diff --git a/sql/swh-func.sql b/sql/swh-func.sql index 7a57be03..4da8fc5c 100644 --- a/sql/swh-func.sql +++ b/sql/swh-func.sql @@ -1,746 +1,659 @@ -- 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_TBLNAME, -- mimicking existing table TBLNAME with an extra dir_id (sha1_git) -- column, and dropping the id column. -- -- This is used to create the tmp_directory_entry_ tables. -- -- Args: -- tblname: name of the table to mimick create or replace function swh_mktemp_dir_entry(tblname regclass) returns void language plpgsql as $$ begin execute format(' create temporary table tmp_%I (like %I including defaults, dir_id sha1_git) on commit drop; alter table tmp_%I drop column id; ', tblname, tblname, tblname, tblname); return; end $$; -- create a temporary table for revisions called tmp_revisions, -- mimicking existing table revision, replacing the foreign keys to -- people with an email and name field -- create or replace function swh_mktemp_revision() returns void language plpgsql as $$ begin create temporary table tmp_revision ( like revision including defaults, author_name text not null default '', author_email text not null default '', committer_name text not null default '', committer_email text not null default '' ) on commit drop; alter table tmp_revision drop column author; alter table tmp_revision drop column committer; return; end $$; -- create a temporary table for releases called tmp_release, -- mimicking existing table release, replacing the foreign keys to -- people with an email and name field -- create or replace function swh_mktemp_release() returns void language plpgsql as $$ begin create temporary table tmp_release ( like release including defaults, author_name text not null default '', author_email text not null default '' ) on commit drop; alter table tmp_release drop column author; return; end $$; -- 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 $$ begin -- This query is critical for (single-algorithm) hash collision detection, -- so we cannot rely only on the fact that a single hash (e.g., sha1) is -- missing from the table content to conclude that a given content is -- missing. Ideally, we would want to (try to) add to content all entries -- in tmp_content that, when considering all columns together, are missing -- from content. -- -- But doing that naively would require a *compound* index on all checksum -- columns; that index would not be significantly smaller than the content -- table itself, and therefore won't be used. Therefore we union together -- all contents that differ on at least one column from what is already -- available. If there is a collision on some (but not all) columns, the -- relevant tmp_content entry will be included in the set of content to be -- added, causing a downstream violation of unicity constraint. return query (select sha1, sha1_git, sha256 from tmp_content as tmp where not exists (select 1 from content as c where c.sha1 = tmp.sha1)) union (select sha1, sha1_git, sha256 from tmp_content as tmp where not exists (select 1 from content as c where c.sha1_git = tmp.sha1_git)) union (select sha1, sha1_git, sha256 from tmp_content as tmp where not exists (select 1 from content as c where c.sha256 = tmp.sha256)); return; end $$; -- check which entries of tmp_skipped_content are missing from skipped_content -- -- operates in bulk: 0. swh_mktemp(skipped_content), 1. COPY to tmp_skipped_content, -- 2. call this function create or replace function swh_skipped_content_missing() returns setof content_signature language plpgsql as $$ begin return query select sha1, sha1_git, sha256 from tmp_skipped_content where not exists (select 1 from skipped_content s where sha1 is not distinct from s.sha1 and sha1_git is not distinct from s.sha1_git and sha256 is not distinct from s.sha256); return; end $$; -- Look up content based on one or several different checksums. Return all -- content information if the content is found; a NULL row otherwise. -- -- At least one checksum should be not NULL. If several are not NULL, they will -- be AND-ed together in the lookup query. -- -- Note: this function is meant to be used to look up individual contents -- (e.g., for the web app), for batch lookup of missing content (e.g., to be -- added) see swh_content_missing create or replace function swh_content_find( sha1 sha1 default NULL, sha1_git sha1_git default NULL, sha256 sha256 default NULL ) returns content language plpgsql as $$ declare con content; filters text[] := array[] :: text[]; -- AND-clauses used to filter content q text; begin if sha1 is not null then filters := filters || format('sha1 = %L', sha1); end if; if sha1_git is not null then filters := filters || format('sha1_git = %L', sha1_git); end if; if sha256 is not null then filters := filters || format('sha256 = %L', sha256); end if; if cardinality(filters) = 0 then return null; else q = format('select * from content where %s', array_to_string(filters, ' and ')); execute q into con; return con; end if; end $$; -- 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 $$ begin insert into content (sha1, sha1_git, sha256, length, status) select distinct sha1, sha1_git, sha256, length, status from tmp_content where (sha1, sha1_git, sha256) in (select * from swh_content_missing()); -- TODO XXX use postgres 9.5 "UPSERT" support here, when available. -- Specifically, using "INSERT .. ON CONFLICT IGNORE" we can avoid -- the extra swh_content_missing() query here. return; end $$; -- add tmp_skipped_content entries to skipped_content, skipping duplicates -- -- operates in bulk: 0. swh_mktemp(skipped_content), 1. COPY to tmp_skipped_content, -- 2. call this function create or replace function swh_skipped_content_add() returns void language plpgsql as $$ begin insert into skipped_content (sha1, sha1_git, sha256, length, status, reason, origin) select distinct sha1, sha1_git, sha256, length, status, reason, origin from tmp_skipped_content where (coalesce(sha1, ''), coalesce(sha1_git, ''), coalesce(sha256, '')) in (select coalesce(sha1, ''), coalesce(sha1_git, ''), coalesce(sha256, '') from swh_skipped_content_missing()); -- TODO XXX use postgres 9.5 "UPSERT" support here, when available. -- Specifically, using "INSERT .. ON CONFLICT IGNORE" we can avoid -- the extra swh_content_missing() query here. return; end $$; -- 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 language plpgsql as $$ begin return query select id from tmp_directory except select id from directory; return; end $$; --- Add tmp_directory_entry_dir entries to directory_entry_dir and --- directory, skipping duplicates in directory_entry_dir. --- --- operates in bulk: 0. swh_mktemp_dir_entry('directory_entry_dir'), 1 COPY to --- tmp_directory_entry_dir, 2. call this function --- --- Assumption: this function is used in the same transaction that inserts the --- context directory in table "directory". --- --- TODO: refactor with other swh_directory_entry_*_add functions -create or replace function swh_directory_entry_dir_add() - returns void - language plpgsql -as $$ -begin - insert into directory_entry_dir (target, name, perms, atime, mtime, ctime) - select distinct t.target, t.name, t.perms, t.atime, t.mtime, t.ctime - from tmp_directory_entry_dir t - where not exists ( - select 1 - from directory_entry_dir i - where t.target = i.target and t.name = i.name and t.perms = i.perms and - t.atime is not distinct from i.atime and - t.mtime is not distinct from i.mtime and - t.ctime is not distinct from i.ctime); - - with new_entries as ( - select t.dir_id, array_agg(i.id) as entries - from tmp_directory_entry_dir t - inner join directory_entry_dir i - on t.target = i.target and t.name = i.name and t.perms = i.perms and - t.atime is not distinct from i.atime and - t.mtime is not distinct from i.mtime and - t.ctime is not distinct from i.ctime - group by t.dir_id - ) - update directory as d - set dir_entries = new_entries.entries - from new_entries - where d.id = new_entries.dir_id; - - return; -end -$$; - --- Add tmp_directory_entry_file entries to directory_entry_file and --- directory, skipping duplicates in directory_entry_file. --- --- operates in bulk: 0. swh_mktemp_dir_entry('directory_entry_file'), 1 COPY to --- tmp_directory_entry_file, 2. call this function --- --- Assumption: this function is used in the same transaction that inserts the --- context directory in table "directory". --- --- TODO: refactor with other swh_directory_entry_*_add functions -create or replace function swh_directory_entry_file_add() - returns void - language plpgsql -as $$ -begin - insert into directory_entry_file (target, name, perms, atime, mtime, ctime) - select distinct t.target, t.name, t.perms, t.atime, t.mtime, t.ctime - from tmp_directory_entry_file t - where not exists ( - select 1 - from directory_entry_file i - where t.target = i.target and t.name = i.name and t.perms = i.perms and - t.atime is not distinct from i.atime and - t.mtime is not distinct from i.mtime and - t.ctime is not distinct from i.ctime); - - with new_entries as ( - select t.dir_id, array_agg(i.id) as entries - from tmp_directory_entry_file t - inner join directory_entry_file i - on t.target = i.target and t.name = i.name and t.perms = i.perms and - t.atime is not distinct from i.atime and - t.mtime is not distinct from i.mtime and - t.ctime is not distinct from i.ctime - group by t.dir_id - ) - update directory as d - set file_entries = new_entries.entries - from new_entries - where d.id = new_entries.dir_id; - - return; -end -$$; +create type directory_entry_type as enum('file', 'dir', 'rev'); --- Add tmp_directory_entry_rev entries to directory_entry_rev and --- directory, skipping duplicates in directory_entry_rev. +-- Add tmp_directory_entry_* entries to directory_entry_* and directory, +-- skipping duplicates in directory_entry_*. This is a generic function that +-- works on all kind of directory entries. -- --- operates in bulk: 0. swh_mktemp_dir_entry('directory_entry_rev'), 1 COPY to --- tmp_directory_entry_rev, 2. call this function +-- operates in bulk: 0. swh_mktemp_dir_entry('directory_entry_*'), 1 COPY to +-- tmp_directory_entry_*, 2. call this function -- -- Assumption: this function is used in the same transaction that inserts the -- context directory in table "directory". --- --- TODO: refactor with other swh_directory_entry_*_add functions -create or replace function swh_directory_entry_rev_add() +create or replace function swh_directory_entry_add(typ directory_entry_type) returns void language plpgsql as $$ begin - insert into directory_entry_rev (target, name, perms, atime, mtime, ctime) + execute format(' + insert into directory_entry_%1$s (target, name, perms, atime, mtime, ctime) select distinct t.target, t.name, t.perms, t.atime, t.mtime, t.ctime - from tmp_directory_entry_rev t + from tmp_directory_entry_%1$s t where not exists ( select 1 - from directory_entry_rev i + from directory_entry_%1$s i where t.target = i.target and t.name = i.name and t.perms = i.perms and t.atime is not distinct from i.atime and t.mtime is not distinct from i.mtime and - t.ctime is not distinct from i.ctime); + t.ctime is not distinct from i.ctime) + ', typ); + execute format(' with new_entries as ( select t.dir_id, array_agg(i.id) as entries - from tmp_directory_entry_rev t - inner join directory_entry_rev i + from tmp_directory_entry_%1$s t + inner join directory_entry_%1$s i on t.target = i.target and t.name = i.name and t.perms = i.perms and t.atime is not distinct from i.atime and t.mtime is not distinct from i.mtime and t.ctime is not distinct from i.ctime group by t.dir_id ) update directory as d - set rev_entries = new_entries.entries + set %1$s_entries = new_entries.entries from new_entries - where d.id = new_entries.dir_id; + where d.id = new_entries.dir_id + ', typ); return; end $$; -create type directory_entry_type as enum('file', 'dir', 'rev'); - -- a directory listing entry with all the metadata -- -- can be used to list a directory, and retrieve all the data in one go. create type directory_entry as ( dir_id sha1_git, -- id of the parent directory type directory_entry_type, -- type of entry target sha1_git, -- id of target 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 ); -- List a single level of directory walked_dir_id create or replace function swh_directory_walk_one(walked_dir_id sha1_git) returns setof directory_entry language plpgsql as $$ begin return query with dir as ( select id as dir_id, dir_entries, file_entries, rev_entries from directory where id = walked_dir_id), ls_d as (select dir_id, unnest(dir_entries) as entry_id from dir), ls_f as (select dir_id, unnest(file_entries) as entry_id from dir), ls_r as (select dir_id, unnest(rev_entries) as entry_id from dir) (select dir_id, 'dir'::directory_entry_type as type, target, name, perms, atime, mtime, ctime from ls_d left join directory_entry_dir d on ls_d.entry_id = d.id) union (select dir_id, 'file'::directory_entry_type as type, target, name, perms, atime, mtime, ctime from ls_f left join directory_entry_file d on ls_f.entry_id = d.id) union (select dir_id, 'rev'::directory_entry_type as type, target, name, perms, atime, mtime, ctime from ls_r left join directory_entry_rev d on ls_r.entry_id = d.id) order by name; return; end $$; -- List all revision IDs starting from a given revision, going back in time -- -- TODO ordering: should be breadth-first right now (what do we want?) -- TODO ordering: ORDER BY parent_rank somewhere? create or replace function swh_revision_list(root_revision sha1_git) returns setof sha1_git language plpgsql as $$ begin return query with recursive rev_list(id) as ( (select id from revision where id = root_revision) union (select parent_id from revision_history as h join rev_list on h.id = rev_list.id) ) select * from rev_list; return; end $$; -- Detailed entry in a revision log create type revision_log_entry as ( id sha1_git, date timestamptz, date_offset smallint, committer_date timestamptz, committer_date_offset smallint, type revision_type, directory sha1_git, message bytea, author_name text, author_email text, committer_name text, committer_email text ); -- "git style" revision log. Similar to swh_revision_list(), but returning all -- information associated to each revision, and expanding authors/committers create or replace function swh_revision_log(root_revision sha1_git) returns setof revision_log_entry language plpgsql as $$ begin return query select revision.id, date, date_offset, committer_date, committer_date_offset, type, directory, message, author.name as author_name, author.email as author_email, committer.name as committer_name, committer.email as committer_email from swh_revision_list(root_revision) as rev_list join revision on revision.id = rev_list join person as author on revision.author = author.id join person as committer on revision.committer = committer.id; return; end $$; -- List missing revisions from tmp_revision create or replace function swh_revision_missing() returns setof sha1_git language plpgsql as $$ begin return query select id from tmp_revision except select id from revision; return; end $$; -- Create entries in person from tmp_revision create or replace function swh_person_add_from_revision() returns void language plpgsql as $$ begin with t as ( select author_name as name, author_email as email from tmp_revision union select committer_name as name, committer_email as email from tmp_revision ) insert into person (name, email) select distinct name, email from t where not exists ( select 1 from person p where t.name = p.name and t.email = p.email ); return; end $$; -- Create entries in revision from tmp_revision create or replace function swh_revision_add() returns void language plpgsql as $$ begin perform swh_person_add_from_revision(); insert into revision (id, date, date_offset, committer_date, committer_date_offset, type, directory, message, author, committer) select t.id, t.date, t.date_offset, t.committer_date, t.committer_date_offset, t.type, t.directory, t.message, a.id, c.id from tmp_revision t left join person a on a.name = t.author_name and a.email = t.author_email left join person c on c.name = t.committer_name and c.email = t.committer_email; return; end $$; -- List missing releases from tmp_release create or replace function swh_release_missing() returns setof sha1_git language plpgsql as $$ begin return query select id from tmp_release except select id from release; return; end $$; -- Create entries in person from tmp_release create or replace function swh_person_add_from_release() returns void language plpgsql as $$ begin with t as ( select distinct author_name as name, author_email as email from tmp_release ) insert into person (name, email) select name, email from t where not exists ( select 1 from person p where t.name = p.name and t.email = p.email ); return; end $$; -- Create entries in release from tmp_release create or replace function swh_release_add() returns void language plpgsql as $$ begin perform swh_person_add_from_release(); insert into release (id, revision, date, date_offset, name, comment, author) select t.id, t.revision, t.date, t.date_offset, t.name, t.comment, a.id from tmp_release t left join person a on a.name = t.author_name and a.email = t.author_email; return; end $$; -- Absolute path: directory reference + complete path relative to it create type content_dir as ( directory sha1_git, path unix_path ); -- Find the containing directory of a given content, specified by sha1 -- (note: *not* sha1_git). -- -- Return a pair (dir_it, path) where path is a UNIX path that, from the -- directory root, reach down to a file with the desired content. -- -- In case of multiple paths (i.e., pretty much always), an arbitrary one is -- chosen. create or replace function swh_content_find_directory(content_id sha1) returns content_dir language plpgsql as $$ declare d content_dir; begin with recursive path as ( -- Recursively build a path from the requested content to a root -- directory. Each iteration returns a pair (dir_id, filename) where -- filename is relative to dir_id. Stops when no parent directory can -- be found. (select dir.id as dir_id, dir_entry_f.name as name, 0 as depth from directory_entry_file as dir_entry_f join content on content.sha1_git = dir_entry_f.target join directory as dir on dir.file_entries @> array[dir_entry_f.id] where content.sha1 = content_id limit 1) union all (select dir.id as dir_id, (dir_entry_d.name || '/' || path.name)::unix_path as name, path.depth + 1 from path join directory_entry_dir as dir_entry_d on dir_entry_d.target = path.dir_id join directory as dir on dir.dir_entries @> array[dir_entry_d.id] limit 1) ) select dir_id, name from path order by depth desc limit 1 into strict d; return d; end $$; -- Walk the revision history starting from a given revision, until a matching -- occurrence is found. Return all occurrence information. create or replace function swh_revision_find_occurrence(revision_id sha1_git) returns occurrence language plpgsql as $$ declare occ occurrence%ROWTYPE; rev sha1_git; begin -- first check to see if revision_id is already pointed by an occurrence select origin, branch, revision from occurrence_history as occ_hist where occ_hist.revision = revision_id order by upper(occ_hist.validity) -- TODO filter by authority? limit 1 into occ; -- no occurrence point to revision_id, walk up the history if not found then -- recursively walk the history, stopping immediately before a revision -- pointed to by an occurrence. -- TODO find a nicer way to stop at, but *including*, that revision with recursive revlog as ( (select revision_id as rev_id, 0 as depth) union all (select hist.parent_id as rev_id, revlog.depth + 1 from revlog join revision_history as hist on hist.id = revlog.rev_id and not exists(select 1 from occurrence_history where revision = hist.parent_id) limit 1) ) select rev_id from revlog order by depth desc limit 1 into strict rev; -- as we stopped before a pointed by revision, look it up again and -- return its data select origin, branch, revision from revision_history as rev_hist, occurrence_history as occ_hist where rev_hist.id = rev and occ_hist.revision = rev_hist.parent_id order by upper(occ_hist.validity) -- TODO filter by authority? limit 1 into strict occ; -- will fail if no occurrence is found, and that's OK end if; return occ; end $$; -- Occurrence of some content in a given context create type content_occurrence as ( origin_type text, origin_url text, branch text, revision_id sha1_git, path unix_path ); -- Given the sha1 of some content, look up an occurrence that points to a -- revision, which in turns reference (transitively) a tree containing the -- content. Answer the question: "where/when did SWH see a given content"? -- Return information about an arbitrary occurrence/revision/tree, with no -- ordering guarantee whatsoever. create or replace function swh_content_find_occurrence(content_id sha1) returns content_occurrence language plpgsql as $$ declare dir content_dir; rev sha1_git; occ occurrence%ROWTYPE; coc content_occurrence; begin -- each step could fail if no results are found, and that's OK select * from swh_content_find_directory(content_id) -- look up directory into strict dir; select id from revision where directory = dir.directory -- look up revision limit 1 into strict rev; select * from swh_revision_find_occurrence(rev) -- look up occurrence into strict occ; select origin.type, origin.url, occ.branch, rev, dir.path from origin where origin.id = occ.origin into strict coc; return coc; end $$; diff --git a/swh/storage/storage.py b/swh/storage/storage.py index 608d03cd..1f770275 100644 --- a/swh/storage/storage.py +++ b/swh/storage/storage.py @@ -1,580 +1,581 @@ # 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 from swh.core.hashutil import ALGORITHMS SWH_HASH_KEYS = ALGORITHMS 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) 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 (default: -1) - one key for each checksum algorithm in swh.core.hashutil.ALGORITHMS, mapped to the corresponding checksum - status (str): one of visible, hidden, absent - reason (str): if status = absent, the reason why - origin (int): if status = absent, the origin we saw the content in """ db = self.db content_by_status = defaultdict(list) for d in content: if 'status' not in d: d['status'] = 'visible' if 'length' not in d: d['length'] = -1 content_by_status[d['status']].append(d) content_with_data = content_by_status['visible'] content_without_data = content_by_status['absent'] missing_content = set(self.content_missing(content_with_data)) missing_skipped = set( sha1_git for sha1, sha1_git, sha256 in self.skipped_content_missing(content_without_data)) with db.transaction() as cur: if missing_content: # 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']) content_filtered = (cont for cont in content_with_data if cont['sha1'] in missing_content) db.copy_to(content_filtered, 'tmp_content', ['sha1', 'sha1_git', 'sha256', 'length', 'status'], cur, item_cb=add_to_objstorage) # move metadata in place db.content_add_from_temp(cur) if missing_skipped: missing_filtered = (cont for cont in content_without_data if cont['sha1_git'] in missing_skipped) db.mktemp('skipped_content', cur) db.copy_to(missing_filtered, 'tmp_skipped_content', ['sha1', 'sha1_git', 'sha256', 'length', 'reason', 'status', 'origin'], cur) # move metadata in place db.skipped_content_add_from_temp(cur) @db_transaction_generator def content_missing(self, content, key_hash='sha1', cur=None): """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. key_hash: the name of the hash used as key (default: 'sha1') Returns: an iterable of `key_hash`es missing from the storage Raises: TODO: an exception when we get a hash collision. """ db = self.db keys = ['sha1', 'sha1_git', 'sha256'] if key_hash not in keys: raise ValueError("key_hash should be one of %s" % keys) key_hash_idx = keys.index(key_hash) # Create temporary table for metadata injection db.mktemp('content', cur) db.copy_to(content, 'tmp_content', keys + ['length'], cur) for obj in db.content_missing_from_temp(cur): yield obj[key_hash_idx] @db_transaction_generator def skipped_content_missing(self, content, cur=None): """List skipped_content missing from storage Args: content: iterable of dictionaries containing the data for each checksum algorithm. Returns: an iterable of signatures missing from the storage """ keys = ['sha1', 'sha1_git', 'sha256'] db = self.db db.mktemp('skipped_content', cur) db.copy_to(content, 'tmp_skipped_content', keys + ['length', 'reason'], cur) yield from db.skipped_content_missing_from_temp(cur) @db_transaction def content_find(self, content, cur=None): """Find a content hash in db. Args: content: a dictionary entry representing one content hash. The dictionary key is one of swh.core.hashutil.ALGORITHMS. The value mapped to the corresponding checksum. Returns: a triplet (sha1, sha1_git, sha256) if the content exist or None otherwise. Raises: ValueError in case the key of the dictionary is not sha1, sha1_git nor sha256. """ db = self.db if content == {}: raise ValueError('Key must be one of sha1, git_sha1, sha256.') for key in content.keys(): if key not in SWH_HASH_KEYS: raise ValueError('Key must be one of sha1, git_sha1, sha256.') # format the output found_hash = db.content_find(sha1=content.get('sha1'), sha1_git=content.get('sha1_git'), sha256=content.get('sha256'), cur=cur) return found_hash @db_transaction def content_exist(self, content, cur=None): """Predicate to check the presence of a content's hashes. Args: content: a dictionary entry representing one content hash. The dictionary key is one of swh.core.hashutil.ALGORITHMS. The value mapped to the corresponding checksum. Returns: a boolean indicator of presence Raises: ValueError in case the key of the dictionary is not sha1, sha1_git nor sha256. """ return self.content_find(content) is not None @db_transaction def content_find_occurrence(self, content, cur=None): """Find the content's occurrence. Args: content: a dictionary entry representing one content hash. The dictionary key is one of swh.core.hashutil.ALGORITHMS. The value mapped to the corresponding checksum. Returns: The occurrence of the content. Raises: ValueError in case the key of the dictionary is not sha1, sha1_git nor sha256. """ db = self.db c = self.content_find(content) if c is None: return None sha1, _, _ = c found_occ = db.content_find_occurrence(sha1, cur=cur) if found_occ is None: return None return {'origin_type': found_occ[0], 'origin_url': found_occ[1], 'branch': found_occ[2], 'revision': found_occ[3], 'path': found_occ[4]} 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) + cur.execute('SELECT swh_directory_entry_add(%s)', + (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())) if not revisions_missing: return 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', '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 - date_offset (int): offset from UTC in minutes 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 """ db = self.db release_ids = set(release['id'] for release in releases) releases_missing = list(self.release_missing(release_ids)) if not releases_missing: return with db.transaction() as cur: db.mktemp_release(cur) releases_filtered = (release for release in releases if release['id'] in releases_missing) db.copy_to(releases_filtered, 'tmp_release', ['id', 'revision', 'date', 'date_offset', 'name', 'comment', 'author_name', 'author_email'], cur) db.release_add_from_temp(cur) @db_transaction_generator def release_missing(self, releases, cur=None): """List releases missing from storage Args: an iterable of release ids Returns: a list of missing release ids """ db = self.db # Create temporary table for metadata injection db.mktemp('release', cur) releases_dicts = ({'id': rel} for rel in releases) db.copy_to(releases_dicts, 'tmp_release', ['id'], cur) for obj in db.release_missing_from_temp(cur): yield obj[0] @db_transaction def occurrence_add(self, occurrences, cur=None): """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 - branch (str): the reference name of the occurrence - revision (sha1_git): the id of the revision pointed to by the occurrence - authority (int): id of the authority giving the validity - validity (datetime.DateTime): the validity date for the given occurrence """ db = self.db processed = [] for occurrence in occurrences: occ = occurrence.copy() occ['validity'] = '[%s,infinity)' % str(occ['validity']) processed.append(occ) # XXX: will fail on second execution db.copy_to(processed, 'occurrence_history', ['origin', 'branch', 'revision', 'authority', 'validity'], cur) @db_transaction def origin_get(self, origin, cur=None): """Return the id of the given origin Args: origin: dictionary representing the individual origin to find. This dict has the following keys: - type (FIXME: enum TBD): the origin type ('git', 'wget', ...) - url (bytes): the url the origin points to Returns: the id of the queried origin """ query = "select id from origin where type=%s and url=%s" cur.execute(query, (origin['type'], origin['url'])) data = cur.fetchone() if not data: return None else: return data[0] @db_transaction def origin_add_one(self, origin, cur=None): """Add origin to the storage Args: origin: dictionary representing the individual origin to add. This dict has the following keys: - type (FIXME: enum TBD): the origin type ('git', 'wget', ...) - url (bytes): the url the origin points to Returns: the id of the added origin, or of the identical one that already exists. """ query = "select id from origin where type=%s and url=%s" cur.execute(query, (origin['type'], origin['url'])) data = cur.fetchone() if data: return data[0] insert = """insert into origin (type, url) values (%s, %s) returning id""" cur.execute(insert, (origin['type'], origin['url'])) return cur.fetchone()[0]