X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/9b2aa42213a7d333bbe93e040c2d152a70e9b5af..95e5ccacf6c1193b313fa90a6d39baafa2ba67d8:/sdk/python/arvados/collection.py diff --git a/sdk/python/arvados/collection.py b/sdk/python/arvados/collection.py index 098a11e1bb..e38a6bd475 100644 --- a/sdk/python/arvados/collection.py +++ b/sdk/python/arvados/collection.py @@ -1,9 +1,20 @@ +# Copyright (C) The Arvados Authors. All rights reserved. +# +# SPDX-License-Identifier: Apache-2.0 + +from __future__ import absolute_import +from future.utils import listitems, listvalues, viewkeys +from builtins import str +from past.builtins import basestring +from builtins import object import functools import logging import os import re import errno import hashlib +import datetime +import ciso8601 import time import threading @@ -11,20 +22,22 @@ from collections import deque from stat import * from .arvfile import split, _FileLikeObjectBase, ArvadosFile, ArvadosFileWriter, ArvadosFileReader, _BlockManager, synchronized, must_be_writable, NoopLock -from keep import KeepLocator, KeepClient +from .keep import KeepLocator, KeepClient from .stream import StreamReader from ._normalize_stream import normalize_stream from ._ranges import Range, LocatorAndRange from .safeapi import ThreadSafeApiCache -import config -import errors -import util -import events +import arvados.config as config +import arvados.errors as errors +import arvados.util +import arvados.events as events from arvados.retry import retry_method _logger = logging.getLogger('arvados.collection') class CollectionBase(object): + """Abstract base class for Collection classes.""" + def __enter__(self): return self @@ -51,7 +64,7 @@ class CollectionBase(object): if fields: clean_fields = fields[:1] + [ (re.sub(r'\+[^\d][^\+]*', '', x) - if re.match(util.keep_locator_pattern, x) + if re.match(arvados.util.keep_locator_pattern, x) else x) for x in fields[1:]] clean += [' '.join(clean_fields), "\n"] @@ -82,6 +95,8 @@ class _WriterFile(_FileLikeObjectBase): class CollectionWriter(CollectionBase): + """Deprecated, use Collection instead.""" + def __init__(self, api_client=None, num_retries=0, replication=None): """Instantiate a CollectionWriter. @@ -180,7 +195,7 @@ class CollectionWriter(CollectionBase): def _work_trees(self): path, stream_name, max_manifest_depth = self._queued_trees[0] - d = util.listdir_recursive( + d = arvados.util.listdir_recursive( path, max_depth = (None if max_manifest_depth == 0 else 0)) if d: self._queue_dirents(stream_name, d) @@ -216,7 +231,11 @@ class CollectionWriter(CollectionBase): self.do_queued_work() def write(self, newdata): - if hasattr(newdata, '__iter__'): + if isinstance(newdata, bytes): + pass + elif isinstance(newdata, str): + newdata = newdata.encode() + elif hasattr(newdata, '__iter__'): for s in newdata: self.write(s) return @@ -256,7 +275,7 @@ class CollectionWriter(CollectionBase): return self._last_open def flush_data(self): - data_buffer = ''.join(self._data_buffer) + data_buffer = b''.join(self._data_buffer) if data_buffer: self._current_stream_locators.append( self._my_keep().put( @@ -346,11 +365,12 @@ class CollectionWriter(CollectionBase): sending manifest_text() to the API server's "create collection" endpoint. """ - return self._my_keep().put(self.manifest_text(), copies=self.replication) + return self._my_keep().put(self.manifest_text().encode(), + copies=self.replication) def portable_data_hash(self): - stripped = self.stripped_manifest() - return hashlib.md5(stripped).hexdigest() + '+' + str(len(stripped)) + stripped = self.stripped_manifest().encode() + return '{}+{}'.format(hashlib.md5(stripped).hexdigest(), len(stripped)) def manifest_text(self): self.finish_current_stream() @@ -372,8 +392,18 @@ class CollectionWriter(CollectionBase): ret += locators return ret + def save_new(self, name=None): + return self._api_client.collections().create( + ensure_unique_name=True, + body={ + 'name': name, + 'manifest_text': self.manifest_text(), + }).execute(num_retries=self.num_retries) + class ResumableCollectionWriter(CollectionWriter): + """Deprecated, use Collection instead.""" + STATE_PROPS = ['_current_stream_files', '_current_stream_length', '_current_stream_locators', '_current_stream_name', '_current_file_name', '_current_file_pos', '_close_file', @@ -418,7 +448,7 @@ class ResumableCollectionWriter(CollectionWriter): return writer def check_dependencies(self): - for path, orig_stat in self._dependencies.items(): + for path, orig_stat in listitems(self._dependencies): if not S_ISREG(orig_stat[ST_MODE]): raise errors.StaleWriterStateError("{} not file".format(path)) try: @@ -612,7 +642,12 @@ class RichCollectionBase(CollectionBase): :path: path to a file in the collection :mode: - one of "r", "r+", "w", "w+", "a", "a+" + a string consisting of "r", "w", or "a", optionally followed + by "b" or "t", optionally followed by "+". + :"b": + binary mode: write() accepts bytes, read() returns bytes. + :"t": + text mode (default): write() accepts strings, read() returns strings. :"r": opens for reading :"r+": @@ -624,33 +659,28 @@ class RichCollectionBase(CollectionBase): the end of the file. Writing does not affect the file pointer for reading. """ - mode = mode.replace("b", "") - if len(mode) == 0 or mode[0] not in ("r", "w", "a"): - raise errors.ArgumentError("Bad mode '%s'" % mode) - create = (mode != "r") - if create and not self.writable(): - raise IOError(errno.EROFS, "Collection is read only") + if not re.search(r'^[rwa][bt]?\+?$', mode): + raise errors.ArgumentError("Invalid mode {!r}".format(mode)) - if create: - arvfile = self.find_or_create(path, FILE) - else: + if mode[0] == 'r' and '+' not in mode: + fclass = ArvadosFileReader arvfile = self.find(path) + elif not self.writable(): + raise IOError(errno.EROFS, "Collection is read only") + else: + fclass = ArvadosFileWriter + arvfile = self.find_or_create(path, FILE) if arvfile is None: raise IOError(errno.ENOENT, "File not found", path) if not isinstance(arvfile, ArvadosFile): raise IOError(errno.EISDIR, "Is a directory", path) - if mode[0] == "w": + if mode[0] == 'w': arvfile.truncate(0) - name = os.path.basename(path) - - if mode == "r": - return ArvadosFileReader(arvfile, num_retries=self.num_retries) - else: - return ArvadosFileWriter(arvfile, mode, num_retries=self.num_retries) + return fclass(arvfile, mode=mode, num_retries=self.num_retries) def modified(self): """Determine if the collection has been modified since last commited.""" @@ -672,7 +702,7 @@ class RichCollectionBase(CollectionBase): if value == self._committed: return if value: - for k,v in self._items.items(): + for k,v in listitems(self._items): v.set_committed(True) self._committed = True else: @@ -683,7 +713,7 @@ class RichCollectionBase(CollectionBase): @synchronized def __iter__(self): """Iterate over names of files and collections contained in this collection.""" - return iter(self._items.keys()) + return iter(viewkeys(self._items)) @synchronized def __getitem__(self, k): @@ -720,12 +750,12 @@ class RichCollectionBase(CollectionBase): @synchronized def values(self): """Get a list of files and collection objects directly contained in this collection.""" - return self._items.values() + return listvalues(self._items) @synchronized def items(self): """Get a list of (name, object) tuples directly contained in this collection.""" - return self._items.items() + return listitems(self._items) def exists(self, path): """Test if there is a file or collection at `path`.""" @@ -758,7 +788,7 @@ class RichCollectionBase(CollectionBase): item.remove(pathcomponents[1]) def _clonefrom(self, source): - for k,v in source.items(): + for k,v in listitems(source): self._items[k] = v.clone(self, k) def clone(self): @@ -1074,8 +1104,8 @@ class RichCollectionBase(CollectionBase): # then return API server's PDH response. return self._portable_data_hash else: - stripped = self.portable_manifest_text() - return hashlib.md5(stripped).hexdigest() + '+' + str(len(stripped)) + stripped = self.portable_manifest_text().encode() + return '{}+{}'.format(hashlib.md5(stripped).hexdigest(), len(stripped)) @synchronized def subscribe(self, callback): @@ -1116,7 +1146,7 @@ class RichCollectionBase(CollectionBase): @synchronized def flush(self): """Flush bufferblocks to Keep.""" - for e in self.values(): + for e in listvalues(self): e.flush() @@ -1171,8 +1201,9 @@ class Collection(RichCollectionBase): """Collection constructor. :manifest_locator_or_text: - One of Arvados collection UUID, block locator of - a manifest, raw manifest text, or None (to create an empty collection). + An Arvados collection UUID, portable data hash, raw manifest + text, or (if creating an empty collection) None. + :parent: the parent Collection, may be None. @@ -1222,11 +1253,11 @@ class Collection(RichCollectionBase): self.events = None if manifest_locator_or_text: - if re.match(util.keep_locator_pattern, manifest_locator_or_text): + if re.match(arvados.util.keep_locator_pattern, manifest_locator_or_text): self._manifest_locator = manifest_locator_or_text - elif re.match(util.collection_uuid_pattern, manifest_locator_or_text): + elif re.match(arvados.util.collection_uuid_pattern, manifest_locator_or_text): self._manifest_locator = manifest_locator_or_text - elif re.match(util.manifest_pattern, manifest_locator_or_text): + elif re.match(arvados.util.manifest_pattern, manifest_locator_or_text): self._manifest_text = manifest_locator_or_text else: raise errors.ArgumentError( @@ -1240,6 +1271,18 @@ class Collection(RichCollectionBase): def root_collection(self): return self + def get_properties(self): + if self._api_response and self._api_response["properties"]: + return self._api_response["properties"] + else: + return {} + + def get_trash_at(self): + if self._api_response and self._api_response["trash_at"]: + return ciso8601.parse_datetime(self._api_response["trash_at"]) + else: + return None + def stream_name(self): return "." @@ -1311,66 +1354,27 @@ class Collection(RichCollectionBase): # it. If instantiation fails, we'll fall back to the except # clause, just like any other Collection lookup # failure. Return an exception, or None if successful. - try: - self._remember_api_response(self._my_api().collections().get( - uuid=self._manifest_locator).execute( - num_retries=self.num_retries)) - self._manifest_text = self._api_response['manifest_text'] - # If not overriden via kwargs, we should try to load the - # replication_desired from the API server - if self.replication_desired is None: - self.replication_desired = self._api_response.get('replication_desired', None) - return None - except Exception as e: - return e - - def _populate_from_keep(self): - # Retrieve a manifest directly from Keep. This has a chance of - # working if [a] the locator includes a permission signature - # or [b] the Keep services are operating in world-readable - # mode. Return an exception, or None if successful. - try: - self._manifest_text = self._my_keep().get( - self._manifest_locator, num_retries=self.num_retries) - except Exception as e: - return e + self._remember_api_response(self._my_api().collections().get( + uuid=self._manifest_locator).execute( + num_retries=self.num_retries)) + self._manifest_text = self._api_response['manifest_text'] + self._portable_data_hash = self._api_response['portable_data_hash'] + # If not overriden via kwargs, we should try to load the + # replication_desired from the API server + if self.replication_desired is None: + self.replication_desired = self._api_response.get('replication_desired', None) def _populate(self): - if self._manifest_locator is None and self._manifest_text is None: - return - error_via_api = None - error_via_keep = None - should_try_keep = ((self._manifest_text is None) and - util.keep_locator_pattern.match( - self._manifest_locator)) - if ((self._manifest_text is None) and - util.signed_locator_pattern.match(self._manifest_locator)): - error_via_keep = self._populate_from_keep() if self._manifest_text is None: - error_via_api = self._populate_from_api_server() - if error_via_api is not None and not should_try_keep: - raise error_via_api - if ((self._manifest_text is None) and - not error_via_keep and - should_try_keep): - # Looks like a keep locator, and we didn't already try keep above - error_via_keep = self._populate_from_keep() - if self._manifest_text is None: - # Nothing worked! - raise errors.NotFoundError( - ("Failed to retrieve collection '{}' " + - "from either API server ({}) or Keep ({})." - ).format( - self._manifest_locator, - error_via_api, - error_via_keep)) - # populate + if self._manifest_locator is None: + return + else: + self._populate_from_api_server() self._baseline_manifest = self._manifest_text self._import_manifest(self._manifest_text) - def _has_collection_uuid(self): - return self._manifest_locator is not None and re.match(util.collection_uuid_pattern, self._manifest_locator) + return self._manifest_locator is not None and re.match(arvados.util.collection_uuid_pattern, self._manifest_locator) def __enter__(self): return self @@ -1446,17 +1450,34 @@ class Collection(RichCollectionBase): @must_be_writable @synchronized @retry_method - def save(self, merge=True, num_retries=None): + def save(self, + properties=None, + storage_classes=None, + trash_at=None, + merge=True, + num_retries=None): """Save collection to an existing collection record. Commit pending buffer blocks to Keep, merge with remote record (if - merge=True, the default), and update the collection record. Returns + merge=True, the default), and update the collection record. Returns the current manifest text. Will raise AssertionError if not associated with a collection record on the API server. If you want to save a manifest to Keep only, see `save_new()`. + :properties: + Additional properties of collection. This value will replace any existing + properties of collection. + + :storage_classes: + Specify desirable storage classes to be used when writing data to Keep. + + :trash_at: + A collection is *expiring* when it has a *trash_at* time in the future. + An expiring collection can be accessed as normal, + but is scheduled to be trashed automatically at the *trash_at* time. + :merge: Update and merge remote changes before saving. Otherwise, any remote changes will be ignored and overwritten. @@ -1465,6 +1486,24 @@ class Collection(RichCollectionBase): Retry count on API calls (if None, use the collection default) """ + if properties and type(properties) is not dict: + raise errors.ArgumentError("properties must be dictionary type.") + + if storage_classes and type(storage_classes) is not list: + raise errors.ArgumentError("storage_classes must be list type.") + + if trash_at and type(trash_at) is not datetime.datetime: + raise errors.ArgumentError("trash_at must be datetime type.") + + body={} + if properties: + body["properties"] = properties + if storage_classes: + body["storage_classes_desired"] = storage_classes + if trash_at: + t = trash_at.strftime("%Y-%m-%dT%H:%M:%S.%fZ") + body["trash_at"] = t + if not self.committed(): if not self._has_collection_uuid(): raise AssertionError("Collection manifest_locator is not a collection uuid. Use save_new() for new collections.") @@ -1475,14 +1514,20 @@ class Collection(RichCollectionBase): self.update() text = self.manifest_text(strip=False) + body['manifest_text'] = text + self._remember_api_response(self._my_api().collections().update( uuid=self._manifest_locator, - body={'manifest_text': text} - ).execute( - num_retries=num_retries)) + body=body + ).execute(num_retries=num_retries)) self._manifest_text = self._api_response["manifest_text"] self._portable_data_hash = self._api_response["portable_data_hash"] self.set_committed(True) + elif body: + self._remember_api_response(self._my_api().collections().update( + uuid=self._manifest_locator, + body=body + ).execute(num_retries=num_retries)) return self._manifest_text @@ -1493,6 +1538,9 @@ class Collection(RichCollectionBase): def save_new(self, name=None, create_collection_record=True, owner_uuid=None, + properties=None, + storage_classes=None, + trash_at=None, ensure_unique_name=False, num_retries=None): """Save collection to a new collection record. @@ -1500,7 +1548,7 @@ class Collection(RichCollectionBase): Commit pending buffer blocks to Keep and, when create_collection_record is True (default), create a new collection record. After creating a new collection record, this Collection object will be associated with - the new record used by `save()`. Returns the current manifest text. + the new record used by `save()`. Returns the current manifest text. :name: The collection name. @@ -1513,6 +1561,18 @@ class Collection(RichCollectionBase): the user, or project uuid that will own this collection. If None, defaults to the current user. + :properties: + Additional properties of collection. This value will replace any existing + properties of collection. + + :storage_classes: + Specify desirable storage classes to be used when writing data to Keep. + + :trash_at: + A collection is *expiring* when it has a *trash_at* time in the future. + An expiring collection can be accessed as normal, + but is scheduled to be trashed automatically at the *trash_at* time. + :ensure_unique_name: If True, ask the API server to rename the collection if it conflicts with a collection with the same name and owner. If @@ -1522,6 +1582,15 @@ class Collection(RichCollectionBase): Retry count on API calls (if None, use the collection default) """ + if properties and type(properties) is not dict: + raise errors.ArgumentError("properties must be dictionary type.") + + if storage_classes and type(storage_classes) is not list: + raise errors.ArgumentError("storage_classes must be list type.") + + if trash_at and type(trash_at) is not datetime.datetime: + raise errors.ArgumentError("trash_at must be datetime type.") + self._my_block_manager().commit_all() text = self.manifest_text(strip=False) @@ -1535,6 +1604,13 @@ class Collection(RichCollectionBase): "replication_desired": self.replication_desired} if owner_uuid: body["owner_uuid"] = owner_uuid + if properties: + body["properties"] = properties + if storage_classes: + body["storage_classes_desired"] = storage_classes + if trash_at: + t = trash_at.strftime("%Y-%m-%dT%H:%M:%S.%fZ") + body["trash_at"] = t self._remember_api_response(self._my_api().collections().create(ensure_unique_name=ensure_unique_name, body=body).execute(num_retries=num_retries)) text = self._api_response["manifest_text"] @@ -1547,6 +1623,10 @@ class Collection(RichCollectionBase): return text + _token_re = re.compile(r'(\S+)(\s+|$)') + _block_re = re.compile(r'[0-9a-f]{32}\+(\d+)(\+\S+)*') + _segment_re = re.compile(r'(\d+):(\d+):(\S+)') + @synchronized def _import_manifest(self, manifest_text): """Import a manifest into a `Collection`. @@ -1565,7 +1645,7 @@ class Collection(RichCollectionBase): stream_name = None state = STREAM_NAME - for token_and_separator in re.finditer(r'(\S+)(\s+|$)', manifest_text): + for token_and_separator in self._token_re.finditer(manifest_text): tok = token_and_separator.group(1) sep = token_and_separator.group(2) @@ -1574,25 +1654,25 @@ class Collection(RichCollectionBase): stream_name = tok.replace('\\040', ' ') blocks = [] segments = [] - streamoffset = 0L + streamoffset = 0 state = BLOCKS self.find_or_create(stream_name, COLLECTION) continue if state == BLOCKS: - block_locator = re.match(r'[0-9a-f]{32}\+(\d+)(\+\S+)*', tok) + block_locator = self._block_re.match(tok) if block_locator: - blocksize = long(block_locator.group(1)) + blocksize = int(block_locator.group(1)) blocks.append(Range(tok, streamoffset, blocksize, 0)) streamoffset += blocksize else: state = SEGMENTS if state == SEGMENTS: - file_segment = re.search(r'^(\d+):(\d+):(\S+)', tok) + file_segment = self._segment_re.match(tok) if file_segment: - pos = long(file_segment.group(1)) - size = long(file_segment.group(2)) + pos = int(file_segment.group(1)) + size = int(file_segment.group(2)) name = file_segment.group(3).replace('\\040', ' ') filepath = os.path.join(stream_name, name) afile = self.find_or_create(filepath, FILE) @@ -1669,9 +1749,8 @@ class Subcollection(RichCollectionBase): class CollectionReader(Collection): """A read-only collection object. - Initialize from an api collection record locator, a portable data hash of a - manifest, or raw manifest text. See `Collection` constructor for detailed - options. + Initialize from a collection UUID or portable data hash, or raw + manifest text. See `Collection` constructor for detailed options. """ def __init__(self, manifest_locator_or_text, *args, **kwargs):