X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/da63a23ebc1d936305324cbc66d9903b9129df94..63b03a39adfd78961c5bbb6a3a2d02ccd8c92e4d:/sdk/python/arvados/collection.py diff --git a/sdk/python/arvados/collection.py b/sdk/python/arvados/collection.py index 42db95e622..b0cdd92251 100644 --- a/sdk/python/arvados/collection.py +++ b/sdk/python/arvados/collection.py @@ -3,16 +3,19 @@ import logging import os import re import errno +import hashlib import time +import threading from collections import deque from stat import * -from .arvfile import ArvadosFileBase, split, ArvadosFile, ArvadosFileWriter, ArvadosFileReader, BlockManager, synchronized, must_be_writable, SYNC_READONLY, SYNC_EXPLICIT, SYNC_LIVE, NoopLock -from keep import * -from .stream import StreamReader, normalize_stream, locator_block_size -from .ranges import Range, LocatorAndRange -from .safeapi import SafeApi +from .arvfile import split, _FileLikeObjectBase, ArvadosFile, ArvadosFileWriter, ArvadosFileReader, _BlockManager, synchronized, must_be_writable, NoopLock +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 @@ -35,7 +38,8 @@ class CollectionBase(object): return self._keep_client def stripped_manifest(self): - """ + """Get the manifest with locator hints stripped. + Return the manifest for the current collection with all non-portable hints (i.e., permission signatures and other hints other than size hints) removed from the locators. @@ -54,197 +58,7 @@ class CollectionBase(object): return ''.join(clean) -class CollectionReader(CollectionBase): - def __init__(self, manifest_locator_or_text, api_client=None, - keep_client=None, num_retries=0): - """Instantiate a CollectionReader. - - This class parses Collection manifests to provide a simple interface - to read its underlying files. - - Arguments: - * manifest_locator_or_text: One of a Collection UUID, portable data - hash, or full manifest text. - * api_client: The API client to use to look up Collections. If not - provided, CollectionReader will build one from available Arvados - configuration. - * keep_client: The KeepClient to use to download Collection data. - If not provided, CollectionReader will build one from available - Arvados configuration. - * num_retries: The default number of times to retry failed - service requests. Default 0. You may change this value - after instantiation, but note those changes may not - propagate to related objects like the Keep client. - """ - self._api_client = api_client - self._keep_client = keep_client - self.num_retries = num_retries - if re.match(util.keep_locator_pattern, manifest_locator_or_text): - self._manifest_locator = manifest_locator_or_text - self._manifest_text = None - elif re.match(util.collection_uuid_pattern, manifest_locator_or_text): - self._manifest_locator = manifest_locator_or_text - self._manifest_text = None - elif re.match(util.manifest_pattern, manifest_locator_or_text): - self._manifest_text = manifest_locator_or_text - self._manifest_locator = None - else: - raise errors.ArgumentError( - "Argument to CollectionReader must be a manifest or a collection UUID") - self._api_response = None - self._streams = None - - def _populate_from_api_server(self): - # As in KeepClient itself, we must wait until the last - # possible moment to instantiate an API client, in order to - # avoid tripping up clients that don't have access to an API - # server. If we do build one, make sure our Keep client uses - # 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: - if self._api_client is None: - self._api_client = arvados.api('v1') - self._keep_client = None # Make a new one with the new api. - self._api_response = self._api_client.collections().get( - uuid=self._manifest_locator).execute( - num_retries=self.num_retries) - self._manifest_text = self._api_response['manifest_text'] - 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 - - def _populate(self): - 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 arvados.errors.NotFoundError( - ("Failed to retrieve collection '{}' " + - "from either API server ({}) or Keep ({})." - ).format( - self._manifest_locator, - error_via_api, - error_via_keep)) - self._streams = [sline.split() - for sline in self._manifest_text.split("\n") - if sline] - - def _populate_first(orig_func): - # Decorator for methods that read actual Collection data. - @functools.wraps(orig_func) - def wrapper(self, *args, **kwargs): - if self._streams is None: - self._populate() - return orig_func(self, *args, **kwargs) - return wrapper - - @_populate_first - def api_response(self): - """api_response() -> dict or None - - Returns information about this Collection fetched from the API server. - If the Collection exists in Keep but not the API server, currently - returns None. Future versions may provide a synthetic response. - """ - return self._api_response - - @_populate_first - def normalize(self): - # Rearrange streams - streams = {} - for s in self.all_streams(): - for f in s.all_files(): - streamname, filename = split(s.name() + "/" + f.name()) - if streamname not in streams: - streams[streamname] = {} - if filename not in streams[streamname]: - streams[streamname][filename] = [] - for r in f.segments: - streams[streamname][filename].extend(s.locators_and_ranges(r.locator, r.range_size)) - - self._streams = [normalize_stream(s, streams[s]) - for s in sorted(streams)] - - # Regenerate the manifest text based on the normalized streams - self._manifest_text = ''.join( - [StreamReader(stream, keep=self._my_keep()).manifest_text() - for stream in self._streams]) - - @_populate_first - def open(self, streampath, filename=None): - """open(streampath[, filename]) -> file-like object - - Pass in the path of a file to read from the Collection, either as a - single string or as two separate stream name and file name arguments. - This method returns a file-like object to read that file. - """ - if filename is None: - streampath, filename = split(streampath) - keep_client = self._my_keep() - for stream_s in self._streams: - stream = StreamReader(stream_s, keep_client, - num_retries=self.num_retries) - if stream.name() == streampath: - break - else: - raise ValueError("stream '{}' not found in Collection". - format(streampath)) - try: - return stream.files()[filename] - except KeyError: - raise ValueError("file '{}' not found in Collection stream '{}'". - format(filename, streampath)) - - @_populate_first - def all_streams(self): - return [StreamReader(s, self._my_keep(), num_retries=self.num_retries) - for s in self._streams] - - def all_files(self): - for s in self.all_streams(): - for f in s.all_files(): - yield f - - @_populate_first - def manifest_text(self, strip=False, normalize=False): - if normalize: - cr = CollectionReader(self.manifest_text()) - cr.normalize() - return cr.manifest_text(strip=strip, normalize=False) - elif strip: - return self.stripped_manifest() - else: - return self._manifest_text - - -class _WriterFile(ArvadosFileBase): +class _WriterFile(_FileLikeObjectBase): def __init__(self, coll_writer, name): super(_WriterFile, self).__init__(name, 'wb') self.dest = coll_writer @@ -253,22 +67,22 @@ class _WriterFile(ArvadosFileBase): super(_WriterFile, self).close() self.dest.finish_current_file() - @ArvadosFileBase._before_close + @_FileLikeObjectBase._before_close def write(self, data): self.dest.write(data) - @ArvadosFileBase._before_close + @_FileLikeObjectBase._before_close def writelines(self, seq): for data in seq: self.write(data) - @ArvadosFileBase._before_close + @_FileLikeObjectBase._before_close def flush(self): self.dest.flush_data() class CollectionWriter(CollectionBase): - def __init__(self, api_client=None, num_retries=0): + def __init__(self, api_client=None, num_retries=0, replication=None): """Instantiate a CollectionWriter. CollectionWriter lets you build a new Arvados Collection from scratch. @@ -284,9 +98,13 @@ class CollectionWriter(CollectionBase): service requests. Default 0. You may change this value after instantiation, but note those changes may not propagate to related objects like the Keep client. + * replication: The number of copies of each block to store. + If this argument is None or not supplied, replication is + the server-provided default if available, otherwise 2. """ self._api_client = api_client self.num_retries = num_retries + self.replication = (2 if replication is None else replication) self._keep_client = None self._data_buffer = [] self._data_buffer_len = 0 @@ -441,7 +259,9 @@ class CollectionWriter(CollectionBase): data_buffer = ''.join(self._data_buffer) if data_buffer: self._current_stream_locators.append( - self._my_keep().put(data_buffer[0:config.KEEP_BLOCK_SIZE])) + self._my_keep().put( + data_buffer[0:config.KEEP_BLOCK_SIZE], + copies=self.replication)) self._data_buffer = [data_buffer[config.KEEP_BLOCK_SIZE:]] self._data_buffer_len = len(self._data_buffer[0]) @@ -516,8 +336,16 @@ class CollectionWriter(CollectionBase): self._current_file_name = None def finish(self): - # Store the manifest in Keep and return its locator. - return self._my_keep().put(self.manifest_text()) + """Store the manifest in Keep and return its locator. + + This is useful for storing manifest fragments (task outputs) + temporarily in Keep during a Crunch job. + + In other cases you should make a collection instead, by + sending manifest_text() to the API server's "create + collection" endpoint. + """ + return self._my_keep().put(self.manifest_text(), copies=self.replication) def portable_data_hash(self): stripped = self.stripped_manifest() @@ -551,10 +379,9 @@ class ResumableCollectionWriter(CollectionWriter): '_data_buffer', '_dependencies', '_finished_streams', '_queued_dirents', '_queued_trees'] - def __init__(self, api_client=None, num_retries=0): + def __init__(self, api_client=None, **kwargs): self._dependencies = {} - super(ResumableCollectionWriter, self).__init__( - api_client, num_retries=num_retries) + super(ResumableCollectionWriter, self).__init__(api_client, **kwargs) @classmethod def from_state(cls, state, *init_args, **init_kwargs): @@ -646,8 +473,17 @@ class ResumableCollectionWriter(CollectionWriter): ADD = "add" DEL = "del" MOD = "mod" +FILE = "file" +COLLECTION = "collection" class SynchronizedCollectionBase(CollectionBase): + """Base class for Collections and Subcollections. + + Implements the majority of functionality relating to accessing items in the + Collection. + + """ + def __init__(self, parent=None): self.parent = parent self._modified = True @@ -662,10 +498,7 @@ class SynchronizedCollectionBase(CollectionBase): def _my_block_manager(self): raise NotImplementedError() - def _populate(self): - raise NotImplementedError() - - def sync_mode(self): + def writable(self): raise NotImplementedError() def root_collection(self): @@ -674,61 +507,90 @@ class SynchronizedCollectionBase(CollectionBase): def notify(self, event, collection, name, item): raise NotImplementedError() + def stream_name(self): + raise NotImplementedError() + + @must_be_writable @synchronized - def find(self, path, create=False, create_collection=False): - """Recursively search the specified file path. May return either a Collection - or ArvadosFile. - - :create: - If true, create path components (i.e. Collections) that are - missing. If "create" is False, return None if a path component is - not found. - - :create_collection: - If the path is not found, "create" is True, and - "create_collection" is False, then create and return a new - ArvadosFile for the last path component. If "create_collection" is - True, then create and return a new Collection for the last path - component. + def find_or_create(self, path, create_type): + """Recursively search the specified file path. - """ - if create and self.sync_mode() == SYNC_READONLY: - raise IOError((errno.EROFS, "Collection is read only")) + May return either a `Collection` or `ArvadosFile`. If not found, will + create a new item at the specified path based on `create_type`. Will + create intermediate subcollections needed to contain the final item in + the path. - p = path.split("/") - if p[0] == '.': - del p[0] + :create_type: + One of `arvados.collection.FILE` or + `arvados.collection.COLLECTION`. If the path is not found, and value + of create_type is FILE then create and return a new ArvadosFile for + the last path component. If COLLECTION, then create and return a new + Collection for the last path component. - if p and p[0]: - item = self._items.get(p[0]) - if len(p) == 1: - # item must be a file - if item is None and create: + """ + + pathcomponents = path.split("/", 1) + if pathcomponents[0]: + item = self._items.get(pathcomponents[0]) + if len(pathcomponents) == 1: + if item is None: # create new file - if create_collection: + if create_type == COLLECTION: item = Subcollection(self) else: item = ArvadosFile(self) - self._items[p[0]] = item + self._items[pathcomponents[0]] = item self._modified = True - self.notify(ADD, self, p[0], item) + self.notify(ADD, self, pathcomponents[0], item) return item else: - if item is None and create: + if item is None: # create new collection item = Subcollection(self) - self._items[p[0]] = item + self._items[pathcomponents[0]] = item self._modified = True - self.notify(ADD, self, p[0], item) - del p[0] + self.notify(ADD, self, pathcomponents[0], item) if isinstance(item, SynchronizedCollectionBase): - return item.find("/".join(p), create=create) + return item.find_or_create(pathcomponents[1], create_type) else: - raise errors.ArgumentError("Interior path components must be subcollection") + raise IOError((errno.ENOTDIR, "Interior path components must be subcollection")) else: return self - def open(self, path, mode): + @synchronized + def find(self, path): + """Recursively search the specified file path. + + May return either a Collection or ArvadosFile. Return None if not + found. + + """ + if not path: + raise errors.ArgumentError("Parameter 'path' must not be empty.") + + pathcomponents = path.split("/", 1) + item = self._items.get(pathcomponents[0]) + if len(pathcomponents) == 1: + return item + else: + if isinstance(item, SynchronizedCollectionBase): + if pathcomponents[1]: + return item.find(pathcomponents[1]) + else: + return item + else: + raise IOError((errno.ENOTDIR, "Interior path components must be subcollection")) + + def mkdirs(path): + """Recursive subcollection create. + + Like `os.mkdirs()`. Will create intermediate subcollections needed to + contain the leaf subcollection path. + + """ + return self.find_or_create(path, COLLECTION) + + def open(self, path, mode="r"): """Open a file-like object for access. :path: @@ -748,31 +610,35 @@ class SynchronizedCollectionBase(CollectionBase): """ mode = mode.replace("b", "") if len(mode) == 0 or mode[0] not in ("r", "w", "a"): - raise ArgumentError("Bad mode '%s'" % mode) + raise errors.ArgumentError("Bad mode '%s'" % mode) create = (mode != "r") - if create and self.sync_mode() == SYNC_READONLY: + if create and not self.writable(): raise IOError((errno.EROFS, "Collection is read only")) - f = self.find(path, create=create) + if create: + arvfile = self.find_or_create(path, FILE) + else: + arvfile = self.find(path) - if f is None: + if arvfile is None: raise IOError((errno.ENOENT, "File not found")) - if not isinstance(f, ArvadosFile): + if not isinstance(arvfile, ArvadosFile): raise IOError((errno.EISDIR, "Path must refer to a file.")) if mode[0] == "w": - f.truncate(0) + arvfile.truncate(0) + + name = os.path.basename(path) if mode == "r": - return ArvadosFileReader(f, path, mode, num_retries=self.num_retries) + return ArvadosFileReader(arvfile, name, mode, num_retries=self.num_retries) else: - return ArvadosFileWriter(f, path, mode, num_retries=self.num_retries) + return ArvadosFileWriter(arvfile, name, mode, num_retries=self.num_retries) @synchronized def modified(self): - """Test if the collection (or any subcollection or file) has been modified - since it was created.""" + """Test if the collection (or any subcollection or file) has been modified.""" if self._modified: return True for k,v in self._items.items(): @@ -782,7 +648,7 @@ class SynchronizedCollectionBase(CollectionBase): @synchronized def set_unmodified(self): - """Recursively clear modified flag""" + """Recursively clear modified flag.""" self._modified = False for k,v in self._items.items(): v.set_unmodified() @@ -790,29 +656,25 @@ class SynchronizedCollectionBase(CollectionBase): @synchronized def __iter__(self): """Iterate over names of files and collections contained in this collection.""" - return self._items.keys().__iter__() - - @synchronized - def iterkeys(self): - """Iterate over names of files and collections directly contained in this collection.""" - return self._items.keys() + return iter(self._items.keys()) @synchronized def __getitem__(self, k): - """Get a file or collection that is directly contained by this collection. If - you want to search a path, use `find()` instead. + """Get a file or collection that is directly contained by this collection. + + If you want to search a path, use `find()` instead. + """ return self._items[k] @synchronized def __contains__(self, k): - """If there is a file or collection a directly contained by this collection - with name "k".""" + """Test if there is a file or collection a directly contained by this collection.""" return k in self._items @synchronized def __len__(self): - """Get the number of items directly contained in this collection""" + """Get the number of items directly contained in this collection.""" return len(self._items) @must_be_writable @@ -839,52 +701,84 @@ class SynchronizedCollectionBase(CollectionBase): return self._items.items() def exists(self, path): - """Test if there is a file or collection at "path" """ + """Test if there is a file or collection at `path`.""" return self.find(path) != None @must_be_writable @synchronized - def remove(self, path, rm_r=False): + def remove(self, path, recursive=False): """Remove the file or subcollection (directory) at `path`. - :rm_r: + + :recursive: Specify whether to remove non-empty subcollections (True), or raise an error (False). """ - p = path.split("/") - if p[0] == '.': - # Remove '.' from the front of the path - del p[0] - - if len(p) > 0: - item = self._items.get(p[0]) - if item is None: - raise IOError((errno.ENOENT, "File not found")) - if len(p) == 1: - if isinstance(self._items[p[0]], SynchronizedCollectionBase) and len(self._items[p[0]]) > 0 and not rm_r: - raise IOError((errno.ENOTEMPTY, "Subcollection not empty")) - d = self._items[p[0]] - del self._items[p[0]] - self._modified = True - self.notify(DEL, self, p[0], d) - else: - del p[0] - item.remove("/".join(p)) - else: + + if not path: + raise errors.ArgumentError("Parameter 'path' must not be empty.") + + pathcomponents = path.split("/", 1) + item = self._items.get(pathcomponents[0]) + if item is None: raise IOError((errno.ENOENT, "File not found")) + if len(pathcomponents) == 1: + if isinstance(self._items[pathcomponents[0]], SynchronizedCollectionBase) and len(self._items[pathcomponents[0]]) > 0 and not recursive: + raise IOError((errno.ENOTEMPTY, "Subcollection not empty")) + deleteditem = self._items[pathcomponents[0]] + del self._items[pathcomponents[0]] + self._modified = True + self.notify(DEL, self, pathcomponents[0], deleteditem) + else: + item.remove(pathcomponents[1]) - def _cloneinto(self, target): - for k,v in self._items.items(): - target._items[k] = v.clone(target) + def _clonefrom(self, source): + for k,v in source.items(): + self._items[k] = v.clone(self) def clone(self): raise NotImplementedError() + @must_be_writable + @synchronized + def add(self, source_obj, target_name, overwrite=False): + """Copy a file or subcollection to this collection. + + :source_obj: + An ArvadosFile, or Subcollection object + + :target_name: + Destination item name. If the target name already exists and is a + file, this will raise an error unless you specify `overwrite=True`. + + :overwrite: + Whether to overwrite target file if it already exists. + + """ + + if target_name in self and not overwrite: + raise IOError((errno.EEXIST, "File already exists")) + + modified_from = None + if target_name in self: + modified_from = self[target_name] + + # Actually make the copy. + dup = source_obj.clone(self) + self._items[target_name] = dup + self._modified = True + + if modified_from: + self.notify(MOD, self, target_name, (modified_from, dup)) + else: + self.notify(ADD, self, target_name, dup) + + @must_be_writable @synchronized def copy(self, source, target_path, source_collection=None, overwrite=False): """Copy a file or subcollection to a new path in this collection. :source: - An ArvadosFile, Subcollection, or string with a path to source file or subcollection + A string with a path to source file or subcollection, or an actual ArvadosFile or Subcollection object. :target_path: Destination file or path. If the target path already exists and is a @@ -906,61 +800,70 @@ class SynchronizedCollectionBase(CollectionBase): source_obj = source_collection.find(source) if source_obj is None: raise IOError((errno.ENOENT, "File not found")) - sp = source.split("/") + sourcecomponents = source.split("/") else: source_obj = source - sp = None + sourcecomponents = None # Find parent collection the target path - tp = target_path.split("/") + targetcomponents = target_path.split("/") # Determine the name to use. - target_name = tp[-1] if tp[-1] else (sp[-1] if sp else None) + target_name = targetcomponents[-1] if targetcomponents[-1] else (sourcecomponents[-1] if sourcecomponents else None) if not target_name: raise errors.ArgumentError("Target path is empty and source is an object. Cannot determine destination filename to use.") - target_dir = self.find("/".join(tp[0:-1]), create=True, create_collection=True) + target_dir = self.find_or_create("/".join(targetcomponents[0:-1]), COLLECTION) - with target_dir.lock: - if target_name in target_dir: - if isinstance(target_dir[target_name], SynchronizedCollectionBase) and sp: - target_dir = target_dir[target_name] - target_name = sp[-1] - elif not overwrite: - raise IOError((errno.EEXIST, "File already exists")) + if target_name in target_dir and isinstance(self[target_name], SynchronizedCollectionBase) and sourcecomponents: + target_dir = target_dir[target_name] + target_name = sourcecomponents[-1] - mod = None - if target_name in target_dir: - mod = target_dir[target_name] - - # Actually make the copy. - dup = source_obj.clone(target_dir) - target_dir._items[target_name] = dup - target_dir._modified = True - - if mod: - self.notify(MOD, target_dir, target_name, (mod, dup)) - else: - self.notify(ADD, target_dir, target_name, dup) + target_dir.add(source_obj, target_name, overwrite) @synchronized - def manifest_text(self, strip=False, normalize=False): + def manifest_text(self, stream_name=".", strip=False, normalize=False): """Get the manifest text for this collection, sub collections and files. + :stream_name: + Name of the stream (directory) + :strip: If True, remove signing tokens from block locators if present. - If False, block locators are left unchanged. + If False (default), block locators are left unchanged. :normalize: If True, always export the manifest text in normalized form - even if the Collection is not modified. If False and the collection + even if the Collection is not modified. If False (default) and the collection is not modified, return the original manifest text even if it is not in normalized form. """ + if self.modified() or self._manifest_text is None or normalize: - return export_manifest(self, stream_name=".", portable_locators=strip) + item = self + stream = {} + buf = [] + sorted_keys = sorted(item.keys()) + for filename in [s for s in sorted_keys if isinstance(item[s], ArvadosFile)]: + # Create a stream per file `k` + arvfile = item[filename] + filestream = [] + for segment in arvfile.segments(): + loc = segment.locator + if arvfile.parent._my_block_manager().is_bufferblock(loc): + loc = arvfile.parent._my_block_manager().get_bufferblock(loc).locator() + if strip: + loc = KeepLocator(loc).stripped() + filestream.append(LocatorAndRange(loc, KeepLocator(loc).size, + segment.segment_offset, segment.range_size)) + stream[filename] = filestream + if stream: + buf.append(" ".join(normalize_stream(stream_name, stream)) + "\n") + for dirname in [s for s in sorted_keys if isinstance(item[s], SynchronizedCollectionBase)]: + buf.append(item[dirname].manifest_text(stream_name=os.path.join(stream_name, dirname), strip=strip)) + return "".join(buf) else: if strip: return self.stripped_manifest() @@ -969,13 +872,14 @@ class SynchronizedCollectionBase(CollectionBase): @synchronized def diff(self, end_collection, prefix=".", holding_collection=None): - """ - Generate list of add/modify/delete actions which, when given to `apply`, will - change `self` to match `end_collection` + """Generate list of add/modify/delete actions. + + When given to `apply`, will change `self` to match `end_collection` + """ changes = [] if holding_collection is None: - holding_collection = Collection() + holding_collection = Collection(api_client=self._my_api(), keep_client=self._my_keep()) for k in self: if k not in end_collection: changes.append((DEL, os.path.join(prefix, k), self[k].clone(holding_collection))) @@ -992,17 +896,20 @@ class SynchronizedCollectionBase(CollectionBase): @must_be_writable @synchronized def apply(self, changes): + """Apply changes from `diff`. + + If a change conflicts with a local change, it will be saved to an + alternate path indicating the conflict. + """ - Apply changes from `diff`. If a change conflicts with a local change, it - will be saved to an alternate path indicating the conflict. - """ - for c in changes: - path = c[1] - initial = c[2] + for change in changes: + event_type = change[0] + path = change[1] + initial = change[2] local = self.find(path) conflictpath = "%s~conflict-%s~" % (path, time.strftime("%Y-%m-%d-%H:%M:%S", time.gmtime())) - if c[0] == ADD: + if event_type == ADD: if local is None: # No local file at path, safe to copy over new file self.copy(initial, path) @@ -1010,25 +917,26 @@ class SynchronizedCollectionBase(CollectionBase): # There is already local file and it is different: # save change to conflict file. self.copy(initial, conflictpath) - elif c[0] == MOD: + elif event_type == MOD: + final = change[3] if local == initial: - # Local matches the "initial" item so assume it hasn't + # Local matches the "initial" item so it has not # changed locally and is safe to update. - if isinstance(local, ArvadosFile) and isinstance(c[3], ArvadosFile): + if isinstance(local, ArvadosFile) and isinstance(final, ArvadosFile): # Replace contents of local file with new contents - local.replace_contents(c[3]) + local.replace_contents(final) else: - # Overwrite path with new item; this can happen if if + # Overwrite path with new item; this can happen if # path was a file and is now a collection or vice versa - self.copy(c[3], path, overwrite=True) + self.copy(final, path, overwrite=True) else: # Local is missing (presumably deleted) or local doesn't # match the "start" value, so save change to conflict file - self.copy(c[3], conflictpath) - elif c[0] == DEL: + self.copy(final, conflictpath) + elif event_type == DEL: if local == initial: # Local item matches "initial" value, so it is safe to remove. - self.remove(path, rm_r=True) + self.remove(path, recursive=True) # else, the file is modified or already removed, in either # case we don't want to try to remove it. @@ -1055,58 +963,89 @@ class SynchronizedCollectionBase(CollectionBase): def __ne__(self, other): return not self.__eq__(other) + class Collection(SynchronizedCollectionBase): - """Store an Arvados collection, consisting of a set of files and - sub-collections. This object + """Represents the root of an Arvados Collection. + + This class is threadsafe. The root collection object, all subcollections + and files are protected by a single lock (i.e. each access locks the entire + collection). + + Brief summary of + useful methods: + + :To read an existing file: + `c.open("myfile", "r")` + + :To write a new file: + `c.open("myfile", "w")` + + :To determine if a file exists: + `c.find("myfile") is not None` + + :To copy a file: + `c.copy("source", "dest")` + + :To delete a file: + `c.remove("myfile")` + + :To save to an existing collection record: + `c.save()` + + :To save a new collection record: + `c.save_new()` + + :To merge remote changes into this object: + `c.update()` + + Must be associated with an API server Collection record (during + initialization, or using `save_new`) to use `save` or `update` + """ def __init__(self, manifest_locator_or_text=None, - parent=None, - config=None, api_client=None, keep_client=None, num_retries=None, - block_manager=None, - sync=SYNC_READONLY): - """:manifest_locator_or_text: + parent=None, + apiconfig=None, + block_manager=None): + """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). :parent: the parent Collection, may be None. - :config: - the arvados configuration to get the hostname and api token. + :apiconfig: + A dict containing keys for ARVADOS_API_HOST and ARVADOS_API_TOKEN. Prefer this over supplying your own api_client and keep_client (except in testing). Will use default config settings if not specified. :api_client: - The API client object to use for requests. If not specified, create one using `config`. + The API client object to use for requests. If not specified, create one using `apiconfig`. :keep_client: - the Keep client to use for requests. If not specified, create one using `config`. + the Keep client to use for requests. If not specified, create one using `apiconfig`. :num_retries: the number of retries for API and Keep requests. :block_manager: the block manager to use. If not specified, create one. - :sync: - Set synchronization policy with API server collection record. - :SYNC_READONLY: - Collection is read only. No synchronization. This mode will - also forego locking, which gives better performance. - :SYNC_EXPLICIT: - Collection is writable. Synchronize on explicit request via `update()` or `save()` - :SYNC_LIVE: - Collection is writable. Synchronize with server in response to - background websocket events, on block write, or on file close. """ super(Collection, self).__init__(parent) self._api_client = api_client self._keep_client = keep_client self._block_manager = block_manager - self._config = config - self.num_retries = num_retries + + if apiconfig: + self._config = apiconfig + else: + self._config = config.settings() + + self.num_retries = num_retries if num_retries is not None else 0 self._manifest_locator = None self._manifest_text = None self._api_response = None - self._sync = sync + self.lock = threading.RLock() self.callbacks = [] self.events = None @@ -1122,43 +1061,37 @@ class Collection(SynchronizedCollectionBase): raise errors.ArgumentError( "Argument to CollectionReader must be a manifest or a collection UUID") - self._populate() - - if self._sync == SYNC_LIVE: - if not self._has_collection_uuid(): - raise errors.ArgumentError("Cannot SYNC_LIVE associated with a collection uuid") - self.events = events.subscribe(arvados.api(), [["object_uuid", "=", self._manifest_locator]], self.on_message) - - @staticmethod - def create(name, owner_uuid=None, sync=SYNC_EXPLICIT): - c = Collection(sync=sync) - c.save_as(name, owner_uuid=owner_uuid, ensure_unique_name=True) - return c + try: + self._populate() + except (IOError, errors.SyntaxError) as e: + raise errors.ArgumentError("Error processing manifest text: %s", e) def root_collection(self): return self - def sync_mode(self): - return self._sync + def stream_name(self): + return "." - def on_message(self): - self.update() + def writable(self): + return True @synchronized @retry_method def update(self, other=None, num_retries=None): + """Merge the latest collection on the API server with the current collection.""" + if other is None: if self._manifest_locator is None: raise errors.ArgumentError("`other` is None but collection does not have a manifest_locator uuid") - n = self._my_api().collections().get(uuid=self._manifest_locator).execute(num_retries=num_retries) - other = import_collection(n["manifest_text"]) - baseline = import_collection(self._manifest_text) - self.apply(other.diff(baseline)) + response = self._my_api().collections().get(uuid=self._manifest_locator).execute(num_retries=num_retries) + other = CollectionReader(response["manifest_text"]) + baseline = CollectionReader(self._manifest_text) + self.apply(baseline.diff(other)) @synchronized def _my_api(self): if self._api_client is None: - self._api_client = arvados.SafeApi(self._config) + self._api_client = ThreadSafeApiCache(self._config) self._keep_client = self._api_client.keep return self._api_client @@ -1168,13 +1101,13 @@ class Collection(SynchronizedCollectionBase): if self._api_client is None: self._my_api() else: - self._keep_client = KeepClient(api=self._api_client) + self._keep_client = KeepClient(api_client=self._api_client) return self._keep_client @synchronized def _my_block_manager(self): if self._block_manager is None: - self._block_manager = BlockManager(self._my_keep()) + self._block_manager = _BlockManager(self._my_keep()) return self._block_manager def _populate_from_api_server(self): @@ -1227,7 +1160,7 @@ class Collection(SynchronizedCollectionBase): error_via_keep = self._populate_from_keep() if self._manifest_text is None: # Nothing worked! - raise arvados.errors.NotFoundError( + raise errors.NotFoundError( ("Failed to retrieve collection '{}' " + "from either API server ({}) or Keep ({})." ).format( @@ -1236,12 +1169,8 @@ class Collection(SynchronizedCollectionBase): error_via_keep)) # populate self._baseline_manifest = self._manifest_text - import_manifest(self._manifest_text, self) + self._import_manifest(self._manifest_text) - if self._sync == SYNC_READONLY: - # Now that we're populated, knowing that this will be readonly, - # forego any further locking. - self.lock = NoopLock() def _has_collection_uuid(self): return self._manifest_locator is not None and re.match(util.collection_uuid_pattern, self._manifest_locator) @@ -1250,80 +1179,117 @@ class Collection(SynchronizedCollectionBase): return self def __exit__(self, exc_type, exc_value, traceback): - """Support scoped auto-commit in a with: block""" - if self._sync != SYNC_READONLY and self._has_collection_uuid(): - self.save() + """Support scoped auto-commit in a with: block.""" + if exc_type is not None: + if self.writable() and self._has_collection_uuid(): + self.save() if self._block_manager is not None: self._block_manager.stop_threads() @synchronized - def clone(self, new_parent=None, new_sync=SYNC_READONLY, new_config=None): + def manifest_locator(self): + """Get the manifest locator. May be None.""" + return self._manifest_locator + + @synchronized + def clone(self, new_parent=None, readonly=False, new_config=None): if new_config is None: new_config = self._config - c = Collection(parent=new_parent, config=new_config, sync=new_sync) - if new_sync == SYNC_READONLY: - c.lock = NoopLock() - c._items = {} - self._cloneinto(c) - return c + if readonly: + newcollection = CollectionReader(parent=new_parent, apiconfig=new_config) + else: + newcollection = Collection(parent=new_parent, apiconfig=new_config) + + newcollection._clonefrom(self) + return newcollection @synchronized def api_response(self): - """ - api_response() -> dict or None + """Returns information about this Collection fetched from the API server. - Returns information about this Collection fetched from the API server. If the Collection exists in Keep but not the API server, currently returns None. Future versions may provide a synthetic response. + """ return self._api_response + def find_or_create(self, path, create_type): + """See `SynchronizedCollectionBase.find_or_create`""" + if path == ".": + return self + else: + return super(Collection, self).find_or_create(path[2:] if path.startswith("./") else path, create_type) + + def find(self, path): + """See `SynchronizedCollectionBase.find`""" + if path == ".": + return self + else: + return super(Collection, self).find(path[2:] if path.startswith("./") else path) + + def remove(self, path, recursive=False): + """See `SynchronizedCollectionBase.remove`""" + if path == ".": + raise errors.ArgumentError("Cannot remove '.'") + else: + return super(Collection, self).remove(path[2:] if path.startswith("./") else path, recursive) + @must_be_writable @synchronized @retry_method def save(self, merge=True, num_retries=None): - """Commit pending buffer blocks to Keep, merge with remote record (if + """Save collection to an existing collection record. + + Commit pending buffer blocks to Keep, merge with remote record (if update=True), write the manifest to Keep, and update the collection - record. 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()`. + record. + + 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()`. - :update: + :merge: Update and merge remote changes before saving. Otherwise, any remote changes will be ignored and overwritten. + :num_retries: + Retry count on API calls (if None, use the collection default) + """ if self.modified(): if not self._has_collection_uuid(): - raise AssertionError("Collection manifest_locator must be a collection uuid. Use save_as() for new collections.") + raise AssertionError("Collection manifest_locator must be a collection uuid. Use save_new() for new collections.") self._my_block_manager().commit_all() if merge: self.update() self._my_keep().put(self.manifest_text(strip=True), num_retries=num_retries) - mt = self.manifest_text(strip=False) + text = self.manifest_text(strip=False) self._api_response = self._my_api().collections().update( uuid=self._manifest_locator, - body={'manifest_text': mt} + body={'manifest_text': text} ).execute( num_retries=num_retries) - self._manifest_text = mt + self._manifest_text = self._api_response["manifest_text"] self.set_unmodified() + @must_be_writable @synchronized @retry_method def save_new(self, name=None, create_collection_record=True, owner_uuid=None, ensure_unique_name=False, num_retries=None): - """Commit pending buffer blocks to Keep, write the manifest to Keep, and create - a new collection record (if create_collection_record True). After - creating a new collection record, this Collection object will be - associated with the new record for `save()` and SYNC_LIVE updates. + """Save collection to a new collection record. + + Commit pending buffer blocks to Keep, write the manifest to Keep, and + create a new collection record (if create_collection_record True). + After creating a new collection record, this Collection object will be + associated with the new record used by `save()`. :name: The collection name. - :keep_only: - Only save the manifest to keep, do not create a collection record. + :create_collection_record: + If True, create a collection record. If False, only save the manifest to keep. :owner_uuid: the user, or project uuid that will own this collection. @@ -1334,31 +1300,29 @@ class Collection(SynchronizedCollectionBase): if it conflicts with a collection with the same name and owner. If False, a name conflict will result in an error. + :num_retries: + Retry count on API calls (if None, use the collection default) + """ self._my_block_manager().commit_all() self._my_keep().put(self.manifest_text(strip=True), num_retries=num_retries) - mt = self.manifest_text(strip=False) + text = self.manifest_text(strip=False) if create_collection_record: if name is None: name = "Collection created %s" % (time.strftime("%Y-%m-%d %H:%M:%S %Z", time.localtime())) - body = {"manifest_text": mt, + body = {"manifest_text": text, "name": name} if owner_uuid: body["owner_uuid"] = owner_uuid self._api_response = self._my_api().collections().create(ensure_unique_name=ensure_unique_name, body=body).execute(num_retries=num_retries) - - if self.events: - self.events.unsubscribe(filters=[["object_uuid", "=", self._manifest_locator]]) + text = self._api_response["manifest_text"] self._manifest_locator = self._api_response["uuid"] - if self.events: - self.events.subscribe(filters=[["object_uuid", "=", self._manifest_locator]]) - - self._manifest_text = mt + self._manifest_text = text self.set_unmodified() @synchronized @@ -1374,19 +1338,87 @@ class Collection(SynchronizedCollectionBase): for c in self.callbacks: c(event, collection, name, item) + @synchronized + def _import_manifest(self, manifest_text): + """Import a manifest into a `Collection`. + + :manifest_text: + The manifest text to import from. + + """ + if len(self) > 0: + raise ArgumentError("Can only import manifest into an empty collection") + + STREAM_NAME = 0 + BLOCKS = 1 + SEGMENTS = 2 + + stream_name = None + state = STREAM_NAME + + for token_and_separator in re.finditer(r'(\S+)(\s+|$)', manifest_text): + tok = token_and_separator.group(1) + sep = token_and_separator.group(2) + + if state == STREAM_NAME: + # starting a new stream + stream_name = tok.replace('\\040', ' ') + blocks = [] + segments = [] + streamoffset = 0L + state = BLOCKS + continue + + if state == BLOCKS: + block_locator = re.match(r'[0-9a-f]{32}\+(\d+)(\+\S+)*', tok) + if block_locator: + blocksize = long(block_locator.group(1)) + blocks.append(Range(tok, streamoffset, blocksize)) + streamoffset += blocksize + else: + state = SEGMENTS + + if state == SEGMENTS: + file_segment = re.search(r'^(\d+):(\d+):(\S+)', tok) + if file_segment: + pos = long(file_segment.group(1)) + size = long(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) + if isinstance(afile, ArvadosFile): + afile.add_segment(blocks, pos, size) + else: + raise errors.SyntaxError("File %s conflicts with stream of the same name.", filepath) + else: + # error! + raise errors.SyntaxError("Invalid manifest format") + + if sep == "\n": + stream_name = None + state = STREAM_NAME + + self.set_unmodified() + + class Subcollection(SynchronizedCollectionBase): """This is a subdirectory within a collection that doesn't have its own API - server record. It falls under the umbrella of the root collection.""" + server record. + + It falls under the umbrella of the root collection. + + """ def __init__(self, parent): super(Subcollection, self).__init__(parent) self.lock = self.root_collection().lock + self._manifest_text = None def root_collection(self): return self.parent.root_collection() - def sync_mode(self): - return self.root_collection().sync_mode() + def writable(self): + return self.root_collection().writable() def _my_api(self): return self.root_collection()._my_api() @@ -1397,149 +1429,89 @@ class Subcollection(SynchronizedCollectionBase): def _my_block_manager(self): return self.root_collection()._my_block_manager() - def _populate(self): - self.root_collection()._populate() - def notify(self, event, collection, name, item): return self.root_collection().notify(event, collection, name, item) + def stream_name(self): + for k, v in self.parent.items(): + if v is self: + return os.path.join(self.parent.stream_name(), k) + return '.' + @synchronized def clone(self, new_parent): c = Subcollection(new_parent) - self._cloneinto(c) + c._clonefrom(self) return c -def import_manifest(manifest_text, - into_collection=None, - api_client=None, - keep=None, - num_retries=None, - sync=SYNC_READONLY): - """Import a manifest into a `Collection`. - :manifest_text: - The manifest text to import from. +class CollectionReader(Collection): + """A read-only collection object. - :into_collection: - The `Collection` that will be initialized (must be empty). - If None, create a new `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. - :api_client: - The API client object that will be used when creating a new `Collection` object. + """ + def __init__(self, manifest_locator_or_text, *args, **kwargs): + self._in_init = True + super(CollectionReader, self).__init__(manifest_locator_or_text, *args, **kwargs) + self._in_init = False - :keep: - The keep client object that will be used when creating a new `Collection` object. + # Forego any locking since it should never change once initialized. + self.lock = NoopLock() - :num_retries: - the default number of api client and keep retries on error. + # Backwards compatability with old CollectionReader + # all_streams() and all_files() + self._streams = None - :sync: - Collection sync mode (only if into_collection is None) - """ - if into_collection is not None: - if len(into_collection) > 0: - raise ArgumentError("Can only import manifest into an empty collection") - c = into_collection - else: - c = Collection(api_client=api_client, keep_client=keep, num_retries=num_retries, sync=sync) - - save_sync = c.sync_mode() - c._sync = None - - STREAM_NAME = 0 - BLOCKS = 1 - SEGMENTS = 2 - - stream_name = None - state = STREAM_NAME - - for n in re.finditer(r'(\S+)(\s+|$)', manifest_text): - tok = n.group(1) - sep = n.group(2) - - if state == STREAM_NAME: - # starting a new stream - stream_name = tok.replace('\\040', ' ') - blocks = [] - segments = [] - streamoffset = 0L - state = BLOCKS - continue - - if state == BLOCKS: - s = re.match(r'[0-9a-f]{32}\+(\d+)(\+\S+)*', tok) - if s: - blocksize = long(s.group(1)) - blocks.append(Range(tok, streamoffset, blocksize)) - streamoffset += blocksize - else: - state = SEGMENTS - - if state == SEGMENTS: - s = re.search(r'^(\d+):(\d+):(\S+)', tok) - if s: - pos = long(s.group(1)) - size = long(s.group(2)) - name = s.group(3).replace('\\040', ' ') - f = c.find("%s/%s" % (stream_name, name), create=True) - f.add_segment(blocks, pos, size) - else: - # error! - raise errors.SyntaxError("Invalid manifest format") + def writable(self): + return self._in_init - if sep == "\n": - stream_name = None - state = STREAM_NAME + def _populate_streams(orig_func): + @functools.wraps(orig_func) + def populate_streams_wrapper(self, *args, **kwargs): + # Defer populating self._streams until needed since it creates a copy of the manifest. + if self._streams is None: + if self._manifest_text: + self._streams = [sline.split() + for sline in self._manifest_text.split("\n") + if sline] + else: + self._streams = [] + return orig_func(self, *args, **kwargs) + return populate_streams_wrapper - c.set_unmodified() - c._sync = save_sync - return c + @_populate_streams + def normalize(self): + """Normalize the streams returned by `all_streams`. -def export_manifest(item, stream_name=".", portable_locators=False): - """ - :item: - Create a manifest for `item` (must be a `Collection` or `ArvadosFile`). If - `item` is a is a `Collection`, this will also export subcollections. + This method is kept for backwards compatability and only affects the + behavior of `all_streams()` and `all_files()` - :stream_name: - the name of the stream when exporting `item`. + """ - :portable_locators: - If True, strip any permission hints on block locators. - If False, use block locators as-is. - """ - buf = "" - if isinstance(item, SynchronizedCollectionBase): - stream = {} - sorted_keys = sorted(item.keys()) - for k in [s for s in sorted_keys if isinstance(item[s], ArvadosFile)]: - v = item[k] - st = [] - for s in v.segments(): - loc = s.locator - if loc.startswith("bufferblock"): - loc = v.parent._my_block_manager()._bufferblocks[loc].locator() - if portable_locators: - loc = KeepLocator(loc).stripped() - st.append(LocatorAndRange(loc, locator_block_size(loc), - s.segment_offset, s.range_size)) - stream[k] = st - if stream: - buf += ' '.join(normalize_stream(stream_name, stream)) - buf += "\n" - for k in [s for s in sorted_keys if isinstance(item[s], SynchronizedCollectionBase)]: - buf += export_manifest(item[k], stream_name=os.path.join(stream_name, k), portable_locators=portable_locators) - elif isinstance(item, ArvadosFile): - st = [] - for s in item.segments: - loc = s.locator - if loc.startswith("bufferblock"): - loc = item._bufferblocks[loc].calculate_locator() - if portable_locators: - loc = KeepLocator(loc).stripped() - st.append(LocatorAndRange(loc, locator_block_size(loc), - s.segment_offset, s.range_size)) - stream[stream_name] = st - buf += ' '.join(normalize_stream(stream_name, stream)) - buf += "\n" - return buf + # Rearrange streams + streams = {} + for s in self.all_streams(): + for f in s.all_files(): + streamname, filename = split(s.name() + "/" + f.name()) + if streamname not in streams: + streams[streamname] = {} + if filename not in streams[streamname]: + streams[streamname][filename] = [] + for r in f.segments: + streams[streamname][filename].extend(s.locators_and_ranges(r.locator, r.range_size)) + + self._streams = [normalize_stream(s, streams[s]) + for s in sorted(streams)] + @_populate_streams + def all_streams(self): + return [StreamReader(s, self._my_keep(), num_retries=self.num_retries) + for s in self._streams] + + @_populate_streams + def all_files(self): + for s in self.all_streams(): + for f in s.all_files(): + yield f