X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/8323e56e9dafb1bd4aee4f548c26854ea3eb05e9..cef69b16e1295e456548813ad95c9db827136c17:/sdk/python/arvados/collection.py diff --git a/sdk/python/arvados/collection.py b/sdk/python/arvados/collection.py index 5b00ea0a2a..20f5c40bfc 100644 --- a/sdk/python/arvados/collection.py +++ b/sdk/python/arvados/collection.py @@ -1,79 +1,31 @@ -import gflags -import httplib -import httplib2 +import functools import logging import os -import pprint -import sys -import types -import subprocess -import json -import UserDict import re -import hashlib -import string -import bz2 -import zlib -import fcntl +import errno 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 * +from .stream import StreamReader, normalize_stream, locator_block_size +from .ranges import Range, LocatorAndRange +from .safeapi import SafeApi import config import errors import util +import events +from arvados.retry import retry_method _logger = logging.getLogger('arvados.collection') -def normalize_stream(s, stream): - stream_tokens = [s] - sortedfiles = list(stream.keys()) - sortedfiles.sort() - - blocks = {} - streamoffset = 0L - for f in sortedfiles: - for b in stream[f]: - if b[arvados.LOCATOR] not in blocks: - stream_tokens.append(b[arvados.LOCATOR]) - blocks[b[arvados.LOCATOR]] = streamoffset - streamoffset += b[arvados.BLOCKSIZE] - - if len(stream_tokens) == 1: - stream_tokens.append(config.EMPTY_BLOCK_LOCATOR) - - for f in sortedfiles: - current_span = None - fout = f.replace(' ', '\\040') - for segment in stream[f]: - segmentoffset = blocks[segment[arvados.LOCATOR]] + segment[arvados.OFFSET] - if current_span == None: - current_span = [segmentoffset, segmentoffset + segment[arvados.SEGMENTSIZE]] - else: - if segmentoffset == current_span[1]: - current_span[1] += segment[arvados.SEGMENTSIZE] - else: - stream_tokens.append("{0}:{1}:{2}".format(current_span[0], current_span[1] - current_span[0], fout)) - current_span = [segmentoffset, segmentoffset + segment[arvados.SEGMENTSIZE]] - - if current_span != None: - stream_tokens.append("{0}:{1}:{2}".format(current_span[0], current_span[1] - current_span[0], fout)) - - if len(stream[f]) == 0: - stream_tokens.append("0:0:{0}".format(fout)) - - return stream_tokens - - class CollectionBase(object): def __enter__(self): - pass + return self - def __exit__(self): + def __exit__(self, exc_type, exc_value, traceback): pass def _my_keep(self): @@ -89,14 +41,18 @@ class CollectionBase(object): hints other than size hints) removed from the locators. """ raw = self.manifest_text() - clean = '' + clean = [] for line in raw.split("\n"): fields = line.split() - if len(fields) > 0: - locators = [ (re.sub(r'\+[^\d][^\+]*', '', x) if re.match(util.keep_locator_pattern, x) else x) - for x in fields[1:-1] ] - clean += fields[0] + ' ' + ' '.join(locators) + ' ' + fields[-1] + "\n" - return clean + if fields: + clean_fields = fields[:1] + [ + (re.sub(r'\+[^\d][^\+]*', '', x) + if re.match(util.keep_locator_pattern, x) + else x) + for x in fields[1:]] + clean += [' '.join(clean_fields), "\n"] + return ''.join(clean) + class CollectionReader(CollectionBase): def __init__(self, manifest_locator_or_text, api_client=None, @@ -135,61 +91,59 @@ class CollectionReader(CollectionBase): 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. - if self._api_client is None: - self._api_client = arvados.api('v1') - self._keep_client = None # Make a new one with the new api. - c = self._api_client.collections().get( - uuid=self._manifest_locator).execute( - num_retries=self.num_retries) - self._manifest_text = c['manifest_text'] + # 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. - self._manifest_text = self._my_keep().get( - self._manifest_locator, num_retries=self.num_retries) + # 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): - if self._streams is not None: - return error_via_api = None error_via_keep = None - should_try_keep = (not self._manifest_text and + should_try_keep = ((self._manifest_text is None) and util.keep_locator_pattern.match( self._manifest_locator)) - if (not self._manifest_text and + if ((self._manifest_text is None) and util.signed_locator_pattern.match(self._manifest_locator)): - try: - self._populate_from_keep() - except e: - error_via_keep = e - if not self._manifest_text: - try: - self._populate_from_api_server() - except Exception as e: - if not should_try_keep: - raise - error_via_api = e - if (not self._manifest_text and + 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 - try: - self._populate_from_keep() - except Exception as e: - error_via_keep = e - if not self._manifest_text: + error_via_keep = self._populate_from_keep() + if self._manifest_text is None: # Nothing worked! raise arvados.errors.NotFoundError( ("Failed to retrieve collection '{}' " + @@ -202,37 +156,74 @@ class CollectionReader(CollectionBase): for sline in self._manifest_text.split("\n") if sline] - def normalize(self): - self._populate() + 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(): - filestream = s.name() + "/" + f.name() - r = filestream.rindex("/") - streamname = filestream[:r] - filename = filestream[r+1:] + 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[0], r[1])) + streams[streamname][filename].extend(s.locators_and_ranges(r.locator, r.range_size)) - self._streams = [] - sortedstreams = list(streams.keys()) - sortedstreams.sort() - for s in sortedstreams: - self._streams.append(normalize_stream(s, streams[s])) + 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]) + self._manifest_text = ''.join( + [StreamReader(stream, keep=self._my_keep()).manifest_text() + for stream in self._streams]) - return self + @_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): - self._populate() return [StreamReader(s, self._my_keep(), num_retries=self.num_retries) for s in self._streams] @@ -241,17 +232,42 @@ class CollectionReader(CollectionBase): for f in s.all_files(): yield f - def manifest_text(self, strip=False): - if strip: + @_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: - self._populate() return self._manifest_text -class CollectionWriter(CollectionBase): - KEEP_BLOCK_SIZE = 2**26 +class _WriterFile(ArvadosFileBase): + def __init__(self, coll_writer, name): + super(_WriterFile, self).__init__(name, 'wb') + self.dest = coll_writer + + def close(self): + super(_WriterFile, self).close() + self.dest.finish_current_file() + + @ArvadosFileBase._before_close + def write(self, data): + self.dest.write(data) + + @ArvadosFileBase._before_close + def writelines(self, seq): + for data in seq: + self.write(data) + @ArvadosFileBase._before_close + def flush(self): + self.dest.flush_data() + + +class CollectionWriter(CollectionBase): def __init__(self, api_client=None, num_retries=0): """Instantiate a CollectionWriter. @@ -285,9 +301,11 @@ class CollectionWriter(CollectionBase): self._queued_file = None self._queued_dirents = deque() self._queued_trees = deque() + self._last_open = None - def __exit__(self): - self.finish() + def __exit__(self, exc_type, exc_value, traceback): + if exc_type is None: + self.finish() def do_queued_work(self): # The work queue consists of three pieces: @@ -315,7 +333,7 @@ class CollectionWriter(CollectionBase): def _work_file(self): while True: - buf = self._queued_file.read(self.KEEP_BLOCK_SIZE) + buf = self._queued_file.read(config.KEEP_BLOCK_SIZE) if not buf: break self.write(buf) @@ -344,10 +362,9 @@ class CollectionWriter(CollectionBase): def _work_trees(self): path, stream_name, max_manifest_depth = self._queued_trees[0] - make_dirents = (util.listdir_recursive if (max_manifest_depth == 0) - else os.listdir) - d = make_dirents(path) - if len(d) > 0: + d = util.listdir_recursive( + path, max_depth = (None if max_manifest_depth == 0 else 0)) + if d: self._queue_dirents(stream_name, d) else: self._queued_trees.popleft() @@ -388,15 +405,44 @@ class CollectionWriter(CollectionBase): self._data_buffer.append(newdata) self._data_buffer_len += len(newdata) self._current_stream_length += len(newdata) - while self._data_buffer_len >= self.KEEP_BLOCK_SIZE: + while self._data_buffer_len >= config.KEEP_BLOCK_SIZE: self.flush_data() + def open(self, streampath, filename=None): + """open(streampath[, filename]) -> file-like object + + Pass in the path of a file to write to the Collection, either as a + single string or as two separate stream name and file name arguments. + This method returns a file-like object you can write to add it to the + Collection. + + You may only have one file object from the Collection open at a time, + so be sure to close the object when you're done. Using the object in + a with statement makes that easy:: + + with cwriter.open('./doc/page1.txt') as outfile: + outfile.write(page1_data) + with cwriter.open('./doc/page2.txt') as outfile: + outfile.write(page2_data) + """ + if filename is None: + streampath, filename = split(streampath) + if self._last_open and not self._last_open.closed: + raise errors.AssertionError( + "can't open '{}' when '{}' is still open".format( + filename, self._last_open.name)) + if streampath != self.current_stream_name(): + self.start_new_stream(streampath) + self.set_current_file_name(filename) + self._last_open = _WriterFile(self, filename) + return self._last_open + def flush_data(self): data_buffer = ''.join(self._data_buffer) if data_buffer: self._current_stream_locators.append( - self._my_keep().put(data_buffer[0:self.KEEP_BLOCK_SIZE])) - self._data_buffer = [data_buffer[self.KEEP_BLOCK_SIZE:]] + self._my_keep().put(data_buffer[0:config.KEEP_BLOCK_SIZE])) + self._data_buffer = [data_buffer[config.KEEP_BLOCK_SIZE:]] self._data_buffer_len = len(self._data_buffer[0]) def start_new_file(self, newfilename=None): @@ -408,13 +454,17 @@ class CollectionWriter(CollectionBase): raise errors.AssertionError( "Manifest filenames cannot contain whitespace: %s" % newfilename) + elif re.search(r'\x00', newfilename): + raise errors.AssertionError( + "Manifest filenames cannot contain NUL characters: %s" % + newfilename) self._current_file_name = newfilename def current_file_name(self): return self._current_file_name def finish_current_file(self): - if self._current_file_name == None: + if self._current_file_name is None: if self._current_file_pos == self._current_stream_length: return raise errors.AssertionError( @@ -485,10 +535,7 @@ class CollectionWriter(CollectionBase): manifest += ' ' + ' '.join("%d:%d:%s" % (sfile[0], sfile[1], sfile[2].replace(' ', '\\040')) for sfile in stream[2]) manifest += "\n" - if manifest: - return manifest - else: - return "" + return manifest def data_locators(self): ret = [] @@ -595,3 +642,967 @@ class ResumableCollectionWriter(CollectionWriter): raise errors.AssertionError( "resumable writer can't accept unsourced data") return super(ResumableCollectionWriter, self).write(data) + +ADD = "add" +DEL = "del" +MOD = "mod" + +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 + self._items = {} + + def _my_api(self): + raise NotImplementedError() + + def _my_keep(self): + raise NotImplementedError() + + def _my_block_manager(self): + raise NotImplementedError() + + def _populate(self): + raise NotImplementedError() + + def sync_mode(self): + raise NotImplementedError() + + def root_collection(self): + raise NotImplementedError() + + def notify(self, event, collection, name, item): + raise NotImplementedError() + + @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. + + """ + if create and self.sync_mode() == SYNC_READONLY: + raise IOError((errno.EROFS, "Collection is read only")) + + p = path.split("/") + if p[0] == '.': + del p[0] + + 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: + # create new file + if create_collection: + item = Subcollection(self) + else: + item = ArvadosFile(self) + self._items[p[0]] = item + self._modified = True + self.notify(ADD, self, p[0], item) + return item + else: + if item is None and create: + # create new collection + item = Subcollection(self) + self._items[p[0]] = item + self._modified = True + self.notify(ADD, self, p[0], item) + del p[0] + if isinstance(item, SynchronizedCollectionBase): + return item.find("/".join(p), create=create) + else: + raise errors.ArgumentError("Interior path components must be subcollection") + else: + return self + + def open(self, path, mode): + """Open a file-like object for access. + + :path: + path to a file in the collection + :mode: + one of "r", "r+", "w", "w+", "a", "a+" + :"r": + opens for reading + :"r+": + opens for reading and writing. Reads/writes share a file pointer. + :"w", "w+": + truncates to 0 and opens for reading and writing. Reads/writes share a file pointer. + :"a", "a+": + opens for reading and writing. All writes are appended to + 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 ArgumentError("Bad mode '%s'" % mode) + create = (mode != "r") + + if create and self.sync_mode() == SYNC_READONLY: + raise IOError((errno.EROFS, "Collection is read only")) + + f = self.find(path, create=create) + + if f is None: + raise IOError((errno.ENOENT, "File not found")) + if not isinstance(f, ArvadosFile): + raise IOError((errno.EISDIR, "Path must refer to a file.")) + + if mode[0] == "w": + f.truncate(0) + + if mode == "r": + return ArvadosFileReader(f, path, mode, num_retries=self.num_retries) + else: + return ArvadosFileWriter(f, path, 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.""" + if self._modified: + return True + for k,v in self._items.items(): + if v.modified(): + return True + return False + + @synchronized + def set_unmodified(self): + """Recursively clear modified flag""" + self._modified = False + for k,v in self._items.items(): + v.set_unmodified() + + @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() + + @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. + """ + 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".""" + return k in self._items + + @synchronized + def __len__(self): + """Get the number of items directly contained in this collection""" + return len(self._items) + + @must_be_writable + @synchronized + def __delitem__(self, p): + """Delete an item by name which is directly contained by this collection.""" + del self._items[p] + self._modified = True + self.notify(DEL, self, p, None) + + @synchronized + def keys(self): + """Get a list of names of files and collections directly contained in this collection.""" + return self._items.keys() + + @synchronized + def values(self): + """Get a list of files and collection objects directly contained in this collection.""" + return self._items.values() + + @synchronized + def items(self): + """Get a list of (name, object) tuples directly contained in this collection.""" + return self._items.items() + + def exists(self, 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): + """Remove the file or subcollection (directory) at `path`. + :rm_r: + 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: + raise IOError((errno.ENOENT, "File not found")) + + def _cloneinto(self, target): + for k,v in self._items.items(): + target._items[k] = v.clone(target) + + def clone(self): + raise NotImplementedError() + + @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 + + :target_path: + Destination file or path. If the target path already exists and is a + subcollection, the item will be placed inside the subcollection. If + the target path already exists and is a file, this will raise an error + unless you specify `overwrite=True`. + + :source_collection: + Collection to copy `source_path` from (default `self`) + + :overwrite: + Whether to overwrite target file if it already exists. + """ + if source_collection is None: + source_collection = self + + # Find the object to copy + if isinstance(source, basestring): + source_obj = source_collection.find(source) + if source_obj is None: + raise IOError((errno.ENOENT, "File not found")) + sp = source.split("/") + else: + source_obj = source + sp = None + + # Find parent collection the target path + tp = target_path.split("/") + + # Determine the name to use. + target_name = tp[-1] if tp[-1] else (sp[-1] if sp 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) + + 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")) + + 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) + + @synchronized + def manifest_text(self, strip=False, normalize=False): + """Get the manifest text for this collection, sub collections and files. + + :strip: + If True, remove signing tokens from block locators if present. + If False, 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 + 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) + else: + if strip: + return self.stripped_manifest() + else: + return self._manifest_text + + @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` + """ + changes = [] + if holding_collection is None: + holding_collection = CollectionRoot(api_client=self._my_api(), keep_client=self._my_keep(), sync=SYNC_READONLY) + for k in self: + if k not in end_collection: + changes.append((DEL, os.path.join(prefix, k), self[k].clone(holding_collection))) + for k in end_collection: + if k in self: + if isinstance(end_collection[k], Subcollection) and isinstance(self[k], Subcollection): + 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((ADD, os.path.join(prefix, k), end_collection[k].clone(holding_collection))) + return changes + + @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. + """ + for c in changes: + path = c[1] + initial = c[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 local is None: + # No local file at path, safe to copy over new file + self.copy(initial, path) + elif local is not None and local != initial: + # There is already local file and it is different: + # save change to conflict file. + self.copy(initial, conflictpath) + elif c[0] == MOD: + if local == initial: + # 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): + # Replace contents of local file with new contents + local.replace_contents(c[3]) + else: + # 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) + 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: + if local == initial: + # Local item matches "initial" value, so it is safe to remove. + self.remove(path, rm_r=True) + # else, the file is modified or already removed, in either + # case we don't want to try to remove it. + + def portable_data_hash(self): + """Get the portable data hash for this collection's manifest.""" + stripped = self.manifest_text(strip=True) + return hashlib.md5(stripped).hexdigest() + '+' + str(len(stripped)) + + @synchronized + def __eq__(self, other): + if other is self: + return True + if not isinstance(other, SynchronizedCollectionBase): + return False + if len(self._items) != len(other): + return False + for k in self._items: + if k not in other: + return False + if self._items[k] != other[k]: + return False + return True + + def __ne__(self, other): + return not self.__eq__(other) + +class CollectionRoot(SynchronizedCollectionBase): + """Represents the root of an Arvados Collection, which may be associated with + an API server Collection record. + + 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()` + + 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). + + """ + + def __init__(self, manifest_locator_or_text=None, + parent=None, + apiconfig=None, + api_client=None, + keep_client=None, + num_retries=None, + block_manager=None, + sync=None): + """: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. + :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 `apiconfig`. + :keep_client: + 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(CollectionRoot, self).__init__(parent) + self._api_client = api_client + self._keep_client = keep_client + self._block_manager = block_manager + + if apiconfig: + self._config = apiconfig + else: + self._config = config.settings() + + self.num_retries = num_retries + self._manifest_locator = None + self._manifest_text = None + self._api_response = None + + if sync is None: + raise errors.ArgumentError("Must specify sync mode") + + self._sync = sync + self.lock = threading.RLock() + self.callbacks = [] + self.events = None + + if manifest_locator_or_text: + if re.match(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): + self._manifest_locator = manifest_locator_or_text + elif re.match(util.manifest_pattern, manifest_locator_or_text): + self._manifest_text = manifest_locator_or_text + else: + 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(apiconfig=self._config), + [["object_uuid", "=", self._manifest_locator]], + self.on_message) + + + def root_collection(self): + return self + + def sync_mode(self): + return self._sync + + def on_message(self, event): + if event.get("object_uuid") == self._manifest_locator: + self.update() + + @staticmethod + def create(name, owner_uuid=None, sync=SYNC_EXPLICIT, apiconfig=None): + """Create a new empty Collection with associated collection record.""" + c = Collection(sync=SYNC_EXPLICIT, apiconfig=apiconfig) + c.save_new(name, owner_uuid=owner_uuid, ensure_unique_name=True) + if sync == SYNC_LIVE: + c.events = events.subscribe(arvados.api(apiconfig=self._config), [["object_uuid", "=", c._manifest_locator]], c.on_message) + return c + + @synchronized + @retry_method + def update(self, other=None, num_retries=None): + 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)) + + @synchronized + def _my_api(self): + if self._api_client is None: + self._api_client = arvados.SafeApi(self._config) + self._keep_client = self._api_client.keep + return self._api_client + + @synchronized + def _my_keep(self): + if self._keep_client is None: + if self._api_client is None: + self._my_api() + else: + self._keep_client = KeepClient(api=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()) + return self._block_manager + + 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: + self._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'] + 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): + 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 arvados.errors.NotFoundError( + ("Failed to retrieve collection '{}' " + + "from either API server ({}) or Keep ({})." + ).format( + self._manifest_locator, + error_via_api, + error_via_keep)) + # populate + self._baseline_manifest = self._manifest_text + import_manifest(self._manifest_text, self) + + 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) + + def __enter__(self): + 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() + 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): + if new_config is None: + new_config = self._config + c = CollectionRoot(parent=new_parent, apiconfig=new_config, sync=new_sync) + if new_sync == SYNC_READONLY: + c.lock = NoopLock() + c._items = {} + self._cloneinto(c) + return c + + @synchronized + 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 + + @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 + 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()`. + + :update: + Update and merge remote changes before saving. Otherwise, any + remote changes will be ignored and overwritten. + + """ + 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.") + 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) + self._api_response = self._my_api().collections().update( + uuid=self._manifest_locator, + body={'manifest_text': mt} + ).execute( + num_retries=num_retries) + self._manifest_text = mt + 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. + + :name: + The collection name. + + :keep_only: + Only save the manifest to keep, do not create a collection record. + + :owner_uuid: + the user, or project uuid that will own this collection. + If None, defaults to the current user. + + :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 + False, a name conflict will result in an error. + + """ + 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) + + 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, + "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]]) + + self._manifest_locator = self._api_response["uuid"] + + if self.events: + self.events.subscribe(filters=[["object_uuid", "=", self._manifest_locator]]) + + self._manifest_text = mt + self.set_unmodified() + + @synchronized + def subscribe(self, callback): + self.callbacks.append(callback) + + @synchronized + def unsubscribe(self, callback): + self.callbacks.remove(callback) + + @synchronized + def notify(self, event, collection, name, item): + for c in self.callbacks: + c(event, collection, name, item) + +def ReadOnlyCollection(*args, **kwargs): + kwargs["sync"] = SYNC_READONLY + return CollectionRoot(*args, **kwargs) + +def WritableCollection(*args, **kwargs): + kwargs["sync"] = SYNC_EXPLICIT + return CollectionRoot(*args, **kwargs) + +def LiveCollection(*args, **kwargs): + kwargs["sync"] = SYNC_LIVE + return CollectionRoot(*args, **kwargs) + + +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.""" + + def __init__(self, parent): + super(Subcollection, self).__init__(parent) + self.lock = self.root_collection().lock + + def root_collection(self): + return self.parent.root_collection() + + def sync_mode(self): + return self.root_collection().sync_mode() + + def _my_api(self): + return self.root_collection()._my_api() + + def _my_keep(self): + return self.root_collection()._my_keep() + + 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) + + @synchronized + def clone(self, new_parent): + c = Subcollection(new_parent) + self._cloneinto(c) + 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. + + :into_collection: + The `Collection` that will be initialized (must be empty). + If None, create a new `Collection` object. + + :api_client: + The API client object that will be used when creating a new `Collection` object. + + :keep: + The keep client object that will be used when creating a new `Collection` object. + + :num_retries: + the default number of api client and keep retries on error. + + :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 = CollectionRoot(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") + + if sep == "\n": + stream_name = None + state = STREAM_NAME + + c.set_unmodified() + c._sync = save_sync + return c + +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. + + :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