X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/f69d2824c997c53caa11d30ba816768bad52e12b..2e437823d2d19edf5c6be392f3604a499308cd6e:/sdk/python/arvados/collection.py diff --git a/sdk/python/arvados/collection.py b/sdk/python/arvados/collection.py index c90f7b3bc7..8450bd1ca0 100644 --- a/sdk/python/arvados/collection.py +++ b/sdk/python/arvados/collection.py @@ -474,6 +474,7 @@ class ResumableCollectionWriter(CollectionWriter): ADD = "add" DEL = "del" MOD = "mod" +TOK = "tok" FILE = "file" COLLECTION = "collection" @@ -487,7 +488,7 @@ class RichCollectionBase(CollectionBase): def __init__(self, parent=None): self.parent = parent - self._modified = True + self._committed = False self._callback = None self._items = {} @@ -542,7 +543,7 @@ class RichCollectionBase(CollectionBase): else: item = ArvadosFile(self, pathcomponents[0]) self._items[pathcomponents[0]] = item - self._modified = True + self._committed = False self.notify(ADD, self, pathcomponents[0], item) return item else: @@ -550,12 +551,12 @@ class RichCollectionBase(CollectionBase): # create new collection item = Subcollection(self, pathcomponents[0]) self._items[pathcomponents[0]] = item - self._modified = True + self._committed = False self.notify(ADD, self, pathcomponents[0], item) if isinstance(item, RichCollectionBase): return item.find_or_create(pathcomponents[1], create_type) else: - raise IOError(errno.ENOTDIR, "Interior path components must be subcollection") + raise IOError(errno.ENOTDIR, "Not a directory: '%s'" % pathcomponents[0]) else: return self @@ -568,7 +569,7 @@ class RichCollectionBase(CollectionBase): """ if not path: - raise errors.ArgumentError("Parameter 'path' must not be empty.") + raise errors.ArgumentError("Parameter 'path' is empty.") pathcomponents = path.split("/", 1) item = self._items.get(pathcomponents[0]) @@ -581,15 +582,20 @@ class RichCollectionBase(CollectionBase): else: return item else: - raise IOError(errno.ENOTDIR, "Interior path components must be subcollection") + raise IOError(errno.ENOTDIR, "Is not a directory: %s" % pathcomponents[0]) + @synchronized def mkdirs(self, path): """Recursive subcollection create. - Like `os.mkdirs()`. Will create intermediate subcollections needed to - contain the leaf subcollection path. + Like `os.makedirs()`. Will create intermediate subcollections needed + to contain the leaf subcollection path. """ + + if self.find(path) != None: + raise IOError(errno.EEXIST, "Directory or file exists: '%s'" % path) + return self.find_or_create(path, COLLECTION) def open(self, path, mode="r"): @@ -626,7 +632,7 @@ class RichCollectionBase(CollectionBase): if arvfile is None: raise IOError(errno.ENOENT, "File not found") if not isinstance(arvfile, ArvadosFile): - raise IOError(errno.EISDIR, "Path must refer to a file.") + raise IOError(errno.EISDIR, "Is a directory: %s" % path) if mode[0] == "w": arvfile.truncate(0) @@ -634,26 +640,31 @@ class RichCollectionBase(CollectionBase): name = os.path.basename(path) if mode == "r": - return ArvadosFileReader(arvfile, mode, num_retries=self.num_retries) + return ArvadosFileReader(arvfile, num_retries=self.num_retries) else: return ArvadosFileWriter(arvfile, mode, num_retries=self.num_retries) - @synchronized def modified(self): - """Test if the collection (or any subcollection or file) has been modified.""" - if self._modified: - return True - for k,v in self._items.items(): - if v.modified(): - return True - return False + """Determine if the collection has been modified since last commited.""" + return not self.committed() + + @synchronized + def committed(self): + """Determine if the collection has been committed to the API server.""" + + if self._committed is False: + return False + for v in self._items.values(): + if v.committed() is False: + return False + return True @synchronized - def set_unmodified(self): - """Recursively clear modified flag.""" - self._modified = False + def set_committed(self): + """Recursively set committed flag to True.""" + self._committed = True for k,v in self._items.items(): - v.set_unmodified() + v.set_committed() @synchronized def __iter__(self): @@ -684,7 +695,7 @@ class RichCollectionBase(CollectionBase): def __delitem__(self, p): """Delete an item by name which is directly contained by this collection.""" del self._items[p] - self._modified = True + self._committed = False self.notify(DEL, self, p, None) @synchronized @@ -716,7 +727,7 @@ class RichCollectionBase(CollectionBase): """ if not path: - raise errors.ArgumentError("Parameter 'path' must not be empty.") + raise errors.ArgumentError("Parameter 'path' is empty.") pathcomponents = path.split("/", 1) item = self._items.get(pathcomponents[0]) @@ -727,7 +738,7 @@ class RichCollectionBase(CollectionBase): raise IOError(errno.ENOTEMPTY, "Subcollection not empty") deleteditem = self._items[pathcomponents[0]] del self._items[pathcomponents[0]] - self._modified = True + self._committed = False self.notify(DEL, self, pathcomponents[0], deleteditem) else: item.remove(pathcomponents[1]) @@ -770,13 +781,13 @@ class RichCollectionBase(CollectionBase): # Actually make the move or copy. if reparent: - source_obj.reparent(self, target_name) + source_obj._reparent(self, target_name) item = source_obj else: item = source_obj.clone(self, target_name) self._items[target_name] = item - self._modified = True + self._committed = False if modified_from: self.notify(MOD, self, target_name, (modified_from, item)) @@ -870,7 +881,7 @@ class RichCollectionBase(CollectionBase): source_obj, target_dir, target_name = self._get_src_target(source, target_path, source_collection, False) if not source_obj.writable(): - raise IOError(errno.EROFS, "Source collection must be writable.") + raise IOError(errno.EROFS, "Source collection is read only.") target_dir.add(source_obj, target_name, overwrite, True) def portable_manifest_text(self, stream_name="."): @@ -885,6 +896,7 @@ class RichCollectionBase(CollectionBase): """ return self._get_manifest_text(stream_name, True, True) + @synchronized def manifest_text(self, stream_name=".", strip=False, normalize=False): """Get the manifest text for this collection, sub collections and files. @@ -928,7 +940,7 @@ class RichCollectionBase(CollectionBase): """ - if self.modified() or self._manifest_text is None or normalize: + if not self.committed() or self._manifest_text is None or normalize: stream = {} buf = [] sorted_keys = sorted(self.keys()) @@ -975,6 +987,8 @@ class RichCollectionBase(CollectionBase): changes.extend(self[k].diff(end_collection[k], os.path.join(prefix, k), holding_collection)) elif end_collection[k] != self[k]: changes.append((MOD, os.path.join(prefix, k), self[k].clone(holding_collection, ""), end_collection[k].clone(holding_collection, ""))) + else: + changes.append((TOK, os.path.join(prefix, k), self[k].clone(holding_collection, ""), end_collection[k].clone(holding_collection, ""))) else: changes.append((ADD, os.path.join(prefix, k), end_collection[k].clone(holding_collection, ""))) return changes @@ -988,12 +1002,14 @@ class RichCollectionBase(CollectionBase): alternate path indicating the conflict. """ + if changes: + self._committed = False 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", + conflictpath = "%s~%s~conflict~" % (path, time.strftime("%Y%m%d-%H%M%S", time.gmtime())) if event_type == ADD: if local is None: @@ -1003,7 +1019,7 @@ class RichCollectionBase(CollectionBase): # There is already local file and it is different: # save change to conflict file. self.copy(initial, conflictpath) - elif event_type == MOD: + elif event_type == MOD or event_type == TOK: final = change[3] if local == initial: # Local matches the "initial" item so it has not @@ -1067,6 +1083,12 @@ class RichCollectionBase(CollectionBase): def __ne__(self, other): return not self.__eq__(other) + @synchronized + def flush(self): + """Flush bufferblocks to Keep.""" + for e in self.values(): + e.flush() + class Collection(RichCollectionBase): """Represents the root of an Arvados Collection. @@ -1149,6 +1171,7 @@ class Collection(RichCollectionBase): self._manifest_locator = None self._manifest_text = None self._api_response = None + self._past_versions = set() self.lock = threading.RLock() self.events = None @@ -1162,7 +1185,7 @@ class Collection(RichCollectionBase): self._manifest_text = manifest_locator_or_text else: raise errors.ArgumentError( - "Argument to CollectionReader must be a manifest or a collection UUID") + "Argument to CollectionReader is not a manifest or a collection UUID") try: self._populate() @@ -1178,6 +1201,10 @@ class Collection(RichCollectionBase): def writable(self): return True + @synchronized + def known_past_version(self, modified_at_and_portable_data_hash): + return modified_at_and_portable_data_hash in self._past_versions + @synchronized @retry_method def update(self, other=None, num_retries=None): @@ -1187,9 +1214,19 @@ class Collection(RichCollectionBase): if self._manifest_locator is None: raise errors.ArgumentError("`other` is None but collection does not have a manifest_locator uuid") response = self._my_api().collections().get(uuid=self._manifest_locator).execute(num_retries=num_retries) + if (self.known_past_version((response.get("modified_at"), response.get("portable_data_hash"))) and + response.get("portable_data_hash") != self.portable_data_hash()): + # The record on the server is different from our current one, but we've seen it before, + # so ignore it because it's already been merged. + # However, if it's the same as our current record, proceed with the update, because we want to update + # our tokens. + return + else: + self._past_versions.add((response.get("modified_at"), response.get("portable_data_hash"))) other = CollectionReader(response["manifest_text"]) baseline = CollectionReader(self._manifest_text) self.apply(baseline.diff(other)) + self._manifest_text = self.manifest_text() @synchronized def _my_api(self): @@ -1213,6 +1250,10 @@ class Collection(RichCollectionBase): self._block_manager = _BlockManager(self._my_keep()) return self._block_manager + def _remember_api_response(self, response): + self._api_response = response + self._past_versions.add((response.get("modified_at"), response.get("portable_data_hash"))) + 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 @@ -1222,9 +1263,9 @@ class Collection(RichCollectionBase): # clause, just like any other Collection lookup # failure. Return an exception, or None if successful. try: - self._api_response = self._my_api().collections().get( + self._remember_api_response(self._my_api().collections().get( uuid=self._manifest_locator).execute( - num_retries=self.num_retries) + num_retries=self.num_retries)) self._manifest_text = self._api_response['manifest_text'] return None except Exception as e: @@ -1283,9 +1324,12 @@ class Collection(RichCollectionBase): def __exit__(self, exc_type, exc_value, traceback): """Support scoped auto-commit in a with: block.""" - if exc_type is not None: + if exc_type is None: if self.writable() and self._has_collection_uuid(): self.save() + self.stop_threads() + + def stop_threads(self): if self._block_manager is not None: self._block_manager.stop_threads() @@ -1368,9 +1412,9 @@ class Collection(RichCollectionBase): Retry count on API calls (if None, use the collection default) """ - if self.modified(): + if not self.committed(): if not self._has_collection_uuid(): - raise AssertionError("Collection manifest_locator must be a collection uuid. Use save_new() for new collections.") + raise AssertionError("Collection manifest_locator is not a collection uuid. Use save_new() for new collections.") self._my_block_manager().commit_all() @@ -1378,13 +1422,13 @@ class Collection(RichCollectionBase): self.update() text = self.manifest_text(strip=False) - self._api_response = self._my_api().collections().update( + self._remember_api_response(self._my_api().collections().update( uuid=self._manifest_locator, body={'manifest_text': text} ).execute( - num_retries=num_retries) + num_retries=num_retries)) self._manifest_text = self._api_response["manifest_text"] - self.set_unmodified() + self.set_committed() return self._manifest_text @@ -1429,7 +1473,7 @@ class Collection(RichCollectionBase): if create_collection_record: if name is None: - name = "Collection created %s" % (time.strftime("%Y-%m-%d %H:%M:%S %Z", time.localtime())) + name = "New collection" ensure_unique_name = True body = {"manifest_text": text, @@ -1437,13 +1481,13 @@ class Collection(RichCollectionBase): 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) + 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"] self._manifest_locator = self._api_response["uuid"] self._manifest_text = text - self.set_unmodified() + self.set_committed() return text @@ -1476,7 +1520,7 @@ class Collection(RichCollectionBase): segments = [] streamoffset = 0L state = BLOCKS - self.mkdirs(stream_name) + self.find_or_create(stream_name, COLLECTION) continue if state == BLOCKS: @@ -1502,13 +1546,13 @@ class Collection(RichCollectionBase): raise errors.SyntaxError("File %s conflicts with stream of the same name.", filepath) else: # error! - raise errors.SyntaxError("Invalid manifest format") + raise errors.SyntaxError("Invalid manifest format, expected file segment but did not match format: '%s'" % tok) if sep == "\n": stream_name = None state = STREAM_NAME - self.set_unmodified() + self.set_committed() @synchronized def notify(self, event, collection, name, item): @@ -1520,7 +1564,7 @@ class Subcollection(RichCollectionBase): """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. + Subcollection locking falls under the umbrella lock of its root collection. """ @@ -1555,6 +1599,16 @@ class Subcollection(RichCollectionBase): c._clonefrom(self) return c + @must_be_writable + @synchronized + def _reparent(self, newparent, newname): + self._committed = False + self.flush() + self.parent.remove(self.name, recursive=True) + self.parent = newparent + self.name = newname + self.lock = self.parent.root_collection().lock + class CollectionReader(Collection): """A read-only collection object.