-%m-%d %H:%M:%S"), time.strftime("%z"))
1336 if args.update_collection:
1337 logger.info(u"Collection updated: '{}'{}".format(
1338 writer.collection_name(), expiration_notice))
1340 logger.info(u"Collection saved as '{}'{}".format(
1341 writer.collection_name(), expiration_notice))
1342 if args.portable_data_hash:
1343 output = writer.portable_data_hash()
1345 output = writer.manifest_locator()
1346 except apiclient_errors.Error as error:
1348 "arv-put: Error creating Collection on project: {}.".format(
1354 # Print the locator (uuid) of the new collection.
1356 status = status or 1
1357 elif not args.silent:
1358 stdout.write(output)
1359 if not output.endswith('\n'):
1362 if install_sig_handlers:
1363 arv_cmd.restore_signal_handlers()
1372 if __name__ == '__main__':