diff --git a/contents.py b/contents.py deleted file mode 100644 index 2cd125c..0000000 --- a/contents.py +++ /dev/null @@ -1,57 +0,0 @@ -import os - -from swh.model.hashutil import hash_to_hex -from swh.provenance.provenance import get_provenance - - -if __name__ == "__main__": - conninfo = { - "host": "localhost", - "database": "new_1000", - "user": "postgres", - "password": "postgres", - } - provenance = get_provenance(conninfo) - - print("content(id, date): ################################################") - provenance.cursor.execute("""SELECT id, date FROM content ORDER BY id""") - for row in provenance.cursor.fetchall(): - print(f"{hash_to_hex(row[0])}, {row[1]}") - print("###################################################################") - - print("content_early_in_rev(blob, rev, path): ############################") - provenance.cursor.execute( - """SELECT blob, rev, path FROM content_early_in_rev ORDER BY blob, rev, path""" - ) - for row in provenance.cursor.fetchall(): - print(f"{row[0]}, {row[1]}, {row[2]}") - print(f"{hash_to_hex(row[0])}, {hash_to_hex(row[1])}, {os.fsdecode(row[2])}") - print("###################################################################") - - print("content_in_dir(blob, dir, path): ##################################") - provenance.cursor.execute( - """SELECT blob, dir, path FROM content_in_dir ORDER BY blob, dir, path""" - ) - for row in provenance.cursor.fetchall(): - print(f"{hash_to_hex(row[0])}, {hash_to_hex(row[1])}, {os.fsdecode(row[2])}") - print("###################################################################") - - print("directory(id, date): ##############################################") - provenance.cursor.execute("""SELECT id, date FROM directory ORDER BY id""") - for row in provenance.cursor.fetchall(): - print(f"{hash_to_hex(row[0])}, {row[1]}") - print("###################################################################") - - print("directory_in_rev(dir, rev, path): #################################") - provenance.cursor.execute( - """SELECT dir, rev, path FROM directory_in_rev ORDER BY dir, rev, path""" - ) - for row in provenance.cursor.fetchall(): - print(f"{hash_to_hex(row[0])}, {hash_to_hex(row[1])}, {os.fsdecode(row[2])}") - print("###################################################################") - - print("revision(id, date): ###############################################") - provenance.cursor.execute("""SELECT id, date FROM revision ORDER BY id""") - for row in provenance.cursor.fetchall(): - print(f"{hash_to_hex(row[0])}, {row[1]}") - print("###################################################################") diff --git a/swh/provenance/__init__.py b/swh/provenance/__init__.py index c227040..7fe778d 100644 --- a/swh/provenance/__init__.py +++ b/swh/provenance/__init__.py @@ -1,24 +1,28 @@ from .archive import ArchiveInterface from .provenance import ProvenanceInterface from .storage.archive import ArchiveStorage from .postgresql.archive import ArchivePostgreSQL from .postgresql.db_utils import connect from .postgresql.provenance import ProvenancePostgreSQL +from .postgresql_nopath.provenance import ProvenancePostgreSQLNoPath def get_archive(cls: str, **kwargs) -> ArchiveInterface: if cls == "api": return ArchiveStorage(**kwargs["storage"]) elif cls == "ps": conn = connect(kwargs["db"]) return ArchivePostgreSQL(conn) else: raise NotImplementedError def get_provenance(cls: str, **kwargs) -> ProvenanceInterface: if cls == "ps": conn = connect(kwargs["db"]) return ProvenancePostgreSQL(conn) + elif cls == "ps_np": + conn = connect(kwargs["db"]) + return ProvenancePostgreSQLNoPath(conn) else: raise NotImplementedError diff --git a/swh/provenance/cli.py b/swh/provenance/cli.py index a79bd46..4a4824f 100644 --- a/swh/provenance/cli.py +++ b/swh/provenance/cli.py @@ -1,201 +1,209 @@ # Copyright (C) 2020 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 # WARNING: do not import unnecessary things here to keep cli startup time under # control import os from typing import Any, Dict, Optional import click import yaml from swh.core import config from swh.core.cli import CONTEXT_SETTINGS from swh.core.cli import swh as swh_cli_group from swh.model.hashutil import hash_to_bytes, hash_to_hex # All generic config code should reside in swh.core.config CONFIG_ENVVAR = "SWH_CONFIG_FILE" DEFAULT_CONFIG_PATH = os.path.join(click.get_app_dir("swh"), "global.yml") DEFAULT_PATH = os.environ.get(CONFIG_ENVVAR, DEFAULT_CONFIG_PATH) DEFAULT_CONFIG: Dict[str, Any] = { "archive": { "cls": "api", "storage": { "cls": "remote", "url": "http://uffizi.internal.softwareheritage.org:5002", } # "cls": "ps", # "db": { # "host": "db.internal.softwareheritage.org", # "dbname": "softwareheritage", # "user": "guest" # } }, "provenance": {"cls": "ps", "db": {"host": "localhost", "dbname": "provenance"}}, } CONFIG_FILE_HELP = f"""Configuration file: \b The CLI option or the environment variable will fail if invalid. CLI option is checked first. Then, environment variable {CONFIG_ENVVAR} is checked. Then, if cannot load the default path, a set of default values are used. Default config path is {DEFAULT_CONFIG_PATH}. Default config values are: \b {yaml.dump(DEFAULT_CONFIG)}""" PROVENANCE_HELP = f"""Software Heritage Scanner tools. {CONFIG_FILE_HELP}""" @swh_cli_group.group( name="provenance", context_settings=CONTEXT_SETTINGS, help=PROVENANCE_HELP ) @click.option( "-C", "--config-file", default=None, type=click.Path(exists=False, dir_okay=False, path_type=str), help="""YAML configuration file.""", ) @click.option( "-P", "--profile", default=None, type=click.Path(exists=False, dir_okay=False, path_type=str), help="""Enable profiling to specified file.""", ) @click.pass_context def cli(ctx, config_file: Optional[str], profile: str): if config_file is None and config.config_exists(DEFAULT_PATH): config_file = DEFAULT_PATH if config_file is None: conf = DEFAULT_CONFIG else: # read_raw_config do not fail on ENOENT if not config.config_exists(config_file): raise FileNotFoundError(config_file) conf = config.read_raw_config(config.config_basepath(config_file)) conf = config.merge_configs(DEFAULT_CONFIG, conf) ctx.ensure_object(dict) ctx.obj["config"] = conf if profile: import cProfile import atexit print("Profiling...") pr = cProfile.Profile() pr.enable() def exit(): pr.disable() pr.dump_stats(profile) atexit.register(exit) @cli.command(name="create") @click.option("--name", default=None) @click.pass_context def create(ctx, name): """Create new provenance database.""" from .postgresql.db_utils import connect - from .postgresql.provenance import create_database # Connect to server without selecting a database conninfo = ctx.obj["config"]["provenance"]["db"] conn = connect(conninfo) - create_database(conn, conninfo, name) + if ctx.obj["config"]["provenance"]["cls"] == "ps": + from .postgresql.provenance import create_database + + create_database(conn, conninfo, name) + elif ctx.obj["config"]["provenance"]["cls"] == "ps_np": + from .postgresql_nopath.provenance import create_database + + create_database(conn, conninfo, name) + else: + raise NotImplementedError @cli.command(name="iter-revisions") @click.argument("filename") @click.option("-l", "--limit", type=int) @click.pass_context def iter_revisions(ctx, filename, limit): """Process a provided list of revisions.""" from . import get_archive, get_provenance from .revision import FileRevisionIterator from .provenance import revision_add archive = get_archive(**ctx.obj["config"]["archive"]) provenance = get_provenance(**ctx.obj["config"]["provenance"]) revisions = FileRevisionIterator(filename, archive, limit=limit) while True: revision = revisions.next() if revision is None: break revision_add(provenance, archive, revision) @cli.command(name="iter-origins") @click.argument("filename") @click.option("-l", "--limit", type=int) @click.pass_context def iter_origins(ctx, filename, limit): """Process a provided list of origins.""" from . import get_archive, get_provenance from .origin import FileOriginIterator from .provenance import origin_add archive = get_archive(**ctx.obj["config"]["archive"]) provenance = get_provenance(**ctx.obj["config"]["provenance"]) for origin in FileOriginIterator(filename, archive, limit=limit): origin_add(provenance, origin) @cli.command(name="find-first") @click.argument("swhid") @click.pass_context def find_first(ctx, swhid): """Find first occurrence of the requested blob.""" from . import get_provenance provenance = get_provenance(**ctx.obj["config"]["provenance"]) # TODO: return a dictionary with proper keys for each field row = provenance.content_find_first(hash_to_bytes(swhid)) if row is not None: print( "{blob}, {rev}, {date}, {path}".format( blob=hash_to_hex(row[0]), rev=hash_to_hex(row[1]), date=row[2], path=os.fsdecode(row[3]), ) ) else: print(f"Cannot find a content with the id {swhid}") @cli.command(name="find-all") @click.argument("swhid") @click.pass_context def find_all(ctx, swhid): """Find all occurrences of the requested blob.""" from swh.provenance import get_provenance provenance = get_provenance(**ctx.obj["config"]["provenance"]) # TODO: return a dictionary with proper keys for each field for row in provenance.content_find_all(hash_to_bytes(swhid)): print( "{blob}, {rev}, {date}, {path}".format( blob=hash_to_hex(row[0]), rev=hash_to_hex(row[1]), date=row[2], path=os.fsdecode(row[3]), ) ) diff --git a/swh/provenance/postgresql/archive.py b/swh/provenance/postgresql/archive.py index 61e9692..f9f227e 100644 --- a/swh/provenance/postgresql/archive.py +++ b/swh/provenance/postgresql/archive.py @@ -1,84 +1,84 @@ import psycopg2 # import threading from ..archive import ArchiveInterface # from functools import lru_cache from methodtools import lru_cache from typing import Any, Dict, List class ArchivePostgreSQL(ArchiveInterface): def __init__(self, conn: psycopg2.extensions.connection): self.conn = conn # self.mutex = threading.Lock() def directory_ls(self, id: bytes) -> List[Dict[str, Any]]: # TODO: only call directory_ls_internal if the id is not being queried by # someone else. Otherwise wait until results get properly cached. # self.mutex.acquire() entries = self.directory_ls_internal(id) # self.mutex.release() return entries @lru_cache(maxsize=1000000) def directory_ls_internal(self, id: bytes) -> List[Dict[str, Any]]: cursor = self.conn.cursor() cursor.execute( """WITH dir AS (SELECT id AS dir_id, dir_entries, file_entries, rev_entries FROM directory WHERE id=%s), - ls_d AS (SELECT dir_id, unnest(dir_entries) AS entry_id from dir), - ls_f AS (SELECT dir_id, unnest(file_entries) AS entry_id from dir), - ls_r AS (SELECT dir_id, unnest(rev_entries) AS entry_id from dir) + ls_d AS (SELECT dir_id, UNNEST(dir_entries) AS entry_id FROM dir), + ls_f AS (SELECT dir_id, UNNEST(file_entries) AS entry_id FROM dir), + ls_r AS (SELECT dir_id, UNNEST(rev_entries) AS entry_id FROM dir) (SELECT 'dir'::directory_entry_type AS type, e.target, e.name, NULL::sha1_git FROM ls_d LEFT JOIN directory_entry_dir e ON ls_d.entry_id=e.id) UNION (WITH known_contents AS (SELECT 'file'::directory_entry_type AS type, e.target, e.name, c.sha1_git FROM ls_f LEFT JOIN directory_entry_file e ON ls_f.entry_id=e.id INNER JOIN content c ON e.target=c.sha1_git) SELECT * FROM known_contents UNION (SELECT 'file'::directory_entry_type AS type, e.target, e.name, c.sha1_git FROM ls_f LEFT JOIN directory_entry_file e ON ls_f.entry_id=e.id LEFT JOIN skipped_content c ON e.target=c.sha1_git WHERE NOT EXISTS ( SELECT 1 FROM known_contents WHERE known_contents.sha1_git=e.target ) ) ) ORDER BY name """, (id,), ) return [ {"type": row[0], "target": row[1], "name": row[2]} for row in cursor.fetchall() ] def iter_origins(self): raise NotImplementedError def iter_origin_visits(self, origin: str): raise NotImplementedError def iter_origin_visit_statuses(self, origin: str, visit: int): raise NotImplementedError def release_get(self, ids: List[bytes]): raise NotImplementedError def revision_get(self, ids: List[bytes]): raise NotImplementedError def snapshot_get_all_branches(self, snapshot: bytes): raise NotImplementedError diff --git a/swh/provenance/postgresql/provenance.py b/swh/provenance/postgresql/provenance.py index 927ca4a..267ba6c 100644 --- a/swh/provenance/postgresql/provenance.py +++ b/swh/provenance/postgresql/provenance.py @@ -1,503 +1,505 @@ import itertools import logging import operator import os import psycopg2 import psycopg2.extras from ..model import DirectoryEntry, FileEntry from ..origin import OriginEntry from .db_utils import connect, execute_sql from ..provenance import ProvenanceInterface from ..revision import RevisionEntry from datetime import datetime from typing import Any, Dict, Generator, List, Optional, Tuple def normalize(path: bytes) -> bytes: return path[2:] if path.startswith(bytes("." + os.path.sep, "utf-8")) else path def create_database(conn: psycopg2.extensions.connection, conninfo: dict, name: str): conn.set_isolation_level(psycopg2.extensions.ISOLATION_LEVEL_AUTOCOMMIT) # Normalize dbname to avoid issues when reconnecting below name = name.casefold() # Create new database dropping previous one if exists cursor = conn.cursor() cursor.execute(f"""DROP DATABASE IF EXISTS {name}""") cursor.execute(f"""CREATE DATABASE {name}""") conn.close() # Reconnect to server selecting newly created database to add tables conninfo["dbname"] = name conn = connect(conninfo) sqldir = os.path.dirname(os.path.realpath(__file__)) execute_sql(conn, os.path.join(sqldir, "provenance.sql")) ######################################################################################## ######################################################################################## ######################################################################################## class ProvenancePostgreSQL(ProvenanceInterface): def __init__(self, conn: psycopg2.extensions.connection): # TODO: consider adding a mutex for thread safety conn.set_isolation_level(psycopg2.extensions.ISOLATION_LEVEL_AUTOCOMMIT) self.conn = conn self.cursor = self.conn.cursor() self.insert_cache: Dict[str, Any] = {} self.remove_cache: Dict[str, Any] = {} self.select_cache: Dict[str, Any] = {} self.clear_caches() def clear_caches(self): self.insert_cache = { "content": dict(), "content_early_in_rev": list(), "content_in_dir": list(), "directory": dict(), "directory_in_rev": list(), "revision": dict(), "revision_before_rev": list(), "revision_in_org": list(), } self.remove_cache = {"directory": dict()} self.select_cache = {"content": dict(), "directory": dict(), "revision": dict()} def commit(self): result = False try: self.insert_all() self.clear_caches() result = True except Exception as error: # Unexpected error occurred, rollback all changes and log message logging.error(f"Unexpected error: {error}") return result def content_add_to_directory( self, directory: DirectoryEntry, blob: FileEntry, prefix: bytes ): self.insert_cache["content_in_dir"].append( (blob.id, directory.id, normalize(os.path.join(prefix, blob.name))) ) def content_add_to_revision( self, revision: RevisionEntry, blob: FileEntry, prefix: bytes ): self.insert_cache["content_early_in_rev"].append( (blob.id, revision.id, normalize(os.path.join(prefix, blob.name))) ) - def content_find_first(self, blobid: bytes) -> Tuple[bytes, bytes, datetime, bytes]: + def content_find_first( + self, blobid: bytes + ) -> Optional[Tuple[bytes, bytes, datetime, bytes]]: self.cursor.execute( """SELECT content_location.sha1 AS blob, revision.sha1 AS rev, revision.date AS date, content_location.path AS path FROM (SELECT content_hex.sha1, content_hex.rev, location.path FROM (SELECT content.sha1, content_early_in_rev.rev, content_early_in_rev.loc FROM content_early_in_rev JOIN content ON content.id=content_early_in_rev.blob WHERE content.sha1=%s ) AS content_hex JOIN location ON location.id=content_hex.loc ) AS content_location JOIN revision ON revision.id=content_location.rev ORDER BY date, rev, path ASC LIMIT 1""", (blobid,), ) return self.cursor.fetchone() def content_find_all( self, blobid: bytes ) -> Generator[Tuple[bytes, bytes, datetime, bytes], None, None]: self.cursor.execute( """(SELECT content_location.sha1 AS blob, revision.sha1 AS rev, revision.date AS date, content_location.path AS path FROM (SELECT content_hex.sha1, content_hex.rev, location.path FROM (SELECT content.sha1, content_early_in_rev.rev, content_early_in_rev.loc FROM content_early_in_rev JOIN content ON content.id=content_early_in_rev.blob WHERE content.sha1=%s ) AS content_hex JOIN location ON location.id=content_hex.loc ) AS content_location JOIN revision ON revision.id=content_location.rev ) UNION (SELECT content_prefix.sha1 AS blob, revision.sha1 AS rev, revision.date AS date, content_prefix.path AS path FROM (SELECT content_in_rev.sha1, content_in_rev.rev, CASE location.path WHEN '' THEN content_in_rev.suffix WHEN '.' THEN content_in_rev.suffix ELSE (location.path || '/' || content_in_rev.suffix)::unix_path END AS path FROM (SELECT content_suffix.sha1, directory_in_rev.rev, directory_in_rev.loc, content_suffix.path AS suffix FROM (SELECT content_hex.sha1, content_hex.dir, location.path FROM (SELECT content.sha1, content_in_dir.dir, content_in_dir.loc FROM content_in_dir JOIN content ON content_in_dir.blob=content.id WHERE content.sha1=%s ) AS content_hex JOIN location ON location.id=content_hex.loc ) AS content_suffix JOIN directory_in_rev ON directory_in_rev.dir=content_suffix.dir ) AS content_in_rev JOIN location ON location.id=content_in_rev.loc ) AS content_prefix JOIN revision ON revision.id=content_prefix.rev ) ORDER BY date, rev, path""", (blobid, blobid), ) # TODO: use POSTGRESQL EXPLAIN looking for query optimizations. yield from self.cursor.fetchall() def content_get_early_date(self, blob: FileEntry) -> Optional[datetime]: # First check if the date is being modified by current transection. date = self.insert_cache["content"].get(blob.id, None) if date is None: # If not, check whether it's been query before date = self.select_cache["content"].get(blob.id, None) if date is None: # Otherwise, query the database and cache the value self.cursor.execute( """SELECT date FROM content WHERE sha1=%s""", (blob.id,) ) row = self.cursor.fetchone() date = row[0] if row is not None else None self.select_cache["content"][blob.id] = date return date def content_get_early_dates(self, blobs: List[FileEntry]) -> Dict[bytes, datetime]: dates = {} pending = [] for blob in blobs: # First check if the date is being modified by current transection. date = self.insert_cache["content"].get(blob.id, None) if date is not None: dates[blob.id] = date else: # If not, check whether it's been query before date = self.select_cache["content"].get(blob.id, None) if date is not None: dates[blob.id] = date else: pending.append(blob.id) if pending: # Otherwise, query the database and cache the values values = ", ".join(itertools.repeat("%s", len(pending))) self.cursor.execute( f"""SELECT sha1, date FROM content WHERE sha1 IN ({values})""", tuple(pending), ) for row in self.cursor.fetchall(): dates[row[0]] = row[1] self.select_cache["content"][row[0]] = row[1] return dates def content_set_early_date(self, blob: FileEntry, date: datetime): self.insert_cache["content"][blob.id] = date def directory_add_to_revision( self, revision: RevisionEntry, directory: DirectoryEntry, path: bytes ): self.insert_cache["directory_in_rev"].append( (directory.id, revision.id, normalize(path)) ) def directory_get_date_in_isochrone_frontier( self, directory: DirectoryEntry ) -> Optional[datetime]: # First check if the date is being modified by current transection. date = self.insert_cache["directory"].get(directory.id, None) if date is None and directory.id not in self.remove_cache["directory"]: # If not, check whether it's been query before date = self.select_cache["directory"].get(directory.id, None) if date is None: # Otherwise, query the database and cache the value self.cursor.execute( """SELECT date FROM directory WHERE sha1=%s""", (directory.id,) ) row = self.cursor.fetchone() date = row[0] if row is not None else None self.select_cache["directory"][directory.id] = date return date def directory_get_dates_in_isochrone_frontier( self, dirs: List[DirectoryEntry] ) -> Dict[bytes, datetime]: dates = {} pending = [] for directory in dirs: # First check if the date is being modified by current transection. date = self.insert_cache["directory"].get(directory.id, None) if date is not None: dates[directory.id] = date elif directory.id not in self.remove_cache["directory"]: # If not, check whether it's been query before date = self.select_cache["directory"].get(directory.id, None) if date is not None: dates[directory.id] = date else: pending.append(directory.id) if pending: # Otherwise, query the database and cache the values values = ", ".join(itertools.repeat("%s", len(pending))) self.cursor.execute( f"""SELECT sha1, date FROM directory WHERE sha1 IN ({values})""", tuple(pending), ) for row in self.cursor.fetchall(): dates[row[0]] = row[1] self.select_cache["directory"][row[0]] = row[1] return dates def directory_invalidate_in_isochrone_frontier(self, directory: DirectoryEntry): self.remove_cache["directory"][directory.id] = None self.insert_cache["directory"].pop(directory.id, None) def directory_set_date_in_isochrone_frontier( self, directory: DirectoryEntry, date: datetime ): self.insert_cache["directory"][directory.id] = date self.remove_cache["directory"].pop(directory.id, None) def insert_all(self): # Performe insertions with cached information if self.insert_cache["content"]: psycopg2.extras.execute_values( self.cursor, """LOCK TABLE ONLY content; INSERT INTO content(sha1, date) VALUES %s ON CONFLICT (sha1) DO UPDATE SET date=LEAST(EXCLUDED.date,content.date)""", self.insert_cache["content"].items(), ) self.insert_cache["content"].clear() if self.insert_cache["directory"]: psycopg2.extras.execute_values( self.cursor, """LOCK TABLE ONLY directory; INSERT INTO directory(sha1, date) VALUES %s ON CONFLICT (sha1) DO UPDATE SET date=LEAST(EXCLUDED.date,directory.date)""", self.insert_cache["directory"].items(), ) self.insert_cache["directory"].clear() if self.insert_cache["revision"]: psycopg2.extras.execute_values( self.cursor, """LOCK TABLE ONLY revision; INSERT INTO revision(sha1, date) VALUES %s ON CONFLICT (sha1) DO UPDATE SET date=LEAST(EXCLUDED.date,revision.date)""", self.insert_cache["revision"].items(), ) self.insert_cache["revision"].clear() # Relations should come after ids for elements were resolved if self.insert_cache["content_early_in_rev"]: self.insert_location("content", "revision", "content_early_in_rev") if self.insert_cache["content_in_dir"]: self.insert_location("content", "directory", "content_in_dir") if self.insert_cache["directory_in_rev"]: self.insert_location("directory", "revision", "directory_in_rev") # if self.insert_cache["revision_before_rev"]: # psycopg2.extras.execute_values( # self.cursor, # """INSERT INTO revision_before_rev VALUES %s # ON CONFLICT DO NOTHING""", # self.insert_cache["revision_before_rev"], # ) # self.insert_cache["revision_before_rev"].clear() # if self.insert_cache["revision_in_org"]: # psycopg2.extras.execute_values( # self.cursor, # """INSERT INTO revision_in_org VALUES %s # ON CONFLICT DO NOTHING""", # self.insert_cache["revision_in_org"], # ) # self.insert_cache["revision_in_org"].clear() def insert_location(self, src0_table, src1_table, dst_table): # Resolve src0 ids src0_values = dict().fromkeys( map(operator.itemgetter(0), self.insert_cache[dst_table]) ) values = ", ".join(itertools.repeat("%s", len(src0_values))) self.cursor.execute( f"""SELECT sha1, id FROM {src0_table} WHERE sha1 IN ({values})""", tuple(src0_values), ) src0_values = dict(self.cursor.fetchall()) # Resolve src1 ids src1_values = dict().fromkeys( map(operator.itemgetter(1), self.insert_cache[dst_table]) ) values = ", ".join(itertools.repeat("%s", len(src1_values))) self.cursor.execute( f"""SELECT sha1, id FROM {src1_table} WHERE sha1 IN ({values})""", tuple(src1_values), ) src1_values = dict(self.cursor.fetchall()) # Resolve location ids location = dict().fromkeys( map(operator.itemgetter(2), self.insert_cache[dst_table]) ) location = dict( psycopg2.extras.execute_values( self.cursor, """LOCK TABLE ONLY location; INSERT INTO location(path) VALUES %s ON CONFLICT (path) DO UPDATE SET path=EXCLUDED.path RETURNING path, id""", map(lambda path: (path,), location.keys()), fetch=True, ) ) # Insert values in dst_table rows = map( lambda row: (src0_values[row[0]], src1_values[row[1]], location[row[2]]), self.insert_cache[dst_table], ) psycopg2.extras.execute_values( self.cursor, f"""INSERT INTO {dst_table} VALUES %s ON CONFLICT DO NOTHING""", rows, ) self.insert_cache[dst_table].clear() def origin_get_id(self, origin: OriginEntry) -> int: if origin.id is None: # Insert origin in the DB and return the assigned id self.cursor.execute( """INSERT INTO origin (url) VALUES (%s) ON CONFLICT DO NOTHING RETURNING id""", (origin.url,), ) return self.cursor.fetchone()[0] else: return origin.id def revision_add(self, revision: RevisionEntry): # Add current revision to the compact DB self.insert_cache["revision"][revision.id] = revision.date def revision_add_before_revision( self, relative: RevisionEntry, revision: RevisionEntry ): self.insert_cache["revision_before_rev"].append((revision.id, relative.id)) def revision_add_to_origin(self, origin: OriginEntry, revision: RevisionEntry): self.insert_cache["revision_in_org"].append((revision.id, origin.id)) def revision_get_early_date(self, revision: RevisionEntry) -> Optional[datetime]: date = self.insert_cache["revision"].get(revision.id, None) if date is None: # If not, check whether it's been query before date = self.select_cache["revision"].get(revision.id, None) if date is None: # Otherwise, query the database and cache the value self.cursor.execute( """SELECT date FROM revision WHERE sha1=%s""", (revision.id,) ) row = self.cursor.fetchone() date = row[0] if row is not None else None self.select_cache["revision"][revision.id] = date return date def revision_get_prefered_origin(self, revision: RevisionEntry) -> int: # TODO: adapt this method to consider cached values self.cursor.execute( """SELECT COALESCE(org,0) FROM revision WHERE sha1=%s""", (revision.id,) ) row = self.cursor.fetchone() # None means revision is not in database; # 0 means revision has no prefered origin return row[0] if row is not None and row[0] != 0 else None def revision_in_history(self, revision: RevisionEntry) -> bool: # TODO: adapt this method to consider cached values self.cursor.execute( """SELECT 1 FROM revision_before_rev JOIN revision ON revision.id=revision_before_rev.prev WHERE revision.sha1=%s""", (revision.id,), ) return self.cursor.fetchone() is not None def revision_set_prefered_origin( self, origin: OriginEntry, revision: RevisionEntry ): # TODO: adapt this method to consider cached values self.cursor.execute( """UPDATE revision SET org=%s WHERE sha1=%s""", (origin.id, revision.id) ) def revision_visited(self, revision: RevisionEntry) -> bool: # TODO: adapt this method to consider cached values self.cursor.execute( """SELECT 1 FROM revision_in_org JOIN revision ON revision.id=revision_in_org.rev WHERE revision.sha1=%s""", (revision.id,), ) return self.cursor.fetchone() is not None diff --git a/swh/provenance/postgresql_nopath/__init__.py b/swh/provenance/postgresql_nopath/__init__.py new file mode 100644 index 0000000..e69de29 diff --git a/swh/provenance/postgresql/provenance.py b/swh/provenance/postgresql_nopath/provenance.py similarity index 72% copy from swh/provenance/postgresql/provenance.py copy to swh/provenance/postgresql_nopath/provenance.py index 927ca4a..7728c8d 100644 --- a/swh/provenance/postgresql/provenance.py +++ b/swh/provenance/postgresql_nopath/provenance.py @@ -1,503 +1,442 @@ import itertools import logging import operator import os import psycopg2 import psycopg2.extras from ..model import DirectoryEntry, FileEntry from ..origin import OriginEntry -from .db_utils import connect, execute_sql +from ..postgresql.db_utils import connect, execute_sql from ..provenance import ProvenanceInterface from ..revision import RevisionEntry from datetime import datetime from typing import Any, Dict, Generator, List, Optional, Tuple -def normalize(path: bytes) -> bytes: - return path[2:] if path.startswith(bytes("." + os.path.sep, "utf-8")) else path - - def create_database(conn: psycopg2.extensions.connection, conninfo: dict, name: str): conn.set_isolation_level(psycopg2.extensions.ISOLATION_LEVEL_AUTOCOMMIT) # Normalize dbname to avoid issues when reconnecting below name = name.casefold() # Create new database dropping previous one if exists cursor = conn.cursor() cursor.execute(f"""DROP DATABASE IF EXISTS {name}""") cursor.execute(f"""CREATE DATABASE {name}""") conn.close() # Reconnect to server selecting newly created database to add tables conninfo["dbname"] = name conn = connect(conninfo) sqldir = os.path.dirname(os.path.realpath(__file__)) execute_sql(conn, os.path.join(sqldir, "provenance.sql")) ######################################################################################## ######################################################################################## ######################################################################################## -class ProvenancePostgreSQL(ProvenanceInterface): +class ProvenancePostgreSQLNoPath(ProvenanceInterface): def __init__(self, conn: psycopg2.extensions.connection): # TODO: consider adding a mutex for thread safety conn.set_isolation_level(psycopg2.extensions.ISOLATION_LEVEL_AUTOCOMMIT) self.conn = conn self.cursor = self.conn.cursor() self.insert_cache: Dict[str, Any] = {} self.remove_cache: Dict[str, Any] = {} self.select_cache: Dict[str, Any] = {} self.clear_caches() def clear_caches(self): self.insert_cache = { "content": dict(), - "content_early_in_rev": list(), - "content_in_dir": list(), + "content_early_in_rev": set(), + "content_in_dir": set(), "directory": dict(), - "directory_in_rev": list(), + "directory_in_rev": set(), "revision": dict(), "revision_before_rev": list(), "revision_in_org": list(), } self.remove_cache = {"directory": dict()} self.select_cache = {"content": dict(), "directory": dict(), "revision": dict()} def commit(self): result = False try: self.insert_all() self.clear_caches() result = True except Exception as error: # Unexpected error occurred, rollback all changes and log message logging.error(f"Unexpected error: {error}") return result def content_add_to_directory( self, directory: DirectoryEntry, blob: FileEntry, prefix: bytes ): - self.insert_cache["content_in_dir"].append( - (blob.id, directory.id, normalize(os.path.join(prefix, blob.name))) - ) + self.insert_cache["content_in_dir"].add((blob.id, directory.id)) def content_add_to_revision( self, revision: RevisionEntry, blob: FileEntry, prefix: bytes ): - self.insert_cache["content_early_in_rev"].append( - (blob.id, revision.id, normalize(os.path.join(prefix, blob.name))) - ) + self.insert_cache["content_early_in_rev"].add((blob.id, revision.id)) - def content_find_first(self, blobid: bytes) -> Tuple[bytes, bytes, datetime, bytes]: + def content_find_first( + self, blobid: bytes + ) -> Optional[Tuple[bytes, bytes, datetime, bytes]]: self.cursor.execute( - """SELECT content_location.sha1 AS blob, - revision.sha1 AS rev, - revision.date AS date, - content_location.path AS path - FROM (SELECT content_hex.sha1, - content_hex.rev, - location.path - FROM (SELECT content.sha1, - content_early_in_rev.rev, - content_early_in_rev.loc - FROM content_early_in_rev - JOIN content - ON content.id=content_early_in_rev.blob - WHERE content.sha1=%s - ) AS content_hex - JOIN location - ON location.id=content_hex.loc - ) AS content_location + """SELECT revision.sha1 AS rev, + revision.date AS date + FROM (SELECT content_early_in_rev.rev + FROM content_early_in_rev + JOIN content + ON content.id=content_early_in_rev.blob + WHERE content.sha1=%s + ) AS content_in_rev JOIN revision - ON revision.id=content_location.rev - ORDER BY date, rev, path ASC LIMIT 1""", + ON revision.id=content_in_rev.rev + ORDER BY date, rev ASC LIMIT 1""", (blobid,), ) - return self.cursor.fetchone() + row = self.cursor.fetchone() + if row is not None: + # TODO: query revision from the archive and look for blobid into a + # recursive directory_ls of the revision's root. + return blobid, row[0], row[1], b"" + return None def content_find_all( self, blobid: bytes ) -> Generator[Tuple[bytes, bytes, datetime, bytes], None, None]: self.cursor.execute( - """(SELECT content_location.sha1 AS blob, - revision.sha1 AS rev, - revision.date AS date, - content_location.path AS path - FROM (SELECT content_hex.sha1, - content_hex.rev, - location.path - FROM (SELECT content.sha1, - content_early_in_rev.rev, - content_early_in_rev.loc - FROM content_early_in_rev - JOIN content - ON content.id=content_early_in_rev.blob - WHERE content.sha1=%s - ) AS content_hex - JOIN location - ON location.id=content_hex.loc - ) AS content_location + """(SELECT revision.sha1 AS rev, + revision.date AS date + FROM (SELECT content_early_in_rev.rev + FROM content_early_in_rev + JOIN content + ON content.id=content_early_in_rev.blob + WHERE content.sha1=%s + ) AS content_in_rev JOIN revision - ON revision.id=content_location.rev + ON revision.id=content_in_rev.rev ) UNION - (SELECT content_prefix.sha1 AS blob, - revision.sha1 AS rev, - revision.date AS date, - content_prefix.path AS path - FROM (SELECT content_in_rev.sha1, - content_in_rev.rev, - CASE location.path - WHEN '' THEN content_in_rev.suffix - WHEN '.' THEN content_in_rev.suffix - ELSE (location.path || '/' || - content_in_rev.suffix)::unix_path - END AS path - FROM (SELECT content_suffix.sha1, - directory_in_rev.rev, - directory_in_rev.loc, - content_suffix.path AS suffix - FROM (SELECT content_hex.sha1, - content_hex.dir, - location.path - FROM (SELECT content.sha1, - content_in_dir.dir, - content_in_dir.loc - FROM content_in_dir - JOIN content - ON content_in_dir.blob=content.id - WHERE content.sha1=%s - ) AS content_hex - JOIN location - ON location.id=content_hex.loc - ) AS content_suffix - JOIN directory_in_rev - ON directory_in_rev.dir=content_suffix.dir - ) AS content_in_rev - JOIN location - ON location.id=content_in_rev.loc - ) AS content_prefix + (SELECT revision.sha1 AS rev, + revision.date AS date + FROM (SELECT directory_in_rev.rev + FROM (SELECT content_in_dir.dir + FROM content_in_dir + JOIN content + ON content_in_dir.blob=content.id + WHERE content.sha1=%s + ) AS content_dir + JOIN directory_in_rev + ON directory_in_rev.dir=content_dir.dir + ) AS content_in_rev JOIN revision - ON revision.id=content_prefix.rev + ON revision.id=content_in_rev.rev ) - ORDER BY date, rev, path""", + ORDER BY date, rev""", (blobid, blobid), ) # TODO: use POSTGRESQL EXPLAIN looking for query optimizations. - yield from self.cursor.fetchall() + for row in self.cursor.fetchall(): + # TODO: query revision from the archive and look for blobid into a + # recursive directory_ls of the revision's root. + yield blobid, row[0], row[1], b"" def content_get_early_date(self, blob: FileEntry) -> Optional[datetime]: # First check if the date is being modified by current transection. date = self.insert_cache["content"].get(blob.id, None) if date is None: # If not, check whether it's been query before date = self.select_cache["content"].get(blob.id, None) if date is None: # Otherwise, query the database and cache the value self.cursor.execute( """SELECT date FROM content WHERE sha1=%s""", (blob.id,) ) row = self.cursor.fetchone() date = row[0] if row is not None else None self.select_cache["content"][blob.id] = date return date def content_get_early_dates(self, blobs: List[FileEntry]) -> Dict[bytes, datetime]: dates = {} pending = [] for blob in blobs: # First check if the date is being modified by current transection. date = self.insert_cache["content"].get(blob.id, None) if date is not None: dates[blob.id] = date else: # If not, check whether it's been query before date = self.select_cache["content"].get(blob.id, None) if date is not None: dates[blob.id] = date else: pending.append(blob.id) if pending: # Otherwise, query the database and cache the values values = ", ".join(itertools.repeat("%s", len(pending))) self.cursor.execute( f"""SELECT sha1, date FROM content WHERE sha1 IN ({values})""", tuple(pending), ) for row in self.cursor.fetchall(): dates[row[0]] = row[1] self.select_cache["content"][row[0]] = row[1] return dates def content_set_early_date(self, blob: FileEntry, date: datetime): self.insert_cache["content"][blob.id] = date def directory_add_to_revision( self, revision: RevisionEntry, directory: DirectoryEntry, path: bytes ): - self.insert_cache["directory_in_rev"].append( - (directory.id, revision.id, normalize(path)) - ) + self.insert_cache["directory_in_rev"].add((directory.id, revision.id)) def directory_get_date_in_isochrone_frontier( self, directory: DirectoryEntry ) -> Optional[datetime]: # First check if the date is being modified by current transection. date = self.insert_cache["directory"].get(directory.id, None) if date is None and directory.id not in self.remove_cache["directory"]: # If not, check whether it's been query before date = self.select_cache["directory"].get(directory.id, None) if date is None: # Otherwise, query the database and cache the value self.cursor.execute( """SELECT date FROM directory WHERE sha1=%s""", (directory.id,) ) row = self.cursor.fetchone() date = row[0] if row is not None else None self.select_cache["directory"][directory.id] = date return date def directory_get_dates_in_isochrone_frontier( self, dirs: List[DirectoryEntry] ) -> Dict[bytes, datetime]: dates = {} pending = [] for directory in dirs: # First check if the date is being modified by current transection. date = self.insert_cache["directory"].get(directory.id, None) if date is not None: dates[directory.id] = date elif directory.id not in self.remove_cache["directory"]: # If not, check whether it's been query before date = self.select_cache["directory"].get(directory.id, None) if date is not None: dates[directory.id] = date else: pending.append(directory.id) if pending: # Otherwise, query the database and cache the values values = ", ".join(itertools.repeat("%s", len(pending))) self.cursor.execute( f"""SELECT sha1, date FROM directory WHERE sha1 IN ({values})""", tuple(pending), ) for row in self.cursor.fetchall(): dates[row[0]] = row[1] self.select_cache["directory"][row[0]] = row[1] return dates def directory_invalidate_in_isochrone_frontier(self, directory: DirectoryEntry): self.remove_cache["directory"][directory.id] = None self.insert_cache["directory"].pop(directory.id, None) def directory_set_date_in_isochrone_frontier( self, directory: DirectoryEntry, date: datetime ): self.insert_cache["directory"][directory.id] = date self.remove_cache["directory"].pop(directory.id, None) def insert_all(self): # Performe insertions with cached information if self.insert_cache["content"]: psycopg2.extras.execute_values( self.cursor, """LOCK TABLE ONLY content; INSERT INTO content(sha1, date) VALUES %s ON CONFLICT (sha1) DO UPDATE SET date=LEAST(EXCLUDED.date,content.date)""", self.insert_cache["content"].items(), ) self.insert_cache["content"].clear() if self.insert_cache["directory"]: psycopg2.extras.execute_values( self.cursor, """LOCK TABLE ONLY directory; INSERT INTO directory(sha1, date) VALUES %s ON CONFLICT (sha1) DO UPDATE SET date=LEAST(EXCLUDED.date,directory.date)""", self.insert_cache["directory"].items(), ) self.insert_cache["directory"].clear() if self.insert_cache["revision"]: psycopg2.extras.execute_values( self.cursor, """LOCK TABLE ONLY revision; INSERT INTO revision(sha1, date) VALUES %s ON CONFLICT (sha1) DO UPDATE SET date=LEAST(EXCLUDED.date,revision.date)""", self.insert_cache["revision"].items(), ) self.insert_cache["revision"].clear() # Relations should come after ids for elements were resolved if self.insert_cache["content_early_in_rev"]: self.insert_location("content", "revision", "content_early_in_rev") if self.insert_cache["content_in_dir"]: self.insert_location("content", "directory", "content_in_dir") if self.insert_cache["directory_in_rev"]: self.insert_location("directory", "revision", "directory_in_rev") # if self.insert_cache["revision_before_rev"]: # psycopg2.extras.execute_values( # self.cursor, # """INSERT INTO revision_before_rev VALUES %s # ON CONFLICT DO NOTHING""", # self.insert_cache["revision_before_rev"], # ) # self.insert_cache["revision_before_rev"].clear() # if self.insert_cache["revision_in_org"]: # psycopg2.extras.execute_values( # self.cursor, # """INSERT INTO revision_in_org VALUES %s # ON CONFLICT DO NOTHING""", # self.insert_cache["revision_in_org"], # ) # self.insert_cache["revision_in_org"].clear() def insert_location(self, src0_table, src1_table, dst_table): # Resolve src0 ids src0_values = dict().fromkeys( map(operator.itemgetter(0), self.insert_cache[dst_table]) ) values = ", ".join(itertools.repeat("%s", len(src0_values))) self.cursor.execute( f"""SELECT sha1, id FROM {src0_table} WHERE sha1 IN ({values})""", tuple(src0_values), ) src0_values = dict(self.cursor.fetchall()) # Resolve src1 ids src1_values = dict().fromkeys( map(operator.itemgetter(1), self.insert_cache[dst_table]) ) values = ", ".join(itertools.repeat("%s", len(src1_values))) self.cursor.execute( f"""SELECT sha1, id FROM {src1_table} WHERE sha1 IN ({values})""", tuple(src1_values), ) src1_values = dict(self.cursor.fetchall()) - # Resolve location ids - location = dict().fromkeys( - map(operator.itemgetter(2), self.insert_cache[dst_table]) - ) - location = dict( - psycopg2.extras.execute_values( - self.cursor, - """LOCK TABLE ONLY location; - INSERT INTO location(path) VALUES %s - ON CONFLICT (path) DO - UPDATE SET path=EXCLUDED.path - RETURNING path, id""", - map(lambda path: (path,), location.keys()), - fetch=True, - ) - ) - # Insert values in dst_table rows = map( - lambda row: (src0_values[row[0]], src1_values[row[1]], location[row[2]]), + lambda row: (src0_values[row[0]], src1_values[row[1]]), self.insert_cache[dst_table], ) psycopg2.extras.execute_values( self.cursor, f"""INSERT INTO {dst_table} VALUES %s ON CONFLICT DO NOTHING""", rows, ) self.insert_cache[dst_table].clear() def origin_get_id(self, origin: OriginEntry) -> int: if origin.id is None: # Insert origin in the DB and return the assigned id self.cursor.execute( """INSERT INTO origin (url) VALUES (%s) ON CONFLICT DO NOTHING RETURNING id""", (origin.url,), ) return self.cursor.fetchone()[0] else: return origin.id def revision_add(self, revision: RevisionEntry): # Add current revision to the compact DB self.insert_cache["revision"][revision.id] = revision.date def revision_add_before_revision( self, relative: RevisionEntry, revision: RevisionEntry ): self.insert_cache["revision_before_rev"].append((revision.id, relative.id)) def revision_add_to_origin(self, origin: OriginEntry, revision: RevisionEntry): self.insert_cache["revision_in_org"].append((revision.id, origin.id)) def revision_get_early_date(self, revision: RevisionEntry) -> Optional[datetime]: date = self.insert_cache["revision"].get(revision.id, None) if date is None: # If not, check whether it's been query before date = self.select_cache["revision"].get(revision.id, None) if date is None: # Otherwise, query the database and cache the value self.cursor.execute( """SELECT date FROM revision WHERE sha1=%s""", (revision.id,) ) row = self.cursor.fetchone() date = row[0] if row is not None else None self.select_cache["revision"][revision.id] = date return date def revision_get_prefered_origin(self, revision: RevisionEntry) -> int: # TODO: adapt this method to consider cached values self.cursor.execute( """SELECT COALESCE(org,0) FROM revision WHERE sha1=%s""", (revision.id,) ) row = self.cursor.fetchone() # None means revision is not in database; # 0 means revision has no prefered origin return row[0] if row is not None and row[0] != 0 else None def revision_in_history(self, revision: RevisionEntry) -> bool: # TODO: adapt this method to consider cached values self.cursor.execute( """SELECT 1 FROM revision_before_rev JOIN revision ON revision.id=revision_before_rev.prev WHERE revision.sha1=%s""", (revision.id,), ) return self.cursor.fetchone() is not None def revision_set_prefered_origin( self, origin: OriginEntry, revision: RevisionEntry ): # TODO: adapt this method to consider cached values self.cursor.execute( """UPDATE revision SET org=%s WHERE sha1=%s""", (origin.id, revision.id) ) def revision_visited(self, revision: RevisionEntry) -> bool: # TODO: adapt this method to consider cached values self.cursor.execute( """SELECT 1 FROM revision_in_org JOIN revision ON revision.id=revision_in_org.rev WHERE revision.sha1=%s""", (revision.id,), ) return self.cursor.fetchone() is not None diff --git a/swh/provenance/postgresql_nopath/provenance.sql b/swh/provenance/postgresql_nopath/provenance.sql new file mode 100644 index 0000000..7d7f4ae --- /dev/null +++ b/swh/provenance/postgresql_nopath/provenance.sql @@ -0,0 +1,130 @@ +-- a Git object ID, i.e., a Git-style salted SHA1 checksum +drop domain if exists sha1_git cascade; +create domain sha1_git as bytea check (length(value) = 20); + +-- UNIX path (absolute, relative, individual path component, etc.) +drop domain if exists unix_path cascade; +create domain unix_path as bytea; + + +drop table if exists content; +create table content +( + id bigserial primary key, -- internal identifier of the content blob + sha1 sha1_git unique not null, -- intrinsic identifier of the content blob + date timestamptz not null -- timestamp of the revision where the blob appears early +); + +comment on column content.id is 'Content internal identifier'; +comment on column content.sha1 is 'Content intrinsic identifier'; +comment on column content.date is 'Earliest timestamp for the content (first seen time)'; + + +drop table if exists content_early_in_rev; +create table content_early_in_rev +( + blob bigint not null, -- internal identifier of the content blob + rev bigint not null, -- internal identifier of the revision where the blob appears for the first time + primary key (blob, rev) + -- foreign key (blob) references content (id), + -- foreign key (rev) references revision (id) +); + +comment on column content_early_in_rev.blob is 'Content internal identifier'; +comment on column content_early_in_rev.rev is 'Revision internal identifier'; + + +drop table if exists content_in_dir; +create table content_in_dir +( + blob bigint not null, -- internal identifier of the content blob + dir bigint not null, -- internal identifier of the directory contaning the blob + primary key (blob, dir) + -- foreign key (blob) references content (id), + -- foreign key (dir) references directory (id) +); + +comment on column content_in_dir.blob is 'Content internal identifier'; +comment on column content_in_dir.dir is 'Directory internal identifier'; + + +drop table if exists directory; +create table directory +( + id bigserial primary key, -- internal identifier of the directory appearing in an isochrone inner frontier + sha1 sha1_git unique not null, -- intrinsic identifier of the directory + date timestamptz not null -- max timestamp among those of the directory children's +); + +comment on column directory.id is 'Directory internal identifier'; +comment on column directory.sha1 is 'Directory intrinsic identifier'; +comment on column directory.date is 'Latest timestamp for the content in the directory'; + + +drop table if exists directory_in_rev; +create table directory_in_rev +( + dir bigint not null, -- internal identifier of the directory appearing in the revision + rev bigint not null, -- internal identifier of the revision containing the directory + primary key (dir, rev) + -- foreign key (dir) references directory (id), + -- foreign key (rev) references revision (id) +); + +comment on column directory_in_rev.dir is 'Directory internal identifier'; +comment on column directory_in_rev.rev is 'Revision internal identifier'; + + +drop table if exists origin; +create table origin +( + id bigserial primary key, -- internal identifier of the origin + url unix_path unique not null -- url of the origin +); + +comment on column origin.id is 'Origin internal identifier'; +comment on column origin.url is 'URL of the origin'; + + +drop table if exists revision; +create table revision +( + id bigserial primary key, -- internal identifier of the revision + sha1 sha1_git unique not null, -- intrinsic identifier of the revision + date timestamptz not null, -- timestamp of the revision + org bigint -- id of the prefered origin + -- foreign key (org) references origin (id) +); + +comment on column revision.id is 'Revision internal identifier'; +comment on column revision.sha1 is 'Revision intrinsic identifier'; +comment on column revision.date is 'Revision timestamp'; +comment on column revision.org is 'Prefered origin for the revision'; + + +drop table if exists revision_before_rev; +create table revision_before_rev +( + prev bigserial not null, -- internal identifier of the source revision + next bigserial not null, -- internal identifier of the destination revision + primary key (prev, next) + -- foreign key (prev) references revision (id), + -- foreign key (next) references revision (id) +); + +comment on column revision_before_rev.prev is 'Source revision internal identifier'; +comment on column revision_before_rev.next is 'Destination revision internal identifier'; + + +drop table if exists revision_in_org; +create table revision_in_org +( + rev bigint not null, -- internal identifier of the revision poined by the origin + org bigint not null, -- internal identifier of the origin that points to the revision + primary key (rev, org) + -- foreign key (rev) references revision (id), + -- foreign key (org) references origin (id) +); + +comment on column revision_in_org.rev is 'Revision internal identifier'; +comment on column revision_in_org.org is 'Origin internal identifier'; diff --git a/swh/provenance/provenance.py b/swh/provenance/provenance.py index 26025a8..ccf2317 100644 --- a/swh/provenance/provenance.py +++ b/swh/provenance/provenance.py @@ -1,425 +1,427 @@ import os from .archive import ArchiveInterface from .model import DirectoryEntry, FileEntry from .origin import OriginEntry from .revision import RevisionEntry from datetime import datetime from typing import Dict, Generator, List, Optional, Tuple # TODO: consider moving to path utils file together with normalize. def is_child(path: bytes, prefix: bytes) -> bool: return path != prefix and os.path.dirname(path) == prefix class ProvenanceInterface: def __init__(self, **kwargs): raise NotImplementedError def commit(self): raise NotImplementedError def content_add_to_directory( self, directory: DirectoryEntry, blob: FileEntry, prefix: bytes ): raise NotImplementedError def content_add_to_revision( self, revision: RevisionEntry, blob: FileEntry, prefix: bytes ): raise NotImplementedError - def content_find_first(self, blobid: bytes) -> Tuple[bytes, bytes, datetime, bytes]: + def content_find_first( + self, blobid: bytes + ) -> Optional[Tuple[bytes, bytes, datetime, bytes]]: raise NotImplementedError def content_find_all( self, blobid: bytes ) -> Generator[Tuple[bytes, bytes, datetime, bytes], None, None]: raise NotImplementedError def content_get_early_date(self, blob: FileEntry) -> Optional[datetime]: raise NotImplementedError def content_get_early_dates(self, blobs: List[FileEntry]) -> Dict[bytes, datetime]: raise NotImplementedError def content_set_early_date(self, blob: FileEntry, date: datetime): raise NotImplementedError def directory_add_to_revision( self, revision: RevisionEntry, directory: DirectoryEntry, path: bytes ): raise NotImplementedError def directory_get_date_in_isochrone_frontier( self, directory: DirectoryEntry ) -> Optional[datetime]: raise NotImplementedError def directory_get_dates_in_isochrone_frontier( self, dirs: List[DirectoryEntry] ) -> Dict[bytes, datetime]: raise NotImplementedError def directory_invalidate_in_isochrone_frontier(self, directory: DirectoryEntry): raise NotImplementedError def directory_set_date_in_isochrone_frontier( self, directory: DirectoryEntry, date: datetime ): raise NotImplementedError def origin_get_id(self, origin: OriginEntry) -> int: raise NotImplementedError def revision_add(self, revision: RevisionEntry): raise NotImplementedError def revision_add_before_revision( self, relative: RevisionEntry, revision: RevisionEntry ): raise NotImplementedError def revision_add_to_origin(self, origin: OriginEntry, revision: RevisionEntry): raise NotImplementedError def revision_get_early_date(self, revision: RevisionEntry) -> Optional[datetime]: raise NotImplementedError def revision_get_prefered_origin(self, revision: RevisionEntry) -> int: raise NotImplementedError def revision_in_history(self, revision: RevisionEntry) -> bool: raise NotImplementedError def revision_set_prefered_origin( self, origin: OriginEntry, revision: RevisionEntry ): raise NotImplementedError def revision_visited(self, revision: RevisionEntry) -> bool: raise NotImplementedError def directory_process_content( provenance: ProvenanceInterface, directory: DirectoryEntry, relative: DirectoryEntry ): stack = [(directory, b"")] while stack: current, prefix = stack.pop() for child in iter(current): if isinstance(child, FileEntry): # Add content to the relative directory with the computed prefix. provenance.content_add_to_directory(relative, child, prefix) else: # Recursively walk the child directory. stack.append((child, os.path.join(prefix, child.name))) def directory_update_content( stack: List[Tuple[DirectoryEntry, bytes]], provenance: ProvenanceInterface, revision: RevisionEntry, directory: DirectoryEntry, prefix: bytes, subdirs: Optional[List[DirectoryEntry]] = None, blobs: Optional[List[FileEntry]] = None, blobdates: Optional[Dict[bytes, datetime]] = None, ): assert revision.date is not None # Init optional parameters if not provided. if subdirs is None: subdirs = [child for child in directory if isinstance(child, DirectoryEntry)] if blobs is None: blobs = [child for child in directory if isinstance(child, FileEntry)] if blobdates is None: blobdates = provenance.content_get_early_dates(blobs) # Iterate over blobs updating their date if necessary. for blob in blobs: date = blobdates.get(blob.id, None) if date is None or revision.date < date: provenance.content_set_early_date(blob, revision.date) # Push all subdirectories with its corresponding path to analyze them # recursively. for subdir in subdirs: stack.append((subdir, os.path.join(prefix, subdir.name))) def origin_add(provenance: ProvenanceInterface, origin: OriginEntry): # TODO: refactor to iterate over origin visit statuses and commit only once # per status. origin.id = provenance.origin_get_id(origin) for revision in origin.revisions: origin_add_revision(provenance, origin, revision) # Commit after each revision provenance.commit() # TODO: verify this! def origin_add_revision( provenance: ProvenanceInterface, origin: OriginEntry, revision: RevisionEntry ): stack: List[Tuple[Optional[RevisionEntry], RevisionEntry]] = [(None, revision)] while stack: relative, current = stack.pop() # Check if current revision has no prefered origin and update if necessary. prefered = provenance.revision_get_prefered_origin(current) if prefered is None: provenance.revision_set_prefered_origin(origin, current) ######################################################################## if relative is None: # This revision is pointed directly by the origin. visited = provenance.revision_visited(current) provenance.revision_add_to_origin(origin, current) if not visited: stack.append((current, current)) else: # This revision is a parent of another one in the history of the # relative revision. for parent in iter(current): visited = provenance.revision_visited(parent) if not visited: # The parent revision has never been seen before pointing # directly to an origin. known = provenance.revision_in_history(parent) if known: # The parent revision is already known in some other # revision's history. We should point it directly to # the origin and (eventually) walk its history. stack.append((None, parent)) else: # The parent revision was never seen before. We should # walk its history and associate it with the same # relative revision. provenance.revision_add_before_revision(relative, parent) stack.append((relative, parent)) else: # The parent revision already points to an origin, so its # history was properly processed before. We just need to # make sure it points to the current origin as well. provenance.revision_add_to_origin(origin, parent) def revision_add( provenance: ProvenanceInterface, archive: ArchiveInterface, revision: RevisionEntry ): assert revision.date is not None assert revision.root is not None # Processed content starting from the revision's root directory. date = provenance.revision_get_early_date(revision) if date is None or revision.date < date: provenance.revision_add(revision) revision_process_content( provenance, revision, DirectoryEntry(archive, revision.root, b"") ) # TODO: improve this! Maybe using a max attempt counter? # Idealy Provenance class should guarante that a commit never fails. while not provenance.commit(): continue def revision_process_content( provenance: ProvenanceInterface, revision: RevisionEntry, root: DirectoryEntry ): assert revision.date is not None # Stack of directories (and their paths) to be processed. stack: List[Tuple[DirectoryEntry, bytes]] = [(root, root.name)] # This dictionary will hold the computed dates for visited subdirectories inside the # isochrone frontier. innerdirs: Dict[bytes, Tuple[DirectoryEntry, datetime]] = {} # This dictionary will hold the computed dates for visited subdirectories outside # the isochrone frontier which are candidates to be added to the outer frontier (if # their parent is in the inner frontier). outerdirs: Dict[bytes, Tuple[DirectoryEntry, datetime]] = {} while stack: # Get next directory to process and query its date right before processing to be # sure we get the most recently updated value. current, prefix = stack.pop() date = provenance.directory_get_date_in_isochrone_frontier(current) if date is None: # The directory has never been seen on the outer isochrone frontier of # previously processed revisions. Its children should be analyzed. blobs = [child for child in current if isinstance(child, FileEntry)] subdirs = [child for child in current if isinstance(child, DirectoryEntry)] # Get the list of ids with no duplicates to ensure we have available dates # for all the elements. This prevents taking a wrong decision when a blob # occurs more than once in the same directory. ids = list( dict.fromkeys( [child.id for child in blobs] + [child.id for child in subdirs] ) ) if ids: # Known dates for the blobs in the current directory. blobdates = provenance.content_get_early_dates(blobs) # Known dates for the subdirectories in the current directory that # belong to the outer isochrone frontier of some previously processed # revision. knowndates = provenance.directory_get_dates_in_isochrone_frontier( subdirs ) # Known dates for the subdirectories in the current directory that are # inside the isochrone frontier of the revision. innerdates = { innerdir.id: innerdate for path, (innerdir, innerdate) in innerdirs.items() if is_child(path, prefix) } # Known dates for the subdirectories in the current directory that are # outside the isochrone frontier of the revision. outerdates = { outerdir.id: outerdate for path, (outerdir, outerdate) in outerdirs.items() if is_child(path, prefix) } # All known dates for child nodes of the current directory. assert not (innerdates.keys() & outerdates.keys()) dates = list( {**blobdates, **knowndates, **innerdates, **outerdates}.values() ) if len(dates) == len(ids): # All child nodes of current directory are already known. maxdate = max(dates) if maxdate < revision.date: # The directory is outside the isochrone frontier of the # revision. It is a candidate to be added to the outer frontier. outerdirs[prefix] = (current, maxdate) # Its children are removed since they are no longer candidates. outerdirs = { path: outerdir for path, outerdir in outerdirs.items() if not is_child(path, prefix) } elif maxdate == revision.date: # The current directory is inside the isochrone frontier. innerdirs[prefix] = (current, revision.date) # Add blobs present in this level to the revision. No need to # update dates as they are at most equal to current one. for blob in blobs: provenance.content_add_to_revision(revision, blob, prefix) # If any of its children was found outside the frontier it # should be added to the outer frontier now. if outerdates: for path, (outerdir, outerdate) in outerdirs.items(): if is_child(path, prefix): provenance.directory_set_date_in_isochrone_frontier( outerdir, outerdate ) provenance.directory_add_to_revision( revision, outerdir, path ) directory_process_content( provenance, directory=outerdir, relative=outerdir, ) # Removed processed elements to avoid duplicating work. outerdirs = { path: outerdir for path, outerdir in outerdirs.items() if not is_child(path, prefix) } # There can still be subdirectories that are known to be in the # outter isochrone frontier of previous processed revisions. # Thus, they are not in the list of candidates but have to be # added to current revisions as well. for subdir in subdirs: knowndate = knowndates.get(subdir.id, None) if knowndate is not None and knowndate <= revision.date: # Less or equal since the directory could have been # added to the outer isochrone frontier when processing # a different directory's subtree of this very same # revision. provenance.directory_add_to_revision( revision, subdir, os.path.join(prefix, subdir.name) ) else: # The revision is out of order. The current directory does not # belong to the outer isochrone frontier of any previously # processed revision yet all its children nodes are known. They # should be re-analyzed (and timestamps eventually updated) and # current directory updated after them. stack.append((current, prefix)) directory_update_content( stack, provenance, revision, current, prefix, subdirs=subdirs, blobs=blobs, blobdates=blobdates, ) else: # Al least one child node is unknown, ie. the current directory is # inside the isochrone frontier of the current revision. Its child # nodes should be analyzed and current directory updated after them. stack.append((current, prefix)) directory_update_content( stack, provenance, revision, current, prefix, subdirs=subdirs, blobs=blobs, blobdates=blobdates, ) else: # Empty directory. Just consider it to be in the inner frontier of # current revision (ie. all its children are already "known"). innerdirs[prefix] = (current, revision.date) elif revision.date < date: # The revision is out of order. The current directory belongs to the outer # isochrone frontier of some previously processed revison but current # revision is earlier. The frontier record should be invalidated, children # nodes re-analyzed (and timestamps eventually updated) and current # directory updated after them. stack.append((current, prefix)) provenance.directory_invalidate_in_isochrone_frontier(current) directory_update_content(stack, provenance, revision, current, prefix) else: # The directory has already been seen on the outer isochrone frontier of an - # earlier revision. Just add it to the current revision. - provenance.directory_add_to_revision(revision, current, prefix) + # earlier revision. Just stop the recursion here. + pass if root.name in outerdirs: # Only the root directory should be considered at this point. outerdir, outerdate = outerdirs[root.name] provenance.directory_set_date_in_isochrone_frontier(outerdir, outerdate) provenance.directory_add_to_revision(revision, outerdir, root.name) directory_process_content(provenance, directory=outerdir, relative=outerdir)