diff --git a/sql/swh-schema.sql b/sql/swh-schema.sql index 66a2cc21..fec26f24 100644 --- a/sql/swh-schema.sql +++ b/sql/swh-schema.sql @@ -1,325 +1,328 @@ --- --- Software Heritage Data Model --- -- drop schema if exists swh cascade; -- create schema swh; -- set search_path to swh; create table dbversion ( version int primary key, release timestamptz, description text ); insert into dbversion(version, release, description) values(13, now(), 'Work In Progress'); -- a SHA1 checksum (not necessarily originating from Git) create domain sha1 as bytea check (length(value) = 20); -- a Git object ID, i.e., a SHA1 checksum create domain sha1_git as bytea check (length(value) = 20); -- a SHA256 checksum create domain sha256 as bytea check (length(value) = 32); -- UNIX path (absolute, relative, individual path component, etc.) -- TODO should this be bytea or similar to avoid encoding/decoding issues? create domain unix_path as text; -- a set of UNIX-like access permissions, as manipulated by, e.g., chmod create domain file_perms as int; create type content_status as enum ('absent', 'visible', 'hidden'); -- Checksums about actual file content. Note that the content itself is not -- stored in the DB, but on external (key-value) storage. A single checksum is -- used as key there, but the other can be used to verify that we do not inject -- content collisions not knowingly. create table content ( sha1 sha1 primary key, sha1_git sha1_git not null, sha256 sha256 not null, length bigint not null, ctime timestamptz not null default now(), -- creation time, i.e. time of (first) injection into the storage status content_status not null default 'visible' ); create unique index on content(sha1_git); -- create unique index on content(sha256); -- An organization (or part thereof) that might be in charge of running -- software projects. Examples: Debian, GNU, GitHub, Apache, The Linux -- Foundation. The data model is hierarchical (via parent_id) and might store -- sub-branches of existing organizations. The key feature of an organization -- is that it can be *listed* to retrieve information about its content, i.e: -- sub-organizations, projects, origins. create table organization ( id bigserial primary key, parent_id bigint references organization(id), name text not null, description text, homepage text, list_engine text, -- crawler to be used to org's content list_url text, -- root URL to start the listing list_params json, -- org-specific listing parameter latest_list timestamptz -- last time the org's content has been listed ); -- Log of all organization listings (i.e., organization crawling) that have -- been done in the past, or are still ongoing. Similar to fetch_history, but -- for organizations. create table list_history ( id bigserial primary key, organization bigint references organization(id), date timestamptz not null, status boolean, -- true if and only if the listing has been successful result json, -- more detailed return value, depending on status stdout text, stderr text, duration interval -- fetch duration of NULL if still ongoing ); -- An origin is a place, identified by an URL, where software can be found. We -- support different kinds of origins, e.g., git and other VCS repositories, -- web pages that list tarballs URLs (e.g., http://www.kernel.org), indirect -- tarball URLs (e.g., http://www.example.org/latest.tar.gz), etc. The key -- feature of an origin is that it can be *fetched* (wget, git clone, svn -- checkout, etc.) to retrieve all the contained software. create table origin ( id bigserial primary key, type text, -- TODO use an enum here (?) url text not null ); -- Log of all origin fetches (i.e., origin crawling) that have been done in the -- past, or are still ongoing. Similar to list_history, but for origins. create table fetch_history ( id bigserial primary key, origin bigint references origin(id), date timestamptz not null, status boolean, -- true if and only if the fetch has been successful result json, -- more detailed returned values, times, etc... stdout text, stderr text, -- null when status is true, filled otherwise duration interval -- fetch duration of NULL if still ongoing ); -- A specific software project, e.g., the Linux kernel, Apache httpd. A -- software project is version-less at this level, but is associated to several -- metadata. Metadata can evolve over time, this table only contains the most -- recent version of them; for old versions of project see table -- project_history. create table project ( id bigserial primary key, organization bigint references organization(id), -- the "owning" organization origin bigint references origin(id), -- where to find project releases name text, description text, homepage text, doap jsonb -- other kinds of metadata/software project description ontologies can be -- added here, in addition to DOAP ); -- History of project metadata. Time-sensitive version of the table project. create table project_history ( id bigserial primary key, project bigint references project(id), validity tstzrange, organization bigint references organization(id), origin bigint references origin(id), name text, description text, homepage text, doap jsonb ); -- A file-system directory. A directory is a list of directory entries (see -- tables: directory_entry_{dir,file}). -- -- To list the contents of a directory: -- 1. list the contained directory_entry_dir using table directory_list_dir -- 2. list the contained directory_entry_file using table directory_list_file -- 3. list the contained directory_entry_rev using table directory_list_rev -- 4. UNION -- -- Synonyms/mappings: -- * git: tree create table directory ( id sha1_git primary key ); -- A directory entry pointing to a sub-directory. create table directory_entry_dir ( id bigserial primary key, target sha1_git, -- references directory(id) deferrable initially deferred, -- id of target directory name unix_path, -- path name, relative to containing dir perms file_perms, -- unix-like permissions atime timestamptz, -- time of last access mtime timestamptz, -- time of last modification ctime timestamptz -- time of last status change ); create unique index on directory_entry_dir(target, name, perms, atime, mtime, ctime); -- Mapping between directories and contained sub-directories. create table directory_list_dir ( dir_id sha1_git references directory(id), entry_id bigint references directory_entry_dir(id), primary key (dir_id, entry_id) ); -- A directory entry pointing to a file. create table directory_entry_file ( id bigserial primary key, target sha1_git, -- id of target file name unix_path, -- path name, relative to containing dir perms file_perms, -- unix-like permissions atime timestamptz, -- time of last access mtime timestamptz, -- time of last modification ctime timestamptz -- time of last status change ); create unique index on directory_entry_file(target, name, perms, atime, mtime, ctime); -- Mapping between directories and contained files. create table directory_list_file ( dir_id sha1_git references directory(id), entry_id bigint references directory_entry_file(id), primary key (dir_id, entry_id) ); -- A directory entry pointing to a revision. create table directory_entry_rev ( id bigserial primary key, target sha1_git, -- id of target revision name unix_path, -- path name, relative to containing dir perms file_perms, -- unix-like permissions atime timestamptz, -- time of last access mtime timestamptz, -- time of last modification ctime timestamptz -- time of last status change ); create unique index on directory_entry_rev(target, name, perms, atime, mtime, ctime); -- Mapping between directories and contained files. create table directory_list_rev ( dir_id sha1_git references directory(id), entry_id bigint references directory_entry_rev(id), primary key (dir_id, entry_id) ); create table person ( id bigserial primary key, name text not null default '', email text not null default '' ); create unique index on person(name, email); create type revision_type as enum ('git', 'tar', 'dsc'); -- A snapshot of a software project at a specific point in time. -- -- Synonyms/mappings: -- * git / subversion / etc: commit -- * tarball: a specific tarball -- -- Revisions are organized as DAGs. Each revision points to 0, 1, or more (in -- case of merges) parent revisions. Each revision points to a directory, i.e., -- a file-system tree containing files and directories. create table revision ( - id sha1_git primary key, - date timestamptz, - committer_date timestamptz, - type revision_type not null, - directory sha1_git, -- file-system tree - message text, - author bigint references person(id), - committer bigint references person(id) + id sha1_git primary key, + date timestamptz, + date_offset smallint, + committer_date timestamptz, + committer_date_offset smallint, + type revision_type not null, + directory sha1_git, -- file-system tree + message text, + author bigint references person(id), + committer bigint references person(id) ); -- either this table or the sha1_git[] column on the revision table create table revision_history ( id sha1_git references revision(id), parent_id sha1_git, parent_rank int not null default 0, -- parent position in merge commits, 0-based primary key (id, parent_id), unique (id, parent_rank) ); -- The content of software origins is indexed starting from top-level pointers -- called "branches". Every time we fetch some origin we store in this table -- where the branches pointed to at fetch time. -- -- Synonyms/mappings: -- * git: ref (in the "git update-ref" sense) create table occurrence_history ( origin bigint references origin(id), branch text, -- e.g., "master" (for VCS), or "sid" (for Debian) revision sha1_git references revision(id), -- ref target, e.g., commit id authority bigint references organization(id) not null, -- who is claiming to have seen the occurrence. -- Note: SWH is such an authority, and has an entry in -- the organization table. validity tstzrange, -- The time validity of this table entry. If the upper -- bound is missing, the entry is still valid. exclude using gist (origin with =, branch with =, revision with =, authority with =, validity with &&), -- unicity exclusion constraint on lines where the same value is found for -- `origin`, `reference`, `revision`, `authority` and overlapping values for -- `validity`. primary key (origin, branch, revision, authority, validity) ); -- Materialized view of occurrence_history, storing the *current* value of each -- branch, as last seen by SWH. create table occurrence ( origin bigint references origin(id), branch text, revision sha1_git references revision(id), primary key(origin, branch, revision) ); -- A "memorable" point in the development history of a project. -- -- Synonyms/mappings: -- * git: tag (of the annotated kind, otherwise they are just references) -- * tarball: the release version number create table release ( - id sha1_git primary key, - revision sha1_git references revision(id), - date timestamptz, - name text, - comment text, - author bigint references person(id) + id sha1_git primary key, + revision sha1_git references revision(id), + date timestamptz, + date_offset smallint, + name text, + comment text, + author bigint references person(id) ); diff --git a/swh/storage/storage.py b/swh/storage/storage.py index 4052cfdd..4390f623 100644 --- a/swh/storage/storage.py +++ b/swh/storage/storage.py @@ -1,331 +1,336 @@ # 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 The decorated method must accept a `cur` keyword argument """ @functools.wraps(meth) def _meth(self, *args, **kwargs): with self.db.transaction() as cur: return meth(self, *args, cur=cur, **kwargs) return _meth def db_transaction_generator(meth): """decorator to execute Storage methods within DB transactions, while returning a generator The decorated method must accept a `cur` keyword argument """ @functools.wraps(meth) def _meth(self, *args, **kwargs): with self.db.transaction() as cur: yield from meth(self, *args, cur=cur, **kwargs) 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, cur): """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 = self.db # 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_generator def content_missing(self, content, cur): """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 = self.db # 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] 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) - 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 """ 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: dirs_missing_dict = ({'id': dir} for dir in dirs_missing) db.copy_to(dirs_missing_dict, '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_generator def directory_missing(self, directories, cur): """List directories missing from storage Args: an iterable of directory ids Returns: a list of missing directory ids """ db = self.db # Create temporary table for metadata injection db.mktemp('directory', cur) directories_dicts = ({'id': dir} for dir in directories) db.copy_to(directories_dicts, 'tmp_directory', ['id'], cur) for obj in db.directory_missing_from_temp(cur): yield obj[0] def directory_get(self, directory): """Get the entries for one directory""" yield from self.db.directory_walk_one(directory) 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 + - date_offset (int): offset from UTC in minutes the revision + was written - committer_date (datetime.DateTime): date the revision got added to the origin + - committer_date_offset (int): offset from UTC in minutes the + revision was 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 """ db = self.db parents = {} for revision in revisions: id = revision['id'] cur_parents = enumerate(revision.get('parents', [])) parents[id] = [{ 'id': id, 'parent_id': parent, 'parent_rank': i } for i, parent in cur_parents] revisions_missing = list(self.revision_missing(parents.keys())) with db.transaction() as cur: db.mktemp_revision(cur) revisions_filtered = (revision for revision in revisions if revision['id'] in revisions_missing) db.copy_to(revisions_filtered, 'tmp_revision', - ['id', 'date', 'committer_date', 'type', 'directory', + ['id', 'date', 'date_offset', 'committer_date', + 'committer_date_offset', 'type', 'directory', 'message', 'author_name', 'author_email', 'committer_name', 'committer_email'], cur) db.revision_add_from_temp(cur) parents_filtered = itertools.chain.from_iterable( parents[id] for id in revisions_missing) db.copy_to(parents_filtered, 'revision_history', ['id', 'parent_id', 'parent_rank'], cur) @db_transaction_generator def revision_missing(self, revisions, cur): """List revisions missing from storage Args: an iterable of revision ids Returns: a list of missing revision ids """ db = self.db # Create temporary table for metadata injection db.mktemp('revision', cur) revisions_dicts = ({'id': dir, 'type': 'git'} for dir in revisions) db.copy_to(revisions_dicts, 'tmp_revision', ['id', 'type'], cur) for obj in db.revision_missing_from_temp(cur): yield obj[0] 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 223234e1..945577b5 100644 --- a/swh/storage/tests/test_storage.py +++ b/swh/storage/tests/test_storage.py @@ -1,158 +1,160 @@ # 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 datetime 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, }, ], } self.revision = { 'id': b'56789012345678901234', 'message': 'hello', 'author_name': 'Nicolas Dandrimont', 'author_email': 'nicolas@example.com', 'committer_name': 'Stefano Zacchiroli', 'committer_email': 'stefano@example.com', 'parents': [b'01234567890123456789'], 'date': datetime.datetime(2015, 1, 1, 22, 0, 0), + 'date_offset': 120, 'committer_date': datetime.datetime(2015, 1, 2, 22, 0, 0), + 'committer_date_offset': -120, 'type': 'git', 'directory': self.dir['id'], } 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): init_missing = list(self.storage.directory_missing([self.dir['id']])) self.assertEqual([self.dir['id']], init_missing) self.storage.directory_add([self.dir]) stored_data = list(self.storage.directory_get(self.dir['id'])) data_to_store = [ (self.dir['id'], ent['type'], ent['target'], ent['name'], ent['perms'], ent['atime'], ent['ctime'], ent['mtime']) for ent in sorted(self.dir['entries'], key=lambda ent: ent['name']) ] self.assertEqual(data_to_store, stored_data) after_missing = list(self.storage.directory_missing([self.dir['id']])) self.assertEqual([], after_missing) @istest def revision_add(self): init_missing = self.storage.revision_missing([self.revision['id']]) self.assertEqual([self.revision['id']], list(init_missing)) self.storage.revision_add([self.revision]) end_missing = self.storage.revision_missing([self.revision['id']]) self.assertEqual([], list(end_missing))