-#!/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 datetime
import errno
import fcntl
+import fnmatch
import hashlib
import json
import logging
import tempfile
import threading
import time
+import traceback
+
from apiclient import errors as apiclient_errors
from arvados._version import __version__
help='Print version and exit.')
upload_opts.add_argument('paths', metavar='path', type=str, nargs='*',
help="""
-Local file or directory. Default: read from standard input.
+Local file or directory. If path is a directory reference with a trailing
+slash, then just upload the directory's contents; otherwise upload the
+directory itself. Default: read from standard input.
""")
_group = upload_opts.add_mutually_exclusive_group()
Save the collection with the specified name.
""")
+run_opts.add_argument('--exclude', metavar='PATTERN', default=[],
+ action='append', help="""
+Exclude files and directories whose names match the given pattern. You
+can specify multiple patterns by using this argument more than once.
+""")
+
_group = run_opts.add_mutually_exclusive_group()
_group.add_argument('--progress', action='store_true',
help="""
Do not continue interrupted uploads from cached state.
""")
+_group = run_opts.add_mutually_exclusive_group()
+_group.add_argument('--follow-links', action='store_true', default=True,
+ dest='follow_links', help="""
+Follow file and directory symlinks (default).
+""")
+_group.add_argument('--no-follow-links', action='store_false', dest='follow_links',
+ help="""
+Do not follow file and directory symlinks.
+""")
+
_group = run_opts.add_mutually_exclusive_group()
_group.add_argument('--cache', action='store_true', dest='use_cache', default=True,
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:
if not args.filename:
args.filename = 'stdin'
+ # Remove possible duplicated patterns
+ if len(args.exclude) > 0:
+ args.exclude = list(set(args.exclude))
+
return args
+class PathDoesNotExistError(Exception):
+ pass
+
+
class CollectionUpdateError(Exception):
pass
@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())
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):
+ logger=logging.getLogger('arvados.arv_put'), dry_run=False,
+ follow_links=True, exclude_paths=[], exclude_names=None):
self.paths = paths
self.resume = resume
self.use_cache = use_cache
self._upload_started = False
self.logger = logger
self.dry_run = dry_run
+ self._checkpoint_before_quit = True
+ self.follow_links = follow_links
+ self.exclude_paths = exclude_paths
+ self.exclude_names = exclude_names
if not self.use_cache and self.resume:
raise ArvPutArgumentConflict('resume cannot be True when use_cache is False')
if self.dry_run:
raise ArvPutUploadIsPending()
self._write_stdin(self.filename or 'stdin')
+ elif not os.path.exists(path):
+ raise PathDoesNotExistError("file or directory '{}' does not exist.".format(path))
elif os.path.isdir(path):
# Use absolute paths on cache index so CWD doesn't interfere
# with the caching logic.
- prefixdir = path = os.path.abspath(path)
- if prefixdir != '/':
- prefixdir += '/'
- for root, dirs, files in os.walk(path):
+ orig_path = path
+ path = os.path.abspath(path)
+ if orig_path[-1:] == os.sep:
+ # When passing a directory reference with a trailing slash,
+ # its contents should be uploaded directly to the collection's root.
+ prefixdir = path
+ else:
+ # When passing a directory reference with no trailing slash,
+ # upload the directory to the collection's root.
+ prefixdir = os.path.dirname(path)
+ prefixdir += os.sep
+ for root, dirs, files in os.walk(path, followlinks=self.follow_links):
+ root_relpath = os.path.relpath(root, path)
+ # Exclude files/dirs by full path matching pattern
+ if self.exclude_paths:
+ dirs[:] = filter(
+ lambda d: not any([pathname_match(os.path.join(root_relpath, d),
+ pat)
+ for pat in self.exclude_paths]),
+ dirs)
+ files = filter(
+ lambda f: not any([pathname_match(os.path.join(root_relpath, f),
+ pat)
+ for pat in self.exclude_paths]),
+ files)
+ # Exclude files/dirs by name matching pattern
+ if self.exclude_names is not None:
+ dirs[:] = filter(lambda d: not self.exclude_names.match(d), dirs)
+ files = filter(lambda f: not self.exclude_names.match(f), files)
# Make os.walk()'s dir traversing order deterministic
dirs.sort()
files.sort()
# 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 isinstance(e, PathDoesNotExistError):
+ # We aren't interested in the traceback for this case
+ pass
+ elif 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:
with self._state_lock:
self._state['manifest'] = manifest
if self.use_cache:
- self._save_state()
+ 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.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()
def _check_file(self, source, filename):
- """Check if this file needs to be uploaded"""
+ """
+ Check if this file needs to be uploaded
+ """
+ # Ignore symlinks when requested
+ if (not self.follow_links) and os.path.islink(source):
+ return
resume_offset = 0
should_upload = False
new_file_in_cache = False
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.
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:
- # 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)
- 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+')
+ 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:
def portable_data_hash(self):
pdh = self._my_collection().portable_data_hash()
- m = self._my_collection().stripped_manifest()
- local_pdh = hashlib.md5(m).hexdigest() + '+' + str(len(m))
+ 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.",
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]
datablocks = self._datablocks_on_item(self._my_collection())
return datablocks
-
-def expected_bytes_for(pathlist):
+def expected_bytes_for(pathlist, follow_links=True, exclude={}):
# Walk the given directory trees and stat files, adding up file sizes,
# so we can display progress as percent
bytesum = 0
+ exclude_paths = exclude.get('paths', None)
+ exclude_names = exclude.get('names', None)
for path in pathlist:
if os.path.isdir(path):
- for filename in arvados.util.listdir_recursive(path):
- bytesum += os.path.getsize(os.path.join(path, filename))
+ for root, dirs, files in os.walk(path, followlinks=follow_links):
+ root_relpath = os.path.relpath(root, path)
+ # Exclude files/dirs by full path matching pattern
+ if exclude_paths is not None:
+ dirs[:] = filter(
+ lambda d: not any([pathname_match(os.path.join(root_relpath, d),
+ pat)
+ for pat in exclude_paths]),
+ dirs)
+ files = filter(
+ lambda f: not any([pathname_match(os.path.join(root_relpath, f),
+ pat)
+ for pat in exclude_paths]),
+ files)
+ # Exclude files/dirs by name matching pattern
+ if exclude_names is not None:
+ dirs[:] = filter(lambda d: not exclude_names.match(d), dirs)
+ files = filter(lambda f: not exclude_names.match(f), files)
+ # Sum file sizes
+ for f in files:
+ filepath = os.path.join(root, f)
+ # Ignore symlinked files when requested
+ if (not follow_links) and os.path.islink(filepath):
+ continue
+ bytesum += os.path.getsize(filepath)
elif not os.path.isfile(path):
return None
else:
_machine_format = "{} {}: {{}} written {{}} total\n".format(sys.argv[0],
os.getpid())
+
+# Simulate glob.glob() matching behavior without the need to scan the filesystem
+# Note: fnmatch() doesn't work correctly when used with pathnames. For example the
+# pattern 'tests/*.py' will match 'tests/run_test.py' and also 'tests/subdir/run_test.py',
+# so instead we're using it on every path component.
+def pathname_match(pathname, pattern):
+ name = pathname.split(os.sep)
+ pat = pattern.split(os.sep)
+ if len(name) != len(pat):
+ return False
+ for i in range(len(name)):
+ if not fnmatch.fnmatch(name[i], pat[i]):
+ return False
+ return True
+
def machine_progress(bytes_written, bytes_expected):
return _machine_format.format(
bytes_written, -1 if (bytes_expected is None) else bytes_expected)
else:
reporter = None
+ # Setup exclude regex from all the --exclude arguments provided
+ name_patterns = []
+ exclude_paths = []
+ exclude_names = None
+ if len(args.exclude) > 0:
+ # We're supporting 2 kinds of exclusion patterns:
+ # 1) --exclude '*.jpg' (file/dir name patterns, will only match the name)
+ # 2) --exclude 'foo/bar' (file/dir path patterns, will match the entire path,
+ # and should be relative to any input dir argument)
+ for p in args.exclude:
+ # Only relative paths patterns allowed
+ if p.startswith(os.sep):
+ logger.error("Cannot use absolute paths with --exclude")
+ sys.exit(1)
+ if os.path.dirname(p):
+ # Path search pattern
+ exclude_paths.append(p)
+ else:
+ # Name-only search pattern
+ name_patterns.append(p)
+ # For name only matching, we can combine all patterns into a single regexp,
+ # for better performance.
+ exclude_names = re.compile('|'.join(
+ [fnmatch.translate(p) for p in name_patterns]
+ )) if len(name_patterns) > 0 else None
+ # Show the user the patterns to be used, just in case they weren't specified inside
+ # quotes and got changed by the shell expansion.
+ logger.info("Exclude patterns: {}".format(args.exclude))
+
# 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)
+ bytes_expected = expected_bytes_for(args.paths,
+ follow_links=args.follow_links,
+ exclude={'paths': exclude_paths,
+ 'names': exclude_names})
+
try:
writer = ArvPutUploadJob(paths = args.paths,
ensure_unique_name = True,
update_collection = args.update_collection,
logger=logger,
- dry_run=args.dry_run)
+ dry_run=args.dry_run,
+ follow_links=args.follow_links,
+ exclude_paths=exclude_paths,
+ exclude_names=exclude_names)
except ResumeCacheConflict:
logger.error("\n".join([
"arv-put: Another process is already uploading this data.",
except ArvPutUploadNotPending:
# No files pending for upload
sys.exit(0)
+ except PathDoesNotExistError as error:
+ logger.error("\n".join([
+ "arv-put: %s" % str(error)]))
+ sys.exit(1)
if args.progress: # Print newline to split stderr from stdout for humans.
logger.info("\n")
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:
projects / arvados.git / blobdiff