X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/8f5755f5ec65da08ffe9dfc6856848ec747a31a8..c7f445954df35959174761ba7b1f44ecf377c87a:/sdk/python/arvados/commands/put.py?ds=sidebyside diff --git a/sdk/python/arvados/commands/put.py b/sdk/python/arvados/commands/put.py index 86670266ae..e4e1b6dad2 100644 --- a/sdk/python/arvados/commands/put.py +++ b/sdk/python/arvados/commands/put.py @@ -11,121 +11,130 @@ import fcntl import hashlib import json import os +import signal import sys import tempfile -def parse_arguments(arguments): - parser = argparse.ArgumentParser( - description='Copy data from the local filesystem to Keep.') - - parser.add_argument('paths', metavar='path', type=str, nargs='*', - help=""" - Local file or directory. Default: read from standard input. - """) - - parser.add_argument('--max-manifest-depth', type=int, metavar='N', - default=-1, help=""" - Maximum depth of directory tree to represent in the manifest - structure. A directory structure deeper than this will be represented - as a single stream in the manifest. If N=0, the manifest will contain - a single stream. Default: -1 (unlimited), i.e., exactly one manifest - stream per filesystem directory that contains files. - """) - - group = parser.add_mutually_exclusive_group() - - group.add_argument('--as-stream', action='store_true', dest='stream', - help=""" - Synonym for --stream. - """) - - group.add_argument('--stream', action='store_true', - help=""" - Store the file content and display the resulting manifest on - stdout. Do not write the manifest to Keep or save a Collection object - in Arvados. - """) - - group.add_argument('--as-manifest', action='store_true', dest='manifest', - help=""" - Synonym for --manifest. - """) - - group.add_argument('--in-manifest', action='store_true', dest='manifest', - help=""" - Synonym for --manifest. - """) - - group.add_argument('--manifest', action='store_true', - help=""" - Store the file data and resulting manifest in Keep, save a Collection - object in Arvados, and display the manifest locator (Collection uuid) - on stdout. This is the default behavior. - """) - - group.add_argument('--as-raw', action='store_true', dest='raw', - help=""" - Synonym for --raw. - """) - - group.add_argument('--raw', action='store_true', - help=""" - Store the file content and display the data block locators on stdout, - separated by commas, with a trailing newline. Do not store a - manifest. - """) - - parser.add_argument('--use-filename', type=str, default=None, - dest='filename', help=""" - Synonym for --filename. - """) - - parser.add_argument('--filename', type=str, default=None, - help=""" - Use the given filename in the manifest, instead of the name of the - local file. This is useful when "-" or "/dev/stdin" is given as an - input file. It can be used only if there is exactly one path given and - it is not a directory. Implies --manifest. - """) - - group = parser.add_mutually_exclusive_group() - group.add_argument('--progress', action='store_true', - help=""" - Display human-readable progress on stderr (bytes and, if possible, - percentage of total data size). This is the default behavior when - stderr is a tty. - """) - - group.add_argument('--no-progress', action='store_true', - help=""" - Do not display human-readable progress on stderr, even if stderr is a - tty. - """) +import arvados.commands._util as arv_cmd + +CAUGHT_SIGNALS = [signal.SIGINT, signal.SIGQUIT, signal.SIGTERM] + +upload_opts = argparse.ArgumentParser(add_help=False) + +upload_opts.add_argument('paths', metavar='path', type=str, nargs='*', + help=""" +Local file or directory. Default: read from standard input. +""") + +upload_opts.add_argument('--max-manifest-depth', type=int, metavar='N', + default=-1, help=""" +Maximum depth of directory tree to represent in the manifest +structure. A directory structure deeper than this will be represented +as a single stream in the manifest. If N=0, the manifest will contain +a single stream. Default: -1 (unlimited), i.e., exactly one manifest +stream per filesystem directory that contains files. +""") + +_group = upload_opts.add_mutually_exclusive_group() + +_group.add_argument('--as-stream', action='store_true', dest='stream', + help=""" +Synonym for --stream. +""") + +_group.add_argument('--stream', action='store_true', + help=""" +Store the file content and display the resulting manifest on +stdout. Do not write the manifest to Keep or save a Collection object +in Arvados. +""") + +_group.add_argument('--as-manifest', action='store_true', dest='manifest', + help=""" +Synonym for --manifest. +""") + +_group.add_argument('--in-manifest', action='store_true', dest='manifest', + help=""" +Synonym for --manifest. +""") + +_group.add_argument('--manifest', action='store_true', + help=""" +Store the file data and resulting manifest in Keep, save a Collection +object in Arvados, and display the manifest locator (Collection uuid) +on stdout. This is the default behavior. +""") + +_group.add_argument('--as-raw', action='store_true', dest='raw', + help=""" +Synonym for --raw. +""") + +_group.add_argument('--raw', action='store_true', + help=""" +Store the file content and display the data block locators on stdout, +separated by commas, with a trailing newline. Do not store a +manifest. +""") + +upload_opts.add_argument('--use-filename', type=str, default=None, + dest='filename', help=""" +Synonym for --filename. +""") + +upload_opts.add_argument('--filename', type=str, default=None, + help=""" +Use the given filename in the manifest, instead of the name of the +local file. This is useful when "-" or "/dev/stdin" is given as an +input file. It can be used only if there is exactly one path given and +it is not a directory. Implies --manifest. +""") + +run_opts = argparse.ArgumentParser(add_help=False) +_group = run_opts.add_mutually_exclusive_group() +_group.add_argument('--progress', action='store_true', + help=""" +Display human-readable progress on stderr (bytes and, if possible, +percentage of total data size). This is the default behavior when +stderr is a tty. +""") + +_group.add_argument('--no-progress', action='store_true', + help=""" +Do not display human-readable progress on stderr, even if stderr is a +tty. +""") + +_group.add_argument('--batch-progress', action='store_true', + help=""" +Display machine-readable progress on stderr (bytes and, if known, +total data size). +""") + +_group = run_opts.add_mutually_exclusive_group() +_group.add_argument('--resume', action='store_true', default=True, + help=""" +Continue interrupted uploads from cached state (default). +""") +_group.add_argument('--no-resume', action='store_false', dest='resume', + help=""" +Do not continue interrupted uploads from cached state. +""") + +arg_parser = argparse.ArgumentParser( + description='Copy data from the local filesystem to Keep.', + parents=[upload_opts, run_opts]) - group.add_argument('--batch-progress', action='store_true', - help=""" - Display machine-readable progress on stderr (bytes and, if known, - total data size). - """) - - group = parser.add_mutually_exclusive_group() - group.add_argument('--resume', action='store_true', default=True, - help=""" - Continue interrupted uploads from cached state (default). - """) - group.add_argument('--no-resume', action='store_false', dest='resume', - help=""" - Do not continue interrupted uploads from cached state. - """) - - args = parser.parse_args(arguments) +def parse_arguments(arguments): + args = arg_parser.parse_args(arguments) if len(args.paths) == 0: args.paths += ['/dev/stdin'] if len(args.paths) != 1 or os.path.isdir(args.paths[0]): if args.filename: - parser.error(""" + arg_parser.error(""" --filename argument cannot be used when storing a directory or multiple files. """) @@ -147,17 +156,11 @@ class ResumeCacheConflict(Exception): class ResumeCache(object): - CACHE_DIR = os.path.expanduser('~/.cache/arvados/arv-put') + CACHE_DIR = '.cache/arvados/arv-put' @classmethod def setup_user_cache(cls): - try: - os.makedirs(cls.CACHE_DIR) - except OSError as error: - if error.errno != errno.EEXIST: - raise - else: - os.chmod(cls.CACHE_DIR, 0o700) + return arv_cmd.make_home_conf_dir(cls.CACHE_DIR, 0o700) def __init__(self, file_spec): self.cache_file = open(file_spec, 'a+') @@ -169,7 +172,7 @@ class ResumeCache(object): md5 = hashlib.md5() md5.update(arvados.config.get('ARVADOS_API_HOST', '!nohost')) realpaths = sorted(os.path.realpath(path) for path in args.paths) - md5.update(''.join(realpaths)) + md5.update('\0'.join(realpaths)) if any(os.path.isdir(path) for path in realpaths): md5.update(str(max(args.max_manifest_depth, -1))) elif args.filename: @@ -227,10 +230,7 @@ class ArvPutCollectionWriter(arvados.ResumableCollectionWriter): self.bytes_written = 0 self._seen_inputs = [] self.cache = cache - if reporter is None: - self.report_progress = lambda bytes_w, bytes_e: None - else: - self.report_progress = reporter + self.reporter = reporter self.bytes_expected = bytes_expected super(ArvPutCollectionWriter, self).__init__() @@ -246,11 +246,9 @@ class ArvPutCollectionWriter(arvados.ResumableCollectionWriter): else: return writer - def preresume_hook(self): - print >>sys.stderr, "arv-put: Resuming previous upload. Bypass with the --no-resume option." - self.report_progress(self.bytes_written, self.bytes_expected) - def cache_state(self): + if self.cache is None: + return state = self.dump_state() # Transform attributes for serialization. for attr, value in state.items(): @@ -260,11 +258,19 @@ class ArvPutCollectionWriter(arvados.ResumableCollectionWriter): state[attr] = list(value) self.cache.save(state) + def report_progress(self): + if self.reporter is not None: + self.reporter(self.bytes_written, self.bytes_expected) + def flush_data(self): - bytes_buffered = self._data_buffer_len + start_buffer_len = self._data_buffer_len + start_block_count = self.bytes_written / self.KEEP_BLOCK_SIZE super(ArvPutCollectionWriter, self).flush_data() - self.bytes_written += (bytes_buffered - self._data_buffer_len) - self.report_progress(self.bytes_written, self.bytes_expected) + 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: + self.cache_state() def _record_new_input(self, input_type, source_name, dest_name): # The key needs to be a list because that's what we'll get back @@ -319,8 +325,10 @@ def progress_writer(progress_func, outfile=sys.stderr): outfile.write(progress_func(bytes_written, bytes_expected)) return write_progress +def exit_signal_handler(sigcode, frame): + sys.exit(-sigcode) + def main(arguments=None): - ResumeCache.setup_user_cache() args = parse_arguments(arguments) if args.progress: @@ -329,39 +337,57 @@ def main(arguments=None): reporter = progress_writer(machine_progress) else: reporter = None + bytes_expected = expected_bytes_for(args.paths) + resume_cache = None try: - resume_cache = ResumeCache(ResumeCache.make_path(args)) - if not args.resume: - resume_cache.restart() + if ResumeCache.setup_user_cache() is not None: + resume_cache = ResumeCache(ResumeCache.make_path(args)) + except (IOError, OSError): + pass # Couldn't open cache directory/file. Continue without it. except ResumeCacheConflict: print "arv-put: Another process is already uploading this data." sys.exit(1) - try: + if resume_cache is None: + writer = ArvPutCollectionWriter(resume_cache, reporter, bytes_expected) + else: + if not args.resume: + resume_cache.restart() writer = ArvPutCollectionWriter.from_cache( - resume_cache, reporter, expected_bytes_for(args.paths)) - - # Copy file data to Keep. - for path in args.paths: - if os.path.isdir(path): - writer.write_directory_tree( - path, max_manifest_depth=args.max_manifest_depth) - else: - writer.start_new_stream() - writer.write_file(path, args.filename or os.path.basename(path)) - except (Exception, KeyboardInterrupt): - writer.cache_state() - raise + resume_cache, reporter, bytes_expected) + + # Install our signal handler for each code in CAUGHT_SIGNALS, and save + # the originals. + orig_signal_handlers = {sigcode: signal.signal(sigcode, exit_signal_handler) + for sigcode in CAUGHT_SIGNALS} + + if writer.bytes_written > 0: # We're resuming a previous upload. + print >>sys.stderr, "\n".join([ + "arv-put: Resuming previous upload from last checkpoint.", + " Use the --no-resume option to start over."]) + 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): + writer.write_directory_tree( + path, max_manifest_depth=args.max_manifest_depth) + else: + writer.start_new_stream() + writer.write_file(path, args.filename or os.path.basename(path)) + writer.finish_current_stream() + + if args.progress: # Print newline to split stderr from stdout for humans. + print >>sys.stderr if args.stream: print writer.manifest_text(), elif args.raw: - writer.finish_current_stream() print ','.join(writer.data_locators()) else: # Register the resulting collection in Arvados. - arvados.api().collections().create( + collection = arvados.api().collections().create( body={ 'uuid': writer.finish(), 'manifest_text': writer.manifest_text(), @@ -369,8 +395,13 @@ def main(arguments=None): ).execute() # Print the locator (uuid) of the new collection. - print writer.finish() - resume_cache.destroy() + print collection['uuid'] + + for sigcode, orig_handler in orig_signal_handlers.items(): + signal.signal(sigcode, orig_handler) + + if resume_cache is not None: + resume_cache.destroy() if __name__ == '__main__': main()