diff --git a/sql/swh-func.sql b/sql/swh-func.sql index fd3a61bd..015b3d1e 100644 --- a/sql/swh-func.sql +++ b/sql/swh-func.sql @@ -1,89 +1,168 @@ -- create a temporary table called tmp_TBLNAME, mimicking existing table -- TBLNAME -- -- Args: -- tblname: name of the table to mimick create or replace function swh_mktemp(tblname regclass) returns void language plpgsql as $$ begin execute format(' create temporary table tmp_%I (like %I including defaults) on commit drop ', tblname, tblname); return; end $$; +-- create a temporary table for directory entries called tmp_directory_entry_TBLNAME, mimicking existing table +-- directory_entry_TBLNAME with an extra dir_id (sha1_git) column +-- +-- Args: +-- tblname: name of the table to mimick +create or replace function swh_mktemp_dir_entry(tblname regclass) + returns void + language plpgsql +as $$ +begin + execute format(' + create temporary table tmp_%I + (like %I including defaults, dir_id sha1_git) + on commit drop; alter table tmp_%I alter id set default nextval(''%I_id_seq''); + ', tblname, tblname, tblname, tblname); + return; +end +$$; + -- a content signature is a set of cryptographic checksums that we use to -- uniquely identify content, for the purpose of verifying if we already have -- some content or not during content injection create type content_signature as ( sha1 sha1, sha1_git sha1_git, sha256 sha256 ); -- check which entries of tmp_content are missing from content -- -- operates in bulk: 0. swh_mktemp(content), 1. COPY to tmp_content, -- 2. call this function create or replace function swh_content_missing() returns setof content_signature language plpgsql as $$ begin return query select sha1, sha1_git, sha256 from tmp_content except select sha1, sha1_git, sha256 from content; return; end $$; -- add tmp_content entries to content, skipping duplicates -- -- operates in bulk: 0. swh_mktemp(content), 1. COPY to tmp_content, -- 2. call this function create or replace function swh_content_add() returns void language plpgsql as $$ declare rows bigint; begin insert into content (sha1, sha1_git, sha256, length, status) select distinct sha1, sha1_git, sha256, length, status from tmp_content where (sha1, sha1_git, sha256) in (select * from swh_content_missing()); -- TODO XXX use postgres 9.5 "UPSERT" support here, when available. -- Specifically, using "INSERT .. ON CONFLICT IGNORE" we can avoid -- the extra swh_content_missing() query here. return; end $$; -- check which entries of tmp_directory are missing from directory -- -- operates in bulk: 0. swh_mktemp(directory), 1. COPY to tmp_directory, -- 2. call this function create or replace function swh_directory_missing() - returns setof sha1_git + returns setof directory language plpgsql as $$ begin return query select id from tmp_directory except select id from directory; return; end $$; + +-- Add directory_entry_dir entries +create or replace function swh_directory_entry_dir_add() + returns void + language plpgsql +as $$ +begin + with inserted_entries as ( + insert into directory_entry_dir ( + target, perms, atime, mtime, ctime, name + ) + select t.target, t.perms, t.atime, t.mtime, t.ctime, t.name + from tmp_directory_entry_dir t + returning id, target, perms, atime, mtime, ctime, name + ) insert into directory_list_dir (entry_id, dir_id) + select i.id, t.dir_id from tmp_directory_entry_dir t left join inserted_entries i on + t.target = i.target and t.perms = i.perms and coalesce(t.atime, '-infinity') = coalesce(i.atime, '-infinity') and coalesce(t.ctime, '-infinity') = coalesce(i.ctime, '-infinity') and coalesce(t.mtime, '-infinity') = coalesce(i.mtime, '-infinity') and t.name = i.name; + return; +end +$$; + +-- Add directory_entry_file entries +create or replace function swh_directory_entry_file_add() + returns void + language plpgsql +as $$ +begin + with inserted_entries as ( + insert into directory_entry_file ( + target, perms, atime, mtime, ctime, name + ) + select t.target, t.perms, t.atime, t.mtime, t.ctime, t.name + from tmp_directory_entry_file t + returning id, target, perms, atime, mtime, ctime, name + ) insert into directory_list_file (entry_id, dir_id) + select i.id, t.dir_id from tmp_directory_entry_file t left join inserted_entries i on + t.target = i.target and t.perms = i.perms and coalesce(t.atime, '-infinity') = coalesce(i.atime, '-infinity') and coalesce(t.ctime, '-infinity') = coalesce(i.ctime, '-infinity') and coalesce(t.mtime, '-infinity') = coalesce(i.mtime, '-infinity') and t.name = i.name; + return; +end +$$; + +-- Add directory_entry_rev entries +create or replace function swh_directory_entry_rev_add() + returns void + language plpgsql +as $$ +begin + with inserted_entries as ( + insert into directory_entry_rev ( + target, perms, atime, mtime, ctime, name + ) + select t.target, t.perms, t.atime, t.mtime, t.ctime, t.name + from tmp_directory_entry_rev t + returning id, target, perms, atime, mtime, ctime, name + ) insert into directory_list_rev (entry_id, dir_id) + select i.id, t.dir_id from tmp_directory_entry_rev t left join inserted_entries i on + t.target = i.target and t.perms = i.perms and coalesce(t.atime, '-infinity') = coalesce(i.atime, '-infinity') and coalesce(t.ctime, '-infinity') = coalesce(i.ctime, '-infinity') and coalesce(t.mtime, '-infinity') = coalesce(i.mtime, '-infinity') and t.name = i.name; + return; +end +$$; diff --git a/swh/storage/db.py b/swh/storage/db.py index 57c67ffc..bca7881e 100644 --- a/swh/storage/db.py +++ b/swh/storage/db.py @@ -1,126 +1,132 @@ # Copyright (C) 2015 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import binascii import functools import psycopg2 import tempfile from contextlib import contextmanager TMP_CONTENT_TABLE = 'tmp_content' def stored_procedure(stored_proc): """decorator to execute remote stored procedure, specified as argument Generally, the body of the decorated function should be empty. If it is not, the stored procedure will be executed first; the function body then. """ def wrap(meth): @functools.wraps(meth) def _meth(self, *args, **kwargs): cur = kwargs.get('cur', None) self._cursor(cur).execute('SELECT %s()' % stored_proc) meth(self, *args, **kwargs) return _meth return wrap class Db: """Proxy to the SWH DB, with wrappers around stored procedures """ @classmethod def connect(cls, *args, **kwargs): """factory method to create a DB proxy Accepts all arguments of psycopg2.connect; only some specific possibilities are reported below. Args: connstring: libpq2 connection string """ conn = psycopg2.connect(*args, **kwargs) return cls(conn) def _cursor(self, cur_arg): """get a cursor: from cur_arg if given, or a fresh one otherwise meant to avoid boilerplate if/then/else in methods that proxy stored procedures """ if cur_arg is not None: return cur_arg # elif self.cur is not None: # return self.cur else: return self.conn.cursor() def __init__(self, conn): """create a DB proxy Args: conn: psycopg2 connection to the SWH DB """ self.conn = conn @contextmanager def transaction(self): """context manager to execute within a DB transaction Yields: a psycopg2 cursor """ with self.conn.cursor() as cur: try: yield cur self.conn.commit() except: if not self.conn.closed: self.conn.rollback() raise def mktemp(self, tblname, cur=None): self._cursor(cur).execute('SELECT swh_mktemp(%s)', (tblname,)) + def mktemp_dir_entry(self, entry_type, cur=None): + self._cursor(cur).execute('SELECT swh_mktemp_dir_entry(%s)', + (('directory_entry_%s' % entry_type),)) + def copy_to(self, items, tblname, columns, cur=None, item_cb=None): def escape(data): + if data is None: + return '\\N' if isinstance(data, bytes): return '\\\\x%s' % binascii.hexlify(data).decode('ascii') else: return str(data) with tempfile.TemporaryFile('w+') as f: for d in items: if item_cb is not None: item_cb(d) line = '\t'.join([escape(d[k]) for k in columns]) + '\n' f.write(line) f.seek(0) self._cursor(cur).copy_from(f, tblname, columns=columns) @stored_procedure('swh_content_add') def content_add_from_temp(self, cur=None): pass def content_missing_from_temp(self, cur=None): cur = self._cursor(cur) cur.execute("""SELECT sha1, sha1_git, sha256 FROM swh_content_missing()""") yield from cur def directory_missing_from_temp(self, cur=None): cur = self._cursor(cur) cur.execute("SELECT id FROM swh_directory_missing()") yield from cur diff --git a/swh/storage/storage.py b/swh/storage/storage.py index 95539137..0c2aaff9 100644 --- a/swh/storage/storage.py +++ b/swh/storage/storage.py @@ -1,228 +1,271 @@ # Copyright (C) 2015 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import functools +import itertools import psycopg2 +from collections import defaultdict + from .db import Db from .objstorage import ObjStorage def db_transaction(meth): """decorator to execute Storage methods within DB transactions Decorated methods will have access to the following attributes: self.cur: psycopg2 DB cursor """ @functools.wraps(meth) def _meth(self, *args, **kwargs): with self.db.transaction() as cur: try: self.cur = cur return meth(self, *args, **kwargs) finally: self.cur = None return _meth class Storage(): """SWH storage proxy, encompassing DB and object storage """ def __init__(self, db_conn, obj_root): """ Args: db_conn: either a libpq connection string, or a psycopg2 connection obj_root: path to the root of the object storage """ if isinstance(db_conn, psycopg2.extensions.connection): self.db = Db(db_conn) else: self.db = Db.connect(db_conn) self.objstorage = ObjStorage(obj_root) @db_transaction def content_add(self, content): """Add content blobs to the storage Note: in case of DB errors, objects might have already been added to the object storage and will not be removed. Since addition to the object storage is idempotent, that should not be a problem. Args: content: iterable of dictionaries representing individual pieces of content to add. Each dictionary has the following keys: - data (bytes): the actual content - length (int): content length - one key for each checksum algorithm in swh.core.hashutil.ALGORITHMS, mapped to the corresponding checksum """ (db, cur) = (self.db, self.cur) + # create temporary table for metadata injection db.mktemp('content', cur) def add_to_objstorage(cont): self.objstorage.add_bytes(cont['data'], obj_id=cont['sha1']) db.copy_to(content, 'tmp_content', ['sha1', 'sha1_git', 'sha256', 'length'], cur, item_cb=add_to_objstorage) # move metadata in place db.content_add_from_temp(cur) db.conn.commit() @db_transaction def content_missing(self, content): """List content missing from storage Args: content: iterable of dictionaries containing one key for each checksum algorithm in swh.core.hashutil.ALGORITHMS, mapped to the corresponding checksum, and a length key mapped to the content length. Returns: an iterable of sha1s missing from the storage Raises: TODO: an exception when we get a hash collision. """ (db, cur) = (self.db, self.cur) # Create temporary table for metadata injection db.mktemp('content', cur) db.copy_to(content, 'tmp_content', ['sha1', 'sha1_git', 'sha256', 'length'], cur) for obj in db.content_missing_from_temp(cur): yield obj[0].tobytes() def directory_add(self, directories): """Add directories to the storage Args: directories: iterable of dictionaries representing the individual directories to add. Each dict has the following keys: - id (sha1_git): the id of the directory to add - entries (list): list of dicts for each entry in the directory. Each dict has the following keys: - name (bytes) - type (one of 'file', 'dir', 'rev'): type of the directory entry (file, directory, revision) - - id (sha1_git): id of the object pointed at by the + - target (sha1_git): id of the object pointed at by the directory entry - perms (int): entry permissions - atime (datetime.DateTime): entry access time - ctime (datetime.DateTime): entry creation time - mtime (datetime.DateTime): entry modification time """ - pass + dirs = set() + dir_entries = { + 'file': defaultdict(list), + 'dir': defaultdict(list), + 'rev': defaultdict(list), + } + + for cur_dir in directories: + dir_id = cur_dir['id'] + dirs.add(dir_id) + for entry in cur_dir['entries']: + entry['dir_id'] = dir_id + dir_entries[entry['type']][dir_id].append(entry) + + dirs_missing = set(self.directory_missing(dirs)) + if not dirs_missing: + return + + db = self.db + with db.transaction() as cur: + db.copy_to(({'id': dir} for dir in dirs), 'directory', ['id'], cur) + for entry_type, entry_list in dir_entries.items(): + entries = itertools.chain.from_iterable( + entries_for_dir + for dir_id, entries_for_dir + in entry_list.items() + if dir_id in dirs_missing) + + db.mktemp_dir_entry(entry_type) + + db.copy_to( + entries, + 'tmp_directory_entry_%s' % entry_type, + ['target', 'name', 'perms', 'atime', + 'mtime', 'ctime', 'dir_id'], + cur, + ) + + cur.execute('SELECT swh_directory_entry_%s_add()' % entry_type) @db_transaction def directory_missing(self, directories): """List directories missing from storage Args: an iterable of directory ids Returns: a list of missing directory ids """ (db, cur) = (self.db, self.cur) + # Create temporary table for metadata injection db.mktemp('directory', cur) - db.copy_to(directories, 'tmp_directory', ['id'], cur) + db.copy_to(({'id': dir} for dir in directories), 'tmp_directory', ['id'], cur) for obj in db.directory_missing_from_temp(cur): yield obj[0].tobytes() def revision_add(self, revisions): """Add revisions to the storage Args: revisions: iterable of dictionaries representing the individual revisions to add. Each dict has the following keys: - id (sha1_git): id of the revision to add - date (datetime.DateTime): date the revision was written - commit_date (datetime.DateTime): date the revision got added to the origin - type (one of 'git', 'tar'): type of the revision added - directory (sha1_git): the directory the revision points at - message (bytes): the message associated with the revision - author_name (bytes): the name of the revision author - author_email (bytes): the email of the revision author - committer_name (bytes): the name of the revision committer - committer_email (bytes): the email of the revision committer - parents (list of sha1_git): the parents of this revision """ pass def revision_missing(self, revisions): """List revisions missing from storage Args: an iterable of revision ids Returns: a list of missing revision ids """ pass def release_add(self, releases): """Add releases to the storage Args: releases: iterable of dictionaries representing the individual releases to add. Each dict has the following keys: - id (sha1_git): id of the release to add - revision (sha1_git): id of the revision the release points to - date (datetime.DateTime): the date the release was made - name (bytes): the name of the release - comment (bytes): the comment associated with the release - author_name (bytes): the name of the release author - author_email (bytes): the email of the release author """ pass def release_missing(self, releases): """List releases missing from storage Args: an iterable of release ids Returns: a list of missing release ids """ pass def occurrence_add(self, occurrences): """Add occurrences to the storage Args: occurrences: iterable of dictionaries representing the individual occurrences to add. Each dict has the following keys: - origin (int): id of the origin corresponding to the occurrence - reference (bytes): the reference name of the occurrence - revision (sha1_git): the id of the revision pointed to by the occurrence - date (datetime.DateTime): the validity date for the given occurrence """ pass def origin_add(self, origins): """Add origins to the storage Args: origins: iterable of dictionaries representing the individual origins to add. Each dict has the following keys: - type (FIXME: enum TBD): the origin type ('git', 'wget', ...) - url (bytes): the url the origin points to """ pass diff --git a/swh/storage/tests/test_storage.py b/swh/storage/tests/test_storage.py index bf0850a5..3f6f4198 100644 --- a/swh/storage/tests/test_storage.py +++ b/swh/storage/tests/test_storage.py @@ -1,89 +1,132 @@ # Copyright (C) 2015 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import shutil import tempfile import unittest from nose.tools import istest from nose.plugins.attrib import attr from .db_testing import DbTestFixture from swh.core.hashutil import hex_to_hash from swh.storage import Storage @attr('db') class TestStorage(DbTestFixture, unittest.TestCase): def setUp(self): super().setUp() self.objroot = tempfile.mkdtemp() self.storage = Storage(self.conn, self.objroot) self.cont = { 'data': b'42\n', 'length': 3, 'sha1': hex_to_hash( '34973274ccef6ab4dfaaf86599792fa9c3fe4689'), 'sha1_git': hex_to_hash( 'd81cc0710eb6cf9efd5b920a8453e1e07157b6cd'), 'sha256': hex_to_hash( '673650f936cb3b0a2f93ce09d81be107' '48b1b203c19e8176b4eefc1964a0cf3a'), } self.cont2 = { 'data': b'4242\n', 'length': 5, 'sha1': hex_to_hash( '61c2b3a30496d329e21af70dd2d7e097046d07b7'), 'sha1_git': hex_to_hash( '36fade77193cb6d2bd826161a0979d64c28ab4fa'), 'sha256': hex_to_hash( '859f0b154fdb2d630f45e1ecae4a8629' '15435e663248bb8461d914696fc047cd'), } self.missing_cont = { 'data': b'missing\n', 'length': 8, 'sha1': hex_to_hash( 'f9c24e2abb82063a3ba2c44efd2d3c797f28ac90'), 'sha1_git': hex_to_hash( '33e45d56f88993aae6a0198013efa80716fd8919'), 'sha256': hex_to_hash( '6bbd052ab054ef222c1c87be60cd191a' 'ddedd24cc882d1f5f7f7be61dc61bb3a'), } + self.dir = { + 'id': b'12345678901234567890', + 'entries': [ + { + 'name': 'foo', + 'type': 'file', + 'target': self.cont['sha1_git'], + 'perms': 0o644, + 'atime': None, + 'ctime': None, + 'mtime': None, + }, + { + 'name': 'bar', + 'type': 'dir', + 'target': b'12345678901234567890', + 'perms': 0o2000, + 'atime': None, + 'ctime': None, + 'mtime': None, + }, + ], + } + def tearDown(self): shutil.rmtree(self.objroot) super().tearDown() @istest def content_add(self): cont = self.cont self.storage.content_add([cont]) self.assertIn(cont['sha1'], self.storage.objstorage) self.cursor.execute('SELECT sha1, sha1_git, sha256, length, status' ' FROM content WHERE sha1 = %s', (cont['sha1'],)) datum = self.cursor.fetchone() self.assertEqual( (datum[0].tobytes(), datum[1].tobytes(), datum[2].tobytes(), datum[3], datum[4]), (cont['sha1'], cont['sha1_git'], cont['sha256'], cont['length'], 'visible')) @istest def content_missing(self): cont2 = self.cont2 missing_cont = self.missing_cont self.storage.content_add([cont2]) gen = self.storage.content_missing([cont2, missing_cont]) self.assertEqual(list(gen), [missing_cont['sha1']]) + + @istest + def directory_add(self): + self.storage.directory_add([self.dir]) + + with self.storage.db.transaction() as cur: + cur.execute('SELECT * FROM directory_entry_file') + print(list(cur)) + cur.execute('SELECT * FROM directory_list_file') + print(list(cur)) + cur.execute('SELECT * FROM directory') + print(list(cur)) + cur.execute('SELECT * FROM directory_entry_dir') + print(list(cur)) + cur.execute('SELECT * FROM directory_list_dir') + print(list(cur)) + + self.assertEqual([], + list(self.storage.directory_missing([self.dir['id']])))