X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/3d1181ac8527cdd355b14872b808cbaede26e487..9756d91f6f1e2472184be0dc45e8849c2c9ab35b:/services/fuse/arvados_fuse/__init__.py diff --git a/services/fuse/arvados_fuse/__init__.py b/services/fuse/arvados_fuse/__init__.py index aecc7a9417..c261bc3197 100644 --- a/services/fuse/arvados_fuse/__init__.py +++ b/services/fuse/arvados_fuse/__init__.py @@ -19,16 +19,10 @@ import logging import time import calendar import threading +from arvados.util import portable_data_hash_pattern, uuid_pattern, collection_uuid_pattern, group_uuid_pattern, user_uuid_pattern, link_uuid_pattern _logger = logging.getLogger('arvados.arvados_fuse') -portable_data_hash_pattern = re.compile(r'[0-9a-f]{32}\+\d+') -uuid_pattern = re.compile(r'[a-z0-9]{5}-[a-z0-9]{5}-[a-z0-9]{15}') -collection_uuid_pattern = re.compile(r'[a-z0-9]{5}-4zz18-[a-z0-9]{15}') -group_uuid_pattern = re.compile(r'[a-z0-9]{5}-j7d0g-[a-z0-9]{15}') -user_uuid_pattern = re.compile(r'[a-z0-9]{5}-tpzed-[a-z0-9]{15}') -link_uuid_pattern = re.compile(r'[a-z0-9]{5}-o0j2j-[a-z0-9]{15}') - class SafeApi(object): '''Threadsafe wrapper for API object. This stores and returns a different api object per thread, because httplib2 which underlies apiclient is not @@ -37,30 +31,36 @@ class SafeApi(object): def __init__(self, config): self.host = config.get('ARVADOS_API_HOST') - self.token = config.get('ARVADOS_API_TOKEN') + self.api_token = config.get('ARVADOS_API_TOKEN') self.insecure = config.flag_is_true('ARVADOS_API_HOST_INSECURE') self.local = threading.local() + self.block_cache = arvados.KeepBlockCache() def localapi(self): if 'api' not in self.local.__dict__: - self.local.api = arvados.api('v1', False, self.host, self.token, self.insecure) + self.local.api = arvados.api('v1', False, self.host, + self.api_token, self.insecure) return self.local.api - def collections(self): - return self.localapi().collections() + def localkeep(self): + if 'keep' not in self.local.__dict__: + self.local.keep = arvados.KeepClient(api_client=self.localapi(), block_cache=self.block_cache) + return self.local.keep - def links(self): - return self.localapi().links() + def __getattr__(self, name): + # Proxy nonexistent attributes to the local API client. + try: + return getattr(self.localapi(), name) + except AttributeError: + return super(SafeApi, self).__getattr__(name) - def groups(self): - return self.localapi().groups() - def users(self): - return self.localapi().users() - def convertTime(t): '''Parse Arvados timestamp to unix time.''' - return calendar.timegm(time.strptime(t, "%Y-%m-%dT%H:%M:%SZ")) + try: + return calendar.timegm(time.strptime(t, "%Y-%m-%dT%H:%M:%SZ")) + except (TypeError, ValueError): + return 0 def sanitize_filename(dirty): '''Remove troublesome characters from filenames.''' @@ -96,9 +96,8 @@ class FreshBase(object): def invalidate(self): self._stale = True - # Test if the entries dict is stale. Also updates atime. + # Test if the entries dict is stale. def stale(self): - self._atime = time.time() if self._stale: return True if self._poll: @@ -158,7 +157,7 @@ class StringFile(File): return len(self.contents) def readfrom(self, off, size): - return self.contents[off:(off+size)] + return self.contents[off:(off+size)] class ObjectFile(StringFile): @@ -259,10 +258,10 @@ class Directory(FreshBase): ent = new_entry(i) if ent is not None: self._entries[name] = self.inodes.add_entry(ent) - changed = True + changed = True # delete any other directory entries that were not in found in 'items' - for i in oldentries: + for i in oldentries: llfuse.invalidate_entry(self.inode, str(i)) self.inodes.del_entry(oldentries[i]) changed = True @@ -290,10 +289,11 @@ class Directory(FreshBase): class CollectionDirectory(Directory): '''Represents the root of a directory tree holding a collection.''' - def __init__(self, parent_inode, inodes, api, collection): + def __init__(self, parent_inode, inodes, api, num_retries, collection): super(CollectionDirectory, self).__init__(parent_inode) self.inodes = inodes self.api = api + self.num_retries = num_retries self.collection_object_file = None self.collection_object = None if isinstance(collection, dict): @@ -311,7 +311,9 @@ class CollectionDirectory(Directory): self.collection_object_file.update(self.collection_object) self.clear() - collection = arvados.CollectionReader(self.collection_object["manifest_text"], self.api) + collection = arvados.CollectionReader( + self.collection_object["manifest_text"], self.api, + self.api.localkeep(), num_retries=self.num_retries) for s in collection.all_streams(): cwd = self for part in s.name().split('/'): @@ -321,7 +323,7 @@ class CollectionDirectory(Directory): cwd._entries[partname] = self.inodes.add_entry(Directory(cwd.inode)) cwd = cwd._entries[partname] for k, v in s.files().items(): - cwd._entries[sanitize_filename(k)] = self.inodes.add_entry(StreamReaderFile(cwd.inode, v, self.mtime())) + cwd._entries[sanitize_filename(k)] = self.inodes.add_entry(StreamReaderFile(cwd.inode, v, self.mtime())) def update(self): try: @@ -329,7 +331,9 @@ class CollectionDirectory(Directory): return True with llfuse.lock_released: - new_collection_object = self.api.collections().get(uuid=self.collection_locator).execute() + new_collection_object = self.api.collections().get( + uuid=self.collection_locator + ).execute(num_retries=self.num_retries) if "portable_data_hash" not in new_collection_object: new_collection_object["portable_data_hash"] = new_collection_object["uuid"] # end with llfuse.lock_released, re-acquire lock @@ -345,11 +349,15 @@ class CollectionDirectory(Directory): else: _logger.error("arv-mount %s: error", self.collection_locator) _logger.exception(detail) + except arvados.errors.ArgumentError as detail: + _logger.warning("arv-mount %s: error %s", self.collection_locator, detail) + if self.collection_object is not None and "manifest_text" in self.collection_object: + _logger.warning("arv-mount manifest_text is: %s", self.collection_object["manifest_text"]) except Exception as detail: _logger.error("arv-mount %s: error", self.collection_locator) - if "manifest_text" in self.collection_object: + if self.collection_object is not None and "manifest_text" in self.collection_object: _logger.error("arv-mount manifest_text is: %s", self.collection_object["manifest_text"]) - _logger.exception(detail) + _logger.exception(detail) return False def __getitem__(self, item): @@ -383,13 +391,16 @@ class MagicDirectory(Directory): to readdir(). ''' - def __init__(self, parent_inode, inodes, api): + def __init__(self, parent_inode, inodes, api, num_retries): super(MagicDirectory, self).__init__(parent_inode) self.inodes = inodes self.api = api + self.num_retries = num_retries + # Have to defer creating readme_file because at this point we don't + # yet have an inode assigned. self.readme_file = None - def __contains__(self, k): + def create_readme(self): if self.readme_file is None: text = '''This directory provides access to Arvados collections as subdirectories listed by uuid (in the form 'zzzzz-4zz18-1234567890abcde') or portable data hash (in @@ -403,6 +414,9 @@ will appear if it exists. self.readme_file = self.inodes.add_entry(StringFile(self.inode, text, time.time())) self._entries["README"] = self.readme_file + def __contains__(self, k): + self.create_readme() + if k in self._entries: return True @@ -410,7 +424,8 @@ will appear if it exists. return False try: - e = self.inodes.add_entry(CollectionDirectory(self.inode, self.inodes, self.api, k)) + e = self.inodes.add_entry(CollectionDirectory( + self.inode, self.inodes, self.api, self.num_retries, k)) if e.update(): self._entries[k] = e return True @@ -420,6 +435,10 @@ will appear if it exists. _logger.debug('arv-mount exception keep %s', e) return False + def items(self): + self.create_readme() + return self._entries.items() + def __getitem__(self, item): if item in self: return self._entries[item] @@ -445,21 +464,25 @@ class RecursiveInvalidateDirectory(Directory): class TagsDirectory(RecursiveInvalidateDirectory): '''A special directory that contains as subdirectories all tags visible to the user.''' - def __init__(self, parent_inode, inodes, api, poll_time=60): + def __init__(self, parent_inode, inodes, api, num_retries, poll_time=60): super(TagsDirectory, self).__init__(parent_inode) self.inodes = inodes self.api = api + self.num_retries = num_retries self._poll = True self._poll_time = poll_time def update(self): with llfuse.lock_released: - tags = self.api.links().list(filters=[['link_class', '=', 'tag']], select=['name'], distinct = True).execute() + tags = self.api.links().list( + filters=[['link_class', '=', 'tag']], + select=['name'], distinct=True + ).execute(num_retries=self.num_retries) if "items" in tags: self.merge(tags['items'], lambda i: i['name'] if 'name' in i else i['uuid'], lambda a, i: a.tag == i, - lambda i: TagDirectory(self.inode, self.inodes, self.api, i['name'], poll=self._poll, poll_time=self._poll_time)) + lambda i: TagDirectory(self.inode, self.inodes, self.api, self.num_retries, i['name'], poll=self._poll, poll_time=self._poll_time)) class TagDirectory(Directory): @@ -467,46 +490,51 @@ class TagDirectory(Directory): to the user that are tagged with a particular tag. ''' - def __init__(self, parent_inode, inodes, api, tag, poll=False, poll_time=60): + def __init__(self, parent_inode, inodes, api, num_retries, tag, + poll=False, poll_time=60): super(TagDirectory, self).__init__(parent_inode) self.inodes = inodes self.api = api + self.num_retries = num_retries self.tag = tag self._poll = poll self._poll_time = poll_time def update(self): with llfuse.lock_released: - taggedcollections = self.api.links().list(filters=[['link_class', '=', 'tag'], - ['name', '=', self.tag], - ['head_uuid', 'is_a', 'arvados#collection']], - select=['head_uuid']).execute() + taggedcollections = self.api.links().list( + filters=[['link_class', '=', 'tag'], + ['name', '=', self.tag], + ['head_uuid', 'is_a', 'arvados#collection']], + select=['head_uuid'] + ).execute(num_retries=self.num_retries) self.merge(taggedcollections['items'], lambda i: i['head_uuid'], lambda a, i: a.collection_locator == i['head_uuid'], - lambda i: CollectionDirectory(self.inode, self.inodes, self.api, i['head_uuid'])) + lambda i: CollectionDirectory(self.inode, self.inodes, self.api, self.num_retries, i['head_uuid'])) -class ProjectDirectory(RecursiveInvalidateDirectory): +class ProjectDirectory(Directory): '''A special directory that contains the contents of a project.''' - def __init__(self, parent_inode, inodes, api, project_object, poll=False, poll_time=60): + def __init__(self, parent_inode, inodes, api, num_retries, project_object, + poll=False, poll_time=60): super(ProjectDirectory, self).__init__(parent_inode) self.inodes = inodes self.api = api + self.num_retries = num_retries self.project_object = project_object - self.project_object_file = ObjectFile(self.inode, self.project_object) - self.inodes.add_entry(self.project_object_file) + self.project_object_file = None self.uuid = project_object['uuid'] def createDirectory(self, i): if collection_uuid_pattern.match(i['uuid']): - return CollectionDirectory(self.inode, self.inodes, self.api, i) + return CollectionDirectory(self.inode, self.inodes, self.api, self.num_retries, i) elif group_uuid_pattern.match(i['uuid']): - return ProjectDirectory(self.inode, self.inodes, self.api, i, self._poll, self._poll_time) + return ProjectDirectory(self.inode, self.inodes, self.api, self.num_retries, i, self._poll, self._poll_time) elif link_uuid_pattern.match(i['uuid']): if i['head_kind'] == 'arvados#collection' or portable_data_hash_pattern.match(i['head_uuid']): - return CollectionDirectory(self.inode, self.inodes, self.api, i['head_uuid']) + return CollectionDirectory(self.inode, self.inodes, self.api, self.num_retries, i['head_uuid']) else: return None elif uuid_pattern.match(i['uuid']): @@ -515,6 +543,10 @@ class ProjectDirectory(RecursiveInvalidateDirectory): return None def update(self): + if self.project_object_file == None: + self.project_object_file = ObjectFile(self.inode, self.project_object) + self.inodes.add_entry(self.project_object_file) + def namefn(i): if 'name' in i: if i['name'] is None or len(i['name']) == 0: @@ -527,7 +559,7 @@ class ProjectDirectory(RecursiveInvalidateDirectory): return i['name'] elif 'kind' in i and i['kind'].startswith('arvados#'): # something else - return "{}.{}".format(i['name'], i['kind'][8:]) + return "{}.{}".format(i['name'], i['kind'][8:]) else: return None @@ -542,14 +574,20 @@ class ProjectDirectory(RecursiveInvalidateDirectory): with llfuse.lock_released: if group_uuid_pattern.match(self.uuid): - self.project_object = self.api.groups().get(uuid=self.uuid).execute() + self.project_object = self.api.groups().get( + uuid=self.uuid).execute(num_retries=self.num_retries) elif user_uuid_pattern.match(self.uuid): - self.project_object = self.api.users().get(uuid=self.uuid).execute() + self.project_object = self.api.users().get( + uuid=self.uuid).execute(num_retries=self.num_retries) - contents = arvados.util.list_all(self.api.groups().contents, uuid=self.uuid) + contents = arvados.util.list_all(self.api.groups().contents, + self.num_retries, uuid=self.uuid) # Name links will be obsolete soon, take this out when there are no more pre-#3036 in use. - contents += arvados.util.list_all(self.api.links().list, filters=[['tail_uuid', '=', self.uuid], ['link_class', '=', 'name']]) - + contents += arvados.util.list_all( + self.api.links().list, self.num_retries, + filters=[['tail_uuid', '=', self.uuid], + ['link_class', '=', 'name']]) + # end with llfuse.lock_released, re-acquire lock self.merge(contents, @@ -571,20 +609,24 @@ class ProjectDirectory(RecursiveInvalidateDirectory): return super(ProjectDirectory, self).__contains__(k) -class SharedDirectory(RecursiveInvalidateDirectory): +class SharedDirectory(Directory): '''A special directory that represents users or groups who have shared projects with me.''' - def __init__(self, parent_inode, inodes, api, exclude, poll=False, poll_time=60): + def __init__(self, parent_inode, inodes, api, num_retries, exclude, + poll=False, poll_time=60): super(SharedDirectory, self).__init__(parent_inode) - self.current_user = api.users().current().execute() self.inodes = inodes self.api = api + self.num_retries = num_retries + self.current_user = api.users().current().execute(num_retries=num_retries) self._poll = True self._poll_time = poll_time def update(self): with llfuse.lock_released: - all_projects = arvados.util.list_all(self.api.groups().list, filters=[['group_class','=','project']]) + all_projects = arvados.util.list_all( + self.api.groups().list, self.num_retries, + filters=[['group_class','=','project']]) objects = {} for ob in all_projects: objects[ob['uuid']] = ob @@ -596,8 +638,12 @@ class SharedDirectory(RecursiveInvalidateDirectory): roots.append(ob) root_owners[ob['owner_uuid']] = True - lusers = arvados.util.list_all(self.api.users().list, filters=[['uuid','in', list(root_owners)]]) - lgroups = arvados.util.list_all(self.api.groups().list, filters=[['uuid','in', list(root_owners)]]) + lusers = arvados.util.list_all( + self.api.users().list, self.num_retries, + filters=[['uuid','in', list(root_owners)]]) + lgroups = arvados.util.list_all( + self.api.groups().list, self.num_retries, + filters=[['uuid','in', list(root_owners)]]) users = {} groups = {} @@ -626,7 +672,7 @@ class SharedDirectory(RecursiveInvalidateDirectory): self.merge(contents.items(), lambda i: i[0], lambda a, i: a.uuid == i[1]['uuid'], - lambda i: ProjectDirectory(self.inode, self.inodes, self.api, i[1], poll=self._poll, poll_time=self._poll_time)) + lambda i: ProjectDirectory(self.inode, self.inodes, self.api, self.num_retries, i[1], poll=self._poll, poll_time=self._poll_time)) except Exception as e: _logger.exception(e) @@ -682,12 +728,13 @@ class Operations(llfuse.Operations): so request handlers do not run concurrently unless the lock is explicitly released using "with llfuse.lock_released:"''' - def __init__(self, uid, gid): + def __init__(self, uid, gid, encoding="utf-8"): super(Operations, self).__init__() self.inodes = Inodes() self.uid = uid self.gid = gid + self.encoding = encoding # dict of inode to filehandle self._filehandles = {} @@ -739,6 +786,7 @@ class Operations(llfuse.Operations): return entry def lookup(self, parent_inode, name): + name = unicode(name, self.encoding) _logger.debug("arv-mount lookup: parent_inode %i name %s", parent_inode, name) inode = None @@ -750,7 +798,7 @@ class Operations(llfuse.Operations): p = self.inodes[parent_inode] if name == '..': inode = p.parent_inode - elif name in p: + elif isinstance(p, Directory) and name in p: inode = p[name].inode if inode != None: @@ -782,10 +830,17 @@ class Operations(llfuse.Operations): else: raise llfuse.FUSEError(errno.EBADF) + # update atime + handle.entry._atime = time.time() + try: with llfuse.lock_released: return handle.entry.readfrom(off, size) - except: + except arvados.errors.NotFoundError as e: + _logger.warning("Block not found: " + str(e)) + raise llfuse.FUSEError(errno.EIO) + except Exception as e: + _logger.exception(e) raise llfuse.FUSEError(errno.EIO) def release(self, fh): @@ -810,6 +865,9 @@ class Operations(llfuse.Operations): else: raise llfuse.FUSEError(errno.EIO) + # update atime + p._atime = time.time() + self._filehandles[fh] = FileHandle(fh, [('.', p), ('..', parent)] + list(p.items())) return fh @@ -826,7 +884,10 @@ class Operations(llfuse.Operations): e = off while e < len(handle.entry): if handle.entry[e][1].inode in self.inodes: - yield (handle.entry[e][0], self.getattr(handle.entry[e][1].inode), e+1) + try: + yield (handle.entry[e][0].encode(self.encoding), self.getattr(handle.entry[e][1].inode), e+1) + except UnicodeEncodeError: + pass e += 1 def releasedir(self, fh):