diff --git a/swh/indexer/indexer.py b/swh/indexer/indexer.py index 7651421..66eabc9 100644 --- a/swh/indexer/indexer.py +++ b/swh/indexer/indexer.py @@ -1,657 +1,657 @@ # Copyright (C) 2016-2022 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import abc from contextlib import contextmanager import logging import os import shutil import tempfile from typing import ( Any, Dict, Generic, Iterable, Iterator, List, Optional, Set, Tuple, TypeVar, Union, ) import warnings import sentry_sdk from typing_extensions import TypedDict from swh.core import utils from swh.core.config import load_from_envvar, merge_configs from swh.indexer.storage import INDEXER_CFG_KEY, Sha1, get_indexer_storage from swh.indexer.storage.interface import IndexerStorageInterface from swh.model import hashutil from swh.model.model import Directory, Origin, Sha1Git from swh.objstorage.exc import ObjNotFoundError from swh.objstorage.factory import get_objstorage from swh.scheduler import CONFIG as SWH_CONFIG from swh.storage import get_storage from swh.storage.interface import StorageInterface class ObjectsDict(TypedDict, total=False): directory: List[Dict] origin: List[Dict] origin_visit_status: List[Dict] @contextmanager def write_to_temp(filename: str, data: bytes, working_directory: str) -> Iterator[str]: """Write the sha1's content in a temporary file. Args: filename: one of sha1's many filenames data: the sha1's content to write in temporary file working_directory: the directory into which the file is written Returns: The path to the temporary file created. That file is filled in with the raw content's data. """ os.makedirs(working_directory, exist_ok=True) temp_dir = tempfile.mkdtemp(dir=working_directory) content_path = os.path.join(temp_dir, filename) with open(content_path, "wb") as f: f.write(data) yield content_path shutil.rmtree(temp_dir) DEFAULT_CONFIG = { INDEXER_CFG_KEY: {"cls": "memory"}, "storage": {"cls": "memory"}, "objstorage": {"cls": "memory"}, } TId = TypeVar("TId") """type of the ids of index()ed objects.""" TData = TypeVar("TData") """type of the objects passed to index().""" TResult = TypeVar("TResult") """return type of index()""" class BaseIndexer(Generic[TId, TData, TResult], metaclass=abc.ABCMeta): """Base class for indexers to inherit from. The main entry point is the :func:`run` function which is in charge of triggering the computations on the batch dict/ids received. Indexers can: - filter out ids whose data has already been indexed. - retrieve ids data from storage or objstorage - index this data depending on the object and store the result in storage. To implement a new object type indexer, inherit from the BaseIndexer and implement indexing: :meth:`~BaseIndexer.run`: object_ids are different depending on object. For example: sha1 for content, sha1_git for revision, directory, release, and id for origin To implement a new concrete indexer, inherit from the object level classes: :class:`ContentIndexer`, :class:`DirectoryIndexer`, :class:`OriginIndexer`. Then you need to implement the following functions: :meth:`~BaseIndexer.filter`: filter out data already indexed (in storage). :meth:`~BaseIndexer.index_object`: compute index on id with data (retrieved from the storage or the objstorage by the id key) and return the resulting index computation. :meth:`~BaseIndexer.persist_index_computations`: persist the results of multiple index computations in the storage. The new indexer implementation can also override the following functions: :meth:`~BaseIndexer.prepare`: Configuration preparation for the indexer. When overriding, this must call the `super().prepare()` instruction. :meth:`~BaseIndexer.check`: Configuration check for the indexer. When overriding, this must call the `super().check()` instruction. :meth:`~BaseIndexer.register_tools`: This should return a dict of the tool(s) to use when indexing or filtering. """ results: List[TResult] USE_TOOLS = True catch_exceptions = True """Prevents exceptions in `index()` from raising too high. Set to False in tests to properly catch all exceptions.""" scheduler: Any storage: StorageInterface objstorage: Any idx_storage: IndexerStorageInterface def __init__(self, config=None, **kw) -> None: """Prepare and check that the indexer is ready to run.""" super().__init__() if config is not None: self.config = config elif SWH_CONFIG: self.config = SWH_CONFIG.copy() else: self.config = load_from_envvar() self.config = merge_configs(DEFAULT_CONFIG, self.config) self.prepare() self.check() self.log.debug("%s: config=%s", self, self.config) def prepare(self) -> None: """Prepare the indexer's needed runtime configuration. Without this step, the indexer cannot possibly run. """ config_storage = self.config.get("storage") if config_storage: self.storage = get_storage(**config_storage) self.objstorage = get_objstorage(**self.config["objstorage"]) idx_storage = self.config[INDEXER_CFG_KEY] self.idx_storage = get_indexer_storage(**idx_storage) _log = logging.getLogger("requests.packages.urllib3.connectionpool") _log.setLevel(logging.WARN) self.log = logging.getLogger("swh.indexer") if self.USE_TOOLS: self.tools = list(self.register_tools(self.config.get("tools", []))) self.results = [] @property def tool(self) -> Dict: return self.tools[0] def check(self) -> None: """Check the indexer's configuration is ok before proceeding. If ok, does nothing. If not raise error. """ if self.USE_TOOLS and not self.tools: raise ValueError("Tools %s is unknown, cannot continue" % self.tools) def _prepare_tool(self, tool: Dict[str, Any]) -> Dict[str, Any]: """Prepare the tool dict to be compliant with the storage api.""" return {"tool_%s" % key: value for key, value in tool.items()} def register_tools( self, tools: Union[Dict[str, Any], List[Dict[str, Any]]] ) -> List[Dict[str, Any]]: """Permit to register tools to the storage. Add a sensible default which can be overridden if not sufficient. (For now, all indexers use only one tool) Expects the self.config['tools'] property to be set with one or more tools. Args: tools: Either a dict or a list of dict. Returns: list: List of dicts with additional id key. Raises: ValueError: if not a list nor a dict. """ if isinstance(tools, list): tools = list(map(self._prepare_tool, tools)) elif isinstance(tools, dict): tools = [self._prepare_tool(tools)] else: raise ValueError("Configuration tool(s) must be a dict or list!") if tools: return self.idx_storage.indexer_configuration_add(tools) else: return [] def index(self, id: TId, data: Optional[TData], **kwargs) -> List[TResult]: """Index computation for the id and associated raw data. Args: id: identifier or Dict object data: id's data from storage or objstorage depending on object type Returns: dict: a dict that makes sense for the :meth:`.persist_index_computations` method. """ raise NotImplementedError() def filter(self, ids: List[TId]) -> Iterator[TId]: """Filter missing ids for that particular indexer. Args: ids: list of ids Yields: iterator of missing ids """ yield from ids @abc.abstractmethod def persist_index_computations(self, results: List[TResult]) -> Dict[str, int]: """Persist the computation resulting from the index. Args: results: List of results. One result is the result of the index function. Returns: a summary dict of what has been inserted in the storage """ return {} class ContentIndexer(BaseIndexer[Sha1, bytes, TResult], Generic[TResult]): """A content indexer working on a list of ids directly. To work on indexer partition, use the :class:`ContentPartitionIndexer` instead. Note: :class:`ContentIndexer` is not an instantiable object. To use it, one should inherit from this class and override the methods mentioned in the :class:`BaseIndexer` class. """ def run(self, ids: List[Sha1], **kwargs) -> Dict: """Given a list of ids: - retrieve the content from the storage - execute the indexing computations - store the results Args: ids (Iterable[Union[bytes, str]]): sha1's identifier list **kwargs: passed to the `index` method Returns: A summary Dict of the task's status """ if "policy_update" in kwargs: warnings.warn( "'policy_update' argument is deprecated and ignored.", DeprecationWarning, ) del kwargs["policy_update"] sha1s = [ hashutil.hash_to_bytes(id_) if isinstance(id_, str) else id_ for id_ in ids ] results = [] summary: Dict = {"status": "uneventful"} try: for sha1 in sha1s: try: raw_content = self.objstorage.get(sha1) except ObjNotFoundError: self.log.warning( "Content %s not found in objstorage" % hashutil.hash_to_hex(sha1) ) continue res = self.index(sha1, raw_content, **kwargs) if res: # If no results, skip it results.extend(res) summary["status"] = "eventful" summary = self.persist_index_computations(results) self.results = results except Exception: if not self.catch_exceptions: raise self.log.exception("Problem when reading contents metadata.") sentry_sdk.capture_exception() summary["status"] = "failed" return summary class ContentPartitionIndexer(BaseIndexer[Sha1, bytes, TResult], Generic[TResult]): """A content partition indexer. This expects as input a partition_id and a nb_partitions. This will then index the contents within that partition. To work on a list of ids, use the :class:`ContentIndexer` instead. Note: :class:`ContentPartitionIndexer` is not an instantiable object. To use it, one should inherit from this class and override the methods mentioned in the :class:`BaseIndexer` class. """ @abc.abstractmethod def indexed_contents_in_partition( self, partition_id: int, nb_partitions: int ) -> Iterable[Sha1]: """Retrieve indexed contents within range [start, end]. Args: partition_id: Index of the partition to fetch nb_partitions: Total number of partitions to split into page_token: opaque token used for pagination """ pass def _list_contents_to_index( self, partition_id: int, nb_partitions: int, indexed: Set[Sha1] ) -> Iterable[Sha1]: """Compute from storage the new contents to index in the partition_id . The already indexed contents are skipped. Args: partition_id: Index of the partition to fetch data from nb_partitions: Total number of partition indexed: Set of content already indexed. Yields: Sha1 id (bytes) of contents to index """ if not isinstance(partition_id, int) or not isinstance(nb_partitions, int): raise TypeError( f"identifiers must be int, not {partition_id!r} and {nb_partitions!r}." ) next_page_token = None while True: result = self.storage.content_get_partition( partition_id, nb_partitions, page_token=next_page_token ) contents = result.results for c in contents: _id = hashutil.hash_to_bytes(c.sha1) if _id in indexed: continue yield _id next_page_token = result.next_page_token if next_page_token is None: break def _index_contents( self, partition_id: int, nb_partitions: int, indexed: Set[Sha1], **kwargs: Any ) -> Iterator[TResult]: """Index the contents within the partition_id. Args: start: Starting bound from range identifier end: End range identifier indexed: Set of content already indexed. Yields: indexing result as dict to persist in the indexer backend """ for sha1 in self._list_contents_to_index(partition_id, nb_partitions, indexed): try: raw_content = self.objstorage.get(sha1) except ObjNotFoundError: self.log.warning(f"Content {sha1.hex()} not found in objstorage") continue yield from self.index(sha1, raw_content, **kwargs) def _index_with_skipping_already_done( self, partition_id: int, nb_partitions: int ) -> Iterator[TResult]: """Index not already indexed contents within the partition partition_id Args: partition_id: Index of the partition to fetch nb_partitions: Total number of partitions to split into Yields: indexing result as dict to persist in the indexer backend """ already_indexed_contents = set( self.indexed_contents_in_partition(partition_id, nb_partitions) ) return self._index_contents( partition_id, nb_partitions, already_indexed_contents ) def run( self, partition_id: int, nb_partitions: int, skip_existing: bool = True, **kwargs, ) -> Dict: """Given a partition of content ids, index the contents within. Either the indexer is incremental (filter out existing computed data) or it computes everything from scratch. Args: partition_id: Index of the partition to fetch nb_partitions: Total number of partitions to split into skip_existing: Skip existing indexed data (default) or not **kwargs: passed to the `index` method Returns: dict with the indexing task status """ summary: Dict[str, Any] = {"status": "uneventful"} count = 0 try: if skip_existing: gen = self._index_with_skipping_already_done( partition_id, nb_partitions ) else: gen = self._index_contents(partition_id, nb_partitions, indexed=set([])) count_object_added_key: Optional[str] = None for contents in utils.grouper(gen, n=self.config["write_batch_size"]): res = self.persist_index_computations(list(contents)) if not count_object_added_key: count_object_added_key = list(res.keys())[0] count += res[count_object_added_key] if count > 0: summary["status"] = "eventful" except Exception: if not self.catch_exceptions: raise self.log.exception("Problem when computing metadata.") sentry_sdk.capture_exception() summary["status"] = "failed" if count > 0 and count_object_added_key: summary[count_object_added_key] = count return summary class OriginIndexer(BaseIndexer[str, None, TResult], Generic[TResult]): """An object type indexer, inherits from the :class:`BaseIndexer` and implements Origin indexing using the run method Note: the :class:`OriginIndexer` is not an instantiable object. To use it in another context one should inherit from this class and override the methods mentioned in the :class:`BaseIndexer` class. """ def run(self, origin_urls: List[str], **kwargs) -> Dict: """Given a list of origin urls: - retrieve origins from storage - execute the indexing computations - store the results Args: origin_urls: list of origin urls. **kwargs: passed to the `index` method """ if "policy_update" in kwargs: warnings.warn( "'policy_update' argument is deprecated and ignored.", DeprecationWarning, ) del kwargs["policy_update"] origins = [{"url": url} for url in origin_urls] return self.process_journal_objects({"origin": origins}) def process_journal_objects(self, objects: ObjectsDict) -> Dict: - """Worker function for ``JournalClient``. Expects ``objects`` to have a single - key, either ``origin`` or ``"origin_visit_status"``.""" + """Worker function for ``JournalClient``.""" origins = [ Origin(url=status["origin"]) for status in objects.get("origin_visit_status", []) if status["status"] == "full" ] + [Origin(url=origin["url"]) for origin in objects.get("origin", [])] summary: Dict[str, Any] = {"status": "uneventful"} try: results = self.index_list( origins, check_origin_known=False, # no need to check they exist, as we just received either an origin or # visit status; which cannot be created by swh-storage unless the origin # already exists ) except Exception: if not self.catch_exceptions: raise summary["status"] = "failed" return summary summary_persist = self.persist_index_computations(results) self.results = results if summary_persist: for value in summary_persist.values(): if value > 0: summary["status"] = "eventful" summary.update(summary_persist) return summary def index_list(self, origins: List[Origin], **kwargs) -> List[TResult]: results = [] for origin in origins: try: results.extend(self.index(origin.url, **kwargs)) except Exception: self.log.exception("Problem when processing origin %s", origin.url) sentry_sdk.capture_exception() raise return results class DirectoryIndexer(BaseIndexer[Sha1Git, Directory, TResult], Generic[TResult]): """An object type indexer, inherits from the :class:`BaseIndexer` and implements Directory indexing using the run method Note: the :class:`DirectoryIndexer` is not an instantiable object. To use it in another context one should inherit from this class and override the methods mentioned in the :class:`BaseIndexer` class. """ def run(self, ids: List[Sha1Git], **kwargs) -> Dict: """Given a list of sha1_gits: - retrieve directories from storage - execute the indexing computations - store the results Args: ids: sha1_git's identifier list """ if "policy_update" in kwargs: warnings.warn( "'policy_update' argument is deprecated and ignored.", DeprecationWarning, ) del kwargs["policy_update"] directory_ids = [ hashutil.hash_to_bytes(id_) if isinstance(id_, str) else id_ for id_ in ids ] return self._process_directories([(dir_id, None) for dir_id in directory_ids]) def process_journal_objects(self, objects: ObjectsDict) -> Dict: - """Worker function for ``JournalClient``. Expects ``objects`` to have a single - key, ``"directory"``.""" - assert set(objects) == {"directory"} + """Worker function for ``JournalClient``.""" return self._process_directories( - [(dir_["id"], Directory.from_dict(dir_)) for dir_ in objects["directory"]] + [ + (dir_["id"], Directory.from_dict(dir_)) + for dir_ in objects.get("directory", []) + ] ) def _process_directories( self, directories: Union[List[Tuple[Sha1Git, Directory]], List[Tuple[Sha1Git, None]]], ) -> Dict: summary: Dict[str, Any] = {"status": "uneventful"} results = [] # TODO: fetch raw_manifest when useful? for (dir_id, dir_) in directories: try: results.extend(self.index(dir_id, dir_)) except Exception: if not self.catch_exceptions: raise self.log.exception("Problem when processing directory") sentry_sdk.capture_exception() summary["status"] = "failed" summary_persist = self.persist_index_computations(results) if summary_persist: for value in summary_persist.values(): if value > 0: summary["status"] = "eventful" summary.update(summary_persist) self.results = results return summary