diff --git a/swh/loader/git/loader.py b/swh/loader/git/loader.py index 7243441..81d2747 100644 --- a/swh/loader/git/loader.py +++ b/swh/loader/git/loader.py @@ -1,582 +1,582 @@ # 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 from dataclasses import dataclass import datetime import logging import os import pickle import sys from tempfile import SpooledTemporaryFile from typing import Any, Callable, Dict, Iterable, Iterator, List, Optional, Set, Type import dulwich.client from dulwich.errors import GitProtocolError, NotGitRepository from dulwich.object_store import ObjectStoreGraphWalker from dulwich.objects import ShaFile from dulwich.pack import PackData, PackInflater from swh.core.statsd import Statsd from swh.loader.core.loader import DVCSLoader from swh.loader.exception import NotFound from swh.model import hashutil from swh.model.model import ( BaseContent, Directory, Release, Revision, Snapshot, SnapshotBranch, TargetType, ) from swh.storage.algos.snapshot import snapshot_get_latest from swh.storage.interface import StorageInterface from . import converters, dumb, utils from .utils import HexBytes logger = logging.getLogger(__name__) class RepoRepresentation: """Repository representation for a Software Heritage origin.""" def __init__( self, storage, base_snapshots: List[Snapshot] = None, incremental: bool = True, statsd: Statsd = None, ): self.storage = storage self.incremental = incremental self.statsd = statsd if base_snapshots and incremental: self.base_snapshots: List[Snapshot] = base_snapshots else: self.base_snapshots = [] self.heads: Set[HexBytes] = set() def get_parents(self, commit: bytes) -> List[bytes]: """This method should return the list of known parents""" return [] def graph_walker(self) -> ObjectStoreGraphWalker: return ObjectStoreGraphWalker(self.heads, self.get_parents) def determine_wants(self, refs: Dict[bytes, HexBytes]) -> List[HexBytes]: """Get the list of bytehex sha1s that the git loader should fetch. This compares the remote refs sent by the server with the base snapshot provided by the loader. """ if not refs: return [] # Cache existing heads local_heads: Set[HexBytes] = set() for base_snapshot in self.base_snapshots: for branch_name, branch in base_snapshot.branches.items(): if not branch or branch.target_type == TargetType.ALIAS: continue local_heads.add(hashutil.hash_to_hex(branch.target).encode()) self.heads = local_heads # Get the remote heads that we want to fetch remote_heads: Set[HexBytes] = set() for ref_name, ref_target in refs.items(): if utils.ignore_branch_name(ref_name): continue remote_heads.add(ref_target) logger.debug("local_heads_count=%s", len(local_heads)) logger.debug("remote_heads_count=%s", len(remote_heads)) wanted_refs = list(remote_heads - local_heads) logger.debug("wanted_refs_count=%s", len(wanted_refs)) if self.statsd is not None: self.statsd.histogram( "git_ignored_refs_percent", len(remote_heads - set(refs.values())) / len(refs), tags={}, ) self.statsd.histogram( "git_known_refs_percent", len(local_heads & remote_heads) / len(remote_heads), tags={}, ) return wanted_refs @dataclass class FetchPackReturn: remote_refs: Dict[bytes, HexBytes] symbolic_refs: Dict[bytes, HexBytes] pack_buffer: SpooledTemporaryFile pack_size: int class GitLoader(DVCSLoader): """A bulk loader for a git repository Emits the following statsd stats: * increments ``swh_loader_git`` * histogram ``swh_loader_git_ignored_refs_percent`` is the ratio of refs ignored over all refs of the remote repository * histogram ``swh_loader_git_known_refs_percent`` is the ratio of (non-ignored) remote heads that are already local over all non-ignored remote heads All three are tagged with ``{{"incremental": ""}}`` where ``incremental_mode`` is one of: * ``from_same_origin`` when the origin was already loaded * ``from_parent_origin`` when the origin was not already loaded, but it was detected as a forge-fork of an origin that was already loaded * ``no_previous_snapshot`` when the origin was not already loaded, and it was detected as a forge-fork of origins that were not already loaded either * ``no_parent_origin`` when the origin was no already loaded, and it was not detected as a forge-fork of any other origin * ``disabled`` when incremental loading is disabled by configuration """ visit_type = "git" def __init__( self, storage: StorageInterface, url: str, incremental: bool = True, repo_representation: Type[RepoRepresentation] = RepoRepresentation, pack_size_bytes: int = 4 * 1024 * 1024 * 1024, temp_file_cutoff: int = 100 * 1024 * 1024, **kwargs: Any, ): """Initialize the bulk updater. Args: repo_representation: swh's repository representation which is in charge of filtering between known and remote data. ... incremental: If True, the default, this starts from the last known snapshot (if any) references. Otherwise, this loads the full repository. """ super().__init__(storage=storage, origin_url=url, **kwargs) self.incremental = incremental self.repo_representation = repo_representation self.pack_size_bytes = pack_size_bytes self.temp_file_cutoff = temp_file_cutoff # state initialized in fetch_data self.remote_refs: Dict[bytes, HexBytes] = {} self.symbolic_refs: Dict[bytes, HexBytes] = {} self.ref_object_types: Dict[bytes, Optional[TargetType]] = {} def fetch_pack_from_origin( self, origin_url: str, base_repo: RepoRepresentation, do_activity: Callable[[bytes], None], ) -> FetchPackReturn: """Fetch a pack from the origin""" pack_buffer = SpooledTemporaryFile(max_size=self.temp_file_cutoff) # Hardcode the use of the tcp transport (for GitHub origins) # Even if the Dulwich API lets us process the packfile in chunks as it's # received, the HTTP transport implementation needs to entirely allocate # the packfile in memory *twice*, once in the HTTP library, and once in # a BytesIO managed by Dulwich, before passing chunks to the `do_pack` # method Overall this triples the memory usage before we can even try to # interrupt the loader before it overruns its memory limit. # In contrast, the Dulwich TCP transport just gives us the read handle # on the underlying socket, doing no processing or copying of the bytes. # We can interrupt it as soon as we've received too many bytes. transport_url = origin_url # Temporarily Disabled due to GitHub removing support for the tcp transport. See # https://forge.softwareheritage.org/T3544 # # if transport_url.startswith("https://github.com/"): # transport_url = "git" + transport_url[5:] logger.debug("Transport url to communicate with server: %s", transport_url) client, path = dulwich.client.get_transport_and_path( transport_url, thin_packs=False ) logger.debug("Client %s to fetch pack at %s", client, path) size_limit = self.pack_size_bytes def do_pack(data: bytes) -> None: cur_size = pack_buffer.tell() would_write = len(data) if cur_size + would_write > size_limit: raise IOError( f"Pack file too big for repository {origin_url}, " f"limit is {size_limit} bytes, current size is {cur_size}, " f"would write {would_write}" ) pack_buffer.write(data) pack_result = client.fetch_pack( path, base_repo.determine_wants, base_repo.graph_walker(), do_pack, progress=do_activity, ) remote_refs = pack_result.refs or {} symbolic_refs = pack_result.symrefs or {} pack_buffer.flush() pack_size = pack_buffer.tell() pack_buffer.seek(0) logger.debug("fetched_pack_size=%s", pack_size) # check if repository only supports git dumb transfer protocol, # fetched pack file will be empty in that case as dulwich do # not support it and do not fetch any refs self.dumb = transport_url.startswith("http") and getattr(client, "dumb", False) return FetchPackReturn( remote_refs=utils.filter_refs(remote_refs), symbolic_refs=utils.filter_refs(symbolic_refs), pack_buffer=pack_buffer, pack_size=pack_size, ) def get_full_snapshot(self, origin_url) -> Optional[Snapshot]: return snapshot_get_latest(self.storage, origin_url) def prepare(self) -> None: assert self.origin is not None self.prev_snapshot = Snapshot(branches={}) """Last snapshot of this origin if any; empty snapshot otherwise""" self.base_snapshots = [] """Last snapshot of this origin and all its parents, if any.""" self.statsd.constant_tags["incremental_enabled"] = self.incremental self.statsd.constant_tags["has_parent_origins"] = bool(self.parent_origins) # May be set to True later self.statsd.constant_tags["has_parent_snapshot"] = False if self.incremental: prev_snapshot = self.get_full_snapshot(self.origin.url) self.statsd.constant_tags["has_previous_snapshot"] = bool(prev_snapshot) if prev_snapshot: self.prev_snapshot = prev_snapshot self.base_snapshots.append(prev_snapshot) if self.parent_origins is not None: - # If this is the first time we load this origin and it is a forge - # fork, load incrementally from one of the origins it was forked from, - # closest parent first + + # If this origin is a forge fork, load incrementally from the + # origins it was forked from for parent_origin in self.parent_origins: parent_snapshot = self.get_full_snapshot(parent_origin.url) if parent_snapshot is not None: self.statsd.constant_tags["has_parent_snapshot"] = True self.base_snapshots.append(parent_snapshot) # Increments a metric with full name 'swh_loader_git'; which is useful to # count how many runs of the loader are with each incremental mode self.statsd.increment("git_total", tags={}) def fetch_data(self) -> bool: assert self.origin is not None base_repo = self.repo_representation( storage=self.storage, base_snapshots=self.base_snapshots, incremental=self.incremental, statsd=self.statsd, ) def do_progress(msg: bytes) -> None: sys.stderr.buffer.write(msg) sys.stderr.flush() try: fetch_info = self.fetch_pack_from_origin( self.origin.url, base_repo, do_progress ) except (dulwich.client.HTTPUnauthorized, NotGitRepository) as e: raise NotFound(e) except GitProtocolError as e: # unfortunately, that kind of error is not specific to a not found # scenario... It depends on the value of message within the exception. for msg in [ "Repository unavailable", # e.g DMCA takedown "Repository not found", "unexpected http resp 401", ]: if msg in e.args[0]: raise NotFound(e) # otherwise transmit the error raise except (AttributeError, NotImplementedError, ValueError): # with old dulwich versions, those exceptions types can be raised # by the fetch_pack operation when encountering a repository with # dumb transfer protocol so we check if the repository supports it # here to continue the loading if it is the case self.dumb = dumb.check_protocol(self.origin.url) if not self.dumb: raise logger.debug( "Protocol used for communication: %s", "dumb" if self.dumb else "smart" ) if self.dumb: self.dumb_fetcher = dumb.GitObjectsFetcher(self.origin.url, base_repo) self.dumb_fetcher.fetch_object_ids() self.remote_refs = utils.filter_refs(self.dumb_fetcher.refs) # type: ignore self.symbolic_refs = self.dumb_fetcher.head else: self.pack_buffer = fetch_info.pack_buffer self.pack_size = fetch_info.pack_size self.remote_refs = fetch_info.remote_refs self.symbolic_refs = fetch_info.symbolic_refs self.ref_object_types = {sha1: None for sha1 in self.remote_refs.values()} logger.info( "Listed %d refs for repo %s", len(self.remote_refs), self.origin.url, extra={ "swh_type": "git_repo_list_refs", "swh_repo": self.origin.url, "swh_num_refs": len(self.remote_refs), }, ) # No more data to fetch return False def save_data(self) -> None: """Store a pack for archival""" assert isinstance(self.visit_date, datetime.datetime) write_size = 8192 pack_dir = self.get_save_data_path() pack_name = "%s.pack" % self.visit_date.isoformat() refs_name = "%s.refs" % self.visit_date.isoformat() with open(os.path.join(pack_dir, pack_name), "xb") as f: self.pack_buffer.seek(0) while True: r = self.pack_buffer.read(write_size) if not r: break f.write(r) self.pack_buffer.seek(0) with open(os.path.join(pack_dir, refs_name), "xb") as f: pickle.dump(self.remote_refs, f) def iter_objects(self, object_type: bytes) -> Iterator[ShaFile]: """Read all the objects of type `object_type` from the packfile""" if self.dumb: yield from self.dumb_fetcher.iter_objects(object_type) else: self.pack_buffer.seek(0) count = 0 for obj in PackInflater.for_pack_data( PackData.from_file(self.pack_buffer, self.pack_size) ): if obj.type_name != object_type: continue yield obj count += 1 logger.debug("packfile_read_count_%s=%s", object_type.decode(), count) def get_contents(self) -> Iterable[BaseContent]: """Format the blobs from the git repository as swh contents""" for raw_obj in self.iter_objects(b"blob"): if raw_obj.id in self.ref_object_types: self.ref_object_types[raw_obj.id] = TargetType.CONTENT yield converters.dulwich_blob_to_content( raw_obj, max_content_size=self.max_content_size ) def get_directories(self) -> Iterable[Directory]: """Format the trees as swh directories""" for raw_obj in self.iter_objects(b"tree"): if raw_obj.id in self.ref_object_types: self.ref_object_types[raw_obj.id] = TargetType.DIRECTORY yield converters.dulwich_tree_to_directory(raw_obj) def get_revisions(self) -> Iterable[Revision]: """Format commits as swh revisions""" for raw_obj in self.iter_objects(b"commit"): if raw_obj.id in self.ref_object_types: self.ref_object_types[raw_obj.id] = TargetType.REVISION yield converters.dulwich_commit_to_revision(raw_obj) def get_releases(self) -> Iterable[Release]: """Retrieve all the release objects from the git repository""" for raw_obj in self.iter_objects(b"tag"): if raw_obj.id in self.ref_object_types: self.ref_object_types[raw_obj.id] = TargetType.RELEASE yield converters.dulwich_tag_to_release(raw_obj) def get_snapshot(self) -> Snapshot: """Get the snapshot for the current visit. The main complexity of this function is mapping target objects to their types, as the `refs` dictionaries returned by the git server only give us the identifiers for the target objects, and not their types. The loader itself only knows the types of the objects that it has fetched from the server (as it has parsed them while loading them to the archive). As we only fetched an increment between the previous snapshot and the current state of the server, we are missing the type information for the objects that would already have been referenced by the previous snapshot, and that the git server didn't send us. We infer the type of these objects from the previous snapshot. """ branches: Dict[bytes, Optional[SnapshotBranch]] = {} unfetched_refs: Dict[bytes, bytes] = {} # Retrieve types from the objects loaded by the current loader for ref_name, ref_object in self.remote_refs.items(): if ref_name in self.symbolic_refs: continue target = hashutil.hash_to_bytes(ref_object.decode()) target_type = self.ref_object_types.get(ref_object) if target_type: branches[ref_name] = SnapshotBranch( target=target, target_type=target_type ) else: # The object pointed at by this ref was not fetched, supposedly # because it existed in the base snapshot. We record it here, # and we can get it from the base snapshot later. unfetched_refs[ref_name] = target dangling_branches = {} # Handle symbolic references as alias branches for ref_name, target in self.symbolic_refs.items(): branches[ref_name] = SnapshotBranch( target_type=TargetType.ALIAS, target=target, ) if target not in branches and target not in unfetched_refs: # This handles the case where the pointer is "dangling". # There's a chance that a further symbolic reference # override this default value, which is totally fine. dangling_branches[target] = ref_name branches[target] = None if unfetched_refs: # Handle inference of object types from the contents of the # previous snapshot unknown_objects = {} base_snapshot_reverse_branches = { branch.target: branch for base_snapshot in reversed(self.base_snapshots) for branch in base_snapshot.branches.values() if branch and branch.target_type != TargetType.ALIAS } assert all( base_snapshot_reverse_branches[branch.target] == branch for branch in self.prev_snapshot.branches.values() if branch and branch.target_type != TargetType.ALIAS ), "base_snapshot_reverse_branches is not a superset of prev_snapshot" for ref_name, target in unfetched_refs.items(): branch = base_snapshot_reverse_branches.get(target) branches[ref_name] = branch if not branch: unknown_objects[ref_name] = target if unknown_objects: # This object was referenced by the server; We did not fetch # it, and we do not know it from the previous snapshot. This is # likely a bug in the loader. raise RuntimeError( "Unknown objects referenced by remote refs: %s" % ( ", ".join( f"{name.decode()}: {hashutil.hash_to_hex(obj)}" for name, obj in unknown_objects.items() ) ) ) utils.warn_dangling_branches( branches, dangling_branches, logger, self.origin.url ) self.snapshot = Snapshot(branches=branches) return self.snapshot def load_status(self) -> Dict[str, Any]: """The load was eventful if the current snapshot is different to the one we retrieved at the beginning of the run""" eventful = False if self.prev_snapshot and self.snapshot: eventful = self.snapshot.id != self.prev_snapshot.id elif self.snapshot: eventful = bool(self.snapshot.branches) return {"status": ("eventful" if eventful else "uneventful")} if __name__ == "__main__": import click logging.basicConfig( level=logging.DEBUG, format="%(asctime)s %(process)d %(message)s" ) from deprecated import deprecated @deprecated(version="1.1", reason="Use `swh loader run git --help` instead") @click.command() @click.option("--origin-url", help="Origin url", required=True) @click.option("--base-url", default=None, help="Optional Base url") @click.option( "--ignore-history/--no-ignore-history", help="Ignore the repository history", default=False, ) def main(origin_url: str, incremental: bool) -> Dict[str, Any]: from swh.storage import get_storage storage = get_storage(cls="memory") loader = GitLoader( storage, origin_url, incremental=incremental, ) return loader.load() main()