Skip to content
Snippets Groups Projects
Commit 370de0d1 authored by vlorentz's avatar vlorentz
Browse files

[WIP] reimplement backfiller using '*_get_partition' instead of direct SQL queries

parent c5b0ec9c
No related branches found
No related tags found
No related merge requests found
Pipeline #1244 failed
Stage: external
Hide whitespace changes
Inline Side-by-side
swh/storage/backfill.py
+ 36
− 553
View file @ 370de0d1
......@@ -43,543 +43,6 @@ from swh.storage.writer import JournalWriter
logger = logging.getLogger(__name__)
PARTITION_KEY = {
"content": "sha1",
"skipped_content": "sha1",
"directory": "id",
"extid": "target",
"metadata_authority": "type, url",
"metadata_fetcher": "name, version",
"raw_extrinsic_metadata": "target",
"revision": "revision.id",
"release": "release.id",
"snapshot": "id",
"origin": "id",
"origin_visit": "origin_visit.origin",
"origin_visit_status": "origin_visit_status.origin",
}
COLUMNS = {
"content": [
"sha1",
"sha1_git",
"sha256",
"blake2s256",
"length",
"status",
"ctime",
],
"skipped_content": [
"sha1",
"sha1_git",
"sha256",
"blake2s256",
"length",
"ctime",
"status",
"reason",
],
"directory": ["id", "dir_entries", "file_entries", "rev_entries", "raw_manifest"],
"extid": ["extid_type", "extid", "extid_version", "target_type", "target"],
"metadata_authority": ["type", "url"],
"metadata_fetcher": ["name", "version"],
"origin": ["url"],
"origin_visit": [
"visit",
"type",
("origin.url", "origin"),
"date",
],
"origin_visit_status": [
("origin_visit_status.visit", "visit"),
("origin.url", "origin"),
("origin_visit_status.date", "date"),
"type",
"snapshot",
"status",
"metadata",
],
"raw_extrinsic_metadata": [
"raw_extrinsic_metadata.type",
"raw_extrinsic_metadata.target",
"metadata_authority.type",
"metadata_authority.url",
"metadata_fetcher.name",
"metadata_fetcher.version",
"discovery_date",
"format",
"raw_extrinsic_metadata.metadata",
"origin",
"visit",
"snapshot",
"release",
"revision",
"path",
"directory",
],
"revision": [
("revision.id", "id"),
"date",
"date_offset_bytes",
"committer_date",
"committer_date_offset_bytes",
"type",
"directory",
"message",
"synthetic",
"metadata",
"extra_headers",
(
"array(select parent_id::bytea from revision_history rh "
"where rh.id = revision.id order by rh.parent_rank asc)",
"parents",
),
"raw_manifest",
("a.id", "author_id"),
("a.name", "author_name"),
("a.email", "author_email"),
("a.fullname", "author_fullname"),
("c.id", "committer_id"),
("c.name", "committer_name"),
("c.email", "committer_email"),
("c.fullname", "committer_fullname"),
],
"release": [
("release.id", "id"),
"date",
"date_offset_bytes",
"comment",
("release.name", "name"),
"synthetic",
"target",
"target_type",
("a.id", "author_id"),
("a.name", "author_name"),
("a.email", "author_email"),
("a.fullname", "author_fullname"),
"raw_manifest",
],
"snapshot": ["id", "object_id"],
}
JOINS = {
"release": ["person a on release.author=a.id"],
"revision": [
"person a on revision.author=a.id",
"person c on revision.committer=c.id",
],
"origin_visit": ["origin on origin_visit.origin=origin.id"],
"origin_visit_status": [
"origin on origin_visit_status.origin=origin.id",
],
"raw_extrinsic_metadata": [
"metadata_authority on "
"raw_extrinsic_metadata.authority_id=metadata_authority.id",
"metadata_fetcher on raw_extrinsic_metadata.fetcher_id=metadata_fetcher.id",
],
}
EXTRA_WHERE = {
# hack to force the right index usage on table extid
"extid": "target_type in ('revision', 'release', 'content', 'directory')"
}
def directory_converter(db: BaseDb, directory_d: Dict[str, Any]) -> Directory:
"""Convert directory from the flat representation to swh model
compatible objects.
"""
columns = ["target", "name", "perms"]
query_template = """
select %(columns)s
from directory_entry_%(type)s
where id in %%s
"""
types = ["file", "dir", "rev"]
entries = []
with db.cursor() as cur:
for type in types:
ids = directory_d.pop("%s_entries" % type)
if not ids:
continue
query = query_template % {
"columns": ",".join(columns),
"type": type,
}
cur.execute(query, (tuple(ids),))
for row in cur:
entry_d = dict(zip(columns, row))
entry = DirectoryEntry(
name=entry_d["name"],
type=type,
target=entry_d["target"],
perms=entry_d["perms"],
)
entries.append(entry)
(is_corrupt, dir_) = Directory.from_possibly_duplicated_entries(
id=directory_d["id"],
entries=tuple(entries),
raw_manifest=directory_d["raw_manifest"],
)
if is_corrupt:
logger.info("%s has duplicated entries", dir_.swhid())
return dir_
def raw_extrinsic_metadata_converter(
db: BaseDb, metadata: Dict[str, Any]
) -> RawExtrinsicMetadata:
"""Convert a raw extrinsic metadata from the flat representation to swh model
compatible objects.
"""
return db_to_raw_extrinsic_metadata(metadata)
def extid_converter(db: BaseDb, extid: Dict[str, Any]) -> ExtID:
"""Convert an extid from the flat representation to swh model
compatible objects.
"""
return db_to_extid(extid)
def revision_converter(db: BaseDb, revision_d: Dict[str, Any]) -> Revision:
"""Convert revision from the flat representation to swh model
compatible objects.
"""
return db_to_revision(revision_d)
def release_converter(db: BaseDb, release_d: Dict[str, Any]) -> Release:
"""Convert release from the flat representation to swh model
compatible objects.
"""
return db_to_release(release_d)
def snapshot_converter(db: BaseDb, snapshot_d: Dict[str, Any]) -> Snapshot:
"""Convert snapshot from the flat representation to swh model
compatible objects.
"""
columns = ["name", "target", "target_type"]
query = """
select %s
from snapshot_branches sbs
inner join snapshot_branch sb on sb.object_id=sbs.branch_id
where sbs.snapshot_id=%%s
""" % ", ".join(
columns
)
with db.cursor() as cur:
cur.execute(query, (snapshot_d["object_id"],))
branches = {}
for name, *row in cur:
branch_d = dict(zip(columns[1:], row))
if branch_d["target"] is not None and branch_d["target_type"] is not None:
branch: Optional[SnapshotBranch] = SnapshotBranch(
target=branch_d["target"],
target_type=TargetType(branch_d["target_type"]),
)
else:
branch = None
branches[name] = branch
return Snapshot(
id=snapshot_d["id"],
branches=branches,
)
CONVERTERS: Dict[str, Callable[[BaseDb, Dict[str, Any]], BaseModel]] = {
"directory": directory_converter,
"extid": extid_converter,
"raw_extrinsic_metadata": raw_extrinsic_metadata_converter,
"revision": revision_converter,
"release": release_converter,
"snapshot": snapshot_converter,
}
def object_to_offset(object_id, numbits):
"""Compute the index of the range containing object id, when dividing
space into 2^numbits.
Args:
object_id (str): The hex representation of object_id
numbits (int): Number of bits in which we divide input space
Returns:
The index of the range containing object id
"""
q, r = divmod(numbits, 8)
length = q + (r != 0)
shift_bits = 8 - r if r else 0
truncated_id = object_id[: length * 2]
if len(truncated_id) < length * 2:
truncated_id += "0" * (length * 2 - len(truncated_id))
truncated_id_bytes = bytes.fromhex(truncated_id)
return int.from_bytes(truncated_id_bytes, byteorder="big") >> shift_bits
def byte_ranges(
numbits: int, start_object: Optional[str] = None, end_object: Optional[str] = None
) -> Iterator[Tuple[Optional[bytes], Optional[bytes]]]:
"""Generate start/end pairs of bytes spanning numbits bits and
constrained by optional start_object and end_object.
Args:
numbits: Number of bits in which we divide input space
start_object: Hex object id contained in the first range
returned
end_object: Hex object id contained in the last range
returned
Yields:
2^numbits pairs of bytes
"""
q, r = divmod(numbits, 8)
length = q + (r != 0)
shift_bits = 8 - r if r else 0
def to_bytes(i):
return int.to_bytes(i << shift_bits, length=length, byteorder="big")
start_offset = 0
end_offset = 1 << numbits
if start_object is not None:
start_offset = object_to_offset(start_object, numbits)
if end_object is not None:
end_offset = object_to_offset(end_object, numbits) + 1
for start in range(start_offset, end_offset):
end = start + 1
if start == 0:
yield None, to_bytes(end)
elif end == 1 << numbits:
yield to_bytes(start), None
else:
yield to_bytes(start), to_bytes(end)
def raw_extrinsic_metadata_target_ranges(
start_object: Optional[str] = None, end_object: Optional[str] = None
) -> Iterator[Tuple[Optional[str], Optional[str]]]:
"""Generate ranges of values for the `target` attribute of `raw_extrinsic_metadata`
objects.
This generates one range for all values before the first SWHID (which would
correspond to raw origin URLs), then a number of hex-based ranges for each
known type of SWHID (2**12 ranges for directories, 2**8 ranges for all other
types). Finally, it generates one extra range for values above all possible
SWHIDs.
"""
if start_object is None:
start_object = ""
swhid_target_types = sorted(type.value for type in ExtendedObjectType)
first_swhid = f"swh:1:{swhid_target_types[0]}:"
# Generate a range for url targets, if the starting object is before SWHIDs
if start_object < first_swhid:
yield start_object, (
first_swhid
if end_object is None or end_object >= first_swhid
else end_object
)
if end_object is not None and end_object <= first_swhid:
return
# Prime the following loop, which uses the upper bound of the previous range
# as lower bound, to account for potential targets between two valid types
# of SWHIDs (even though they would eventually be rejected by the
# RawExtrinsicMetadata parser, they /might/ exist...)
end_swhid = first_swhid
# Generate ranges for swhid targets
for target_type in swhid_target_types:
finished = False
base_swhid = f"swh:1:{target_type}:"
last_swhid = base_swhid + ("f" * 40)
if start_object > last_swhid:
continue
# Generate 2**8 or 2**12 ranges
for _, end in byte_ranges(12 if target_type == "dir" else 8):
# Reuse previous upper bound
start_swhid = end_swhid
# Use last_swhid for this object type if on the last byte range
end_swhid = (base_swhid + end.hex()) if end is not None else last_swhid
# Ignore out of bounds ranges
if start_object >= end_swhid:
continue
# Potentially clamp start of range to the first object requested
start_swhid = max(start_swhid, start_object)
# Handle ending the loop early if the last requested object id is in
# the current range
if end_object is not None and end_swhid >= end_object:
end_swhid = end_object
finished = True
yield start_swhid, end_swhid
if finished:
return
# Generate one final range for potential raw origin URLs after the last
# valid SWHID
start_swhid = max(start_object, end_swhid)
yield start_swhid, end_object
def integer_ranges(
start: str, end: str, block_size: int = 1000
) -> Iterator[Tuple[Optional[int], Optional[int]]]:
for range_start in range(int(start), int(end), block_size):
if range_start == 0:
yield None, block_size
elif range_start + block_size > int(end):
yield range_start, int(end)
else:
yield range_start, range_start + block_size
RANGE_GENERATORS: Dict[
str,
Union[
Callable[[str, str], Iterable[Tuple[Optional[str], Optional[str]]]],
Callable[[str, str], Iterable[Tuple[Optional[bytes], Optional[bytes]]]],
Callable[[str, str], Iterable[Tuple[Optional[int], Optional[int]]]],
],
] = {
"content": lambda start, end: byte_ranges(24, start, end),
"skipped_content": lambda start, end: [(None, None)],
"directory": lambda start, end: byte_ranges(24, start, end),
"extid": lambda start, end: byte_ranges(24, start, end),
"revision": lambda start, end: byte_ranges(24, start, end),
"release": lambda start, end: byte_ranges(16, start, end),
"raw_extrinsic_metadata": raw_extrinsic_metadata_target_ranges,
"snapshot": lambda start, end: byte_ranges(16, start, end),
"origin": integer_ranges,
"origin_visit": integer_ranges,
"origin_visit_status": integer_ranges,
}
def compute_query(obj_type, start, end):
columns = COLUMNS.get(obj_type)
join_specs = JOINS.get(obj_type, [])
join_clause = "\n".join("left join %s" % clause for clause in join_specs)
additional_where = EXTRA_WHERE.get(obj_type)
where = []
where_args = []
if start:
where.append("%(keys)s >= %%s")
where_args.append(start)
if end:
where.append("%(keys)s < %%s")
where_args.append(end)
if additional_where:
where.append(additional_where)
where_clause = ""
if where:
where_clause = ("where " + " and ".join(where)) % {
"keys": "(%s)" % PARTITION_KEY[obj_type]
}
column_specs = []
column_aliases = []
for column in columns:
if isinstance(column, str):
column_specs.append(column)
column_aliases.append(column)
else:
column_specs.append("%s as %s" % column)
column_aliases.append(column[1])
query = """
select %(columns)s
from %(table)s
%(join)s
%(where)s
""" % {
"columns": ",".join(column_specs),
"table": obj_type,
"join": join_clause,
"where": where_clause,
}
return query, where_args, column_aliases
def fetch(db, obj_type, start, end):
"""Fetch all obj_type's identifiers from db.
This opens one connection, stream objects and when done, close
the connection.
Args:
db (BaseDb): Db connection object
obj_type (str): Object type
start (Union[bytes|Tuple]): Range start identifier
end (Union[bytes|Tuple]): Range end identifier
Raises:
ValueError if obj_type is not supported
Yields:
Objects in the given range
"""
query, where_args, column_aliases = compute_query(obj_type, start, end)
converter = CONVERTERS.get(obj_type)
with db.cursor() as cursor:
logger.debug("Fetching data for table %s", obj_type)
logger.debug("query: %s %s", query, where_args)
cursor.execute(query, where_args)
for row in cursor:
record = dict(zip(column_aliases, row))
if converter:
record = converter(db, record)
else:
record = OBJECT_CONVERTERS[obj_type](record)
logger.debug("record: %s", record)
yield record
def _format_range_bound(bound):
if isinstance(bound, bytes):
return bound.hex()
else:
return str(bound)
MANDATORY_KEYS = ["storage", "journal_writer"]
......@@ -649,26 +112,41 @@ class JournalBackfiller:
return start_object, end_object
def run(self, object_type, start_object, end_object, dry_run=False):
def run(self, object_type: str, start_partition_id: int, end_partition_id: int, nb_partitions: int, dry_run=False) -> None:
"""Reads storage's subscribed object types and send them to the
journal's reading topic.
"""
start_object, end_object = self.parse_arguments(
object_type, start_object, end_object
)
for range_start, range_end in RANGE_GENERATORS[object_type](
start_object, end_object
):
logger.info(
"Processing %s range %s to %s",
object_type,
_format_range_bound(range_start),
_format_range_bound(range_end),
)
objects = fetch(self.db, object_type, start=range_start, end=range_end)
for partition_id in range(self.start_partition_id, self.end_partition_id):
self._backfill_partition(object_type, partition_id, nb_partitions, dry_run)
def _backfill_partition(self, object_type: str, partition_id: int, nb_partitions: int, dry_run: bool) -> None:
page_token = None
for page_number in itertools.count():
logger.debug("Fetching page %d of partition %d of %d for %s", page_number, partition_id, nb_partitions, object_type)
if object_type == "release":
page = self.storage.release_get_partition(partition_id, self.nb_partitions, page_token=page_token)
objects = page.results
elif object_type is swhids.ObjectType.REVISION:
page = self.storage.revision_get_partition(partition_id, self.nb_partitions, page_token=page_token)
objects = page.results
elif object_type is swhids.ObjectType.DIRECTORY:
page = self.storage.directory_get_id_partition(
partition_id, self.nb_partitions, page_token=page_token
)
directory_ids = page.results
objects = list(directory_get_many(self.storage, directory_ids))
elif object_type is swhids.ObjectType.SNAPSHOT:
page = self.storage.snapshot_get_id_partition(
partition_id, self.nb_partitions, page_token=page_token
)
snapshot_ids = page.results
objects = [
snapshot_get_all_branches(self.storage, snapshot_id)
for snapshot_id in snapshot_ids
]
else:
assert False, f"Unexpected object type: {object_type}"
if not dry_run:
self.writer.write_additions(object_type, objects)
......@@ -678,6 +156,11 @@ class JournalBackfiller:
for obj in objects:
pass
page_token = page.next_page_token
if page_token is None:
break
logger.info("Done backfilling partition %d of %d for %s", partition_id, nb_partitions, object_type)
if __name__ == "__main__":
print('Please use the "swh-journal backfiller run" command')
swh/storage/cli.py
+ 18
− 5
View file @ 370de0d1
......@@ -103,11 +103,12 @@ def serve(ctx, host, port, debug):
@storage.command()
@click.argument("object_type")
@click.option("--start-object", default=None)
@click.option("--end-object", default=None)
@click.option("--start-partition-id", type=int)
@click.option("--end-partition-id", type=int)
@click.option("--nb-partitions", type=int)
@click.option("--dry-run", is_flag=True, default=False)
@click.pass_context
def backfill(ctx, object_type, start_object, end_object, dry_run):
def backfill(ctx, object_type, start_partition_id, end_partition_id, nb_partitions, dry_run):
"""Run the backfiller
The backfiller list objects from a Storage and produce journal entries from
......@@ -116,6 +117,17 @@ def backfill(ctx, object_type, start_object, end_object, dry_run):
Typically used to rebuild a journal or compensate for missing objects in a
journal (eg. due to a downtime of this later).
This runs a single thread; parallelism is achieved by running this command multiple
times, on disjoint ranges.
All objects of type ``object_type`` are ordered, and split into the given number
of partitions. When running in parallel, the number of partitions should be the
same for all workers or they may work on overlapping or non-exhaustive ranges.
Then, this process will check all partitions in the given
``[start_partition_id, end_partition_id)`` range. When running in parallel, these
ranges should be set so that processes over the whole ``[0, nb_partitions)`` range.
The configuration file requires the following entries:
- brokers: a list of kafka endpoints (the journal) in which entries will be
......@@ -144,8 +156,9 @@ def backfill(ctx, object_type, start_object, end_object, dry_run):
try:
backfiller.run(
object_type=object_type,
start_object=start_object,
end_object=end_object,
start_partition_id=start_partition_id,
end_partition_id=end_partition_id,
nb_partitions=nb_partitions,
dry_run=dry_run,
)
except KeyboardInterrupt:
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment