diff --git a/PKG-INFO b/PKG-INFO index bf733ef5..ab1ab588 100644 --- a/PKG-INFO +++ b/PKG-INFO @@ -1,10 +1,10 @@ Metadata-Version: 1.0 Name: swh.storage -Version: 0.0.7 +Version: 0.0.8 Summary: Software Heritage storage manager Home-page: https://forge.softwareheritage.org/diffusion/DSTO/ Author: Software Heritage developers Author-email: swh-devel@inria.fr License: UNKNOWN Description: UNKNOWN Platform: UNKNOWN diff --git a/debian/control b/debian/control index 2505ebbf..902a9b5d 100644 --- a/debian/control +++ b/debian/control @@ -1,21 +1,21 @@ Source: swh-storage Maintainer: Software Heritage developers Section: python Priority: optional Build-Depends: debhelper (>= 9), dh-python, python3-all, python3-flask, python3-nose, python3-psycopg2, python3-requests, python3-setuptools, - python3-swh.core (>= 0.0.6~), + python3-swh.core (>= 0.0.7~), python3-vcversioner Standards-Version: 3.9.6 Homepage: https://forge.softwareheritage.org/diffusion/DCORE/ Package: python3-swh.storage Architecture: all -Depends: python3-swh.core (>= 0.0.6~), ${misc:Depends}, ${python3:Depends} +Depends: python3-swh.core (>= 0.0.7~), ${misc:Depends}, ${python3:Depends} Description: Software Heritage storage utilities diff --git a/requirements.txt b/requirements.txt index eb8a5be8..cb0c28ab 100644 --- a/requirements.txt +++ b/requirements.txt @@ -1,11 +1,11 @@ psycopg2 vcversioner # remote storage API client requests # remote storage API server flask # Internal dependencies -swh.core >= 0.0.6 +swh.core >= 0.0.7 diff --git a/sql/swh-func.sql b/sql/swh-func.sql index 1bc1708f..2e70ba78 100644 --- a/sql/swh-func.sql +++ b/sql/swh-func.sql @@ -1,724 +1,664 @@ -- 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 + language sql 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 + language sql 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 $$ begin + -- This query is critical for (single-algorithm) hash collision detection, + -- so we cannot rely only on the fact that a single hash (e.g., sha1) is + -- missing from the table content to conclude that a given content is + -- missing. Ideally, we would want to (try to) add to content all entries + -- in tmp_content that, when considering all columns together, are missing + -- from content. + -- + -- But doing that naively would require a *compound* index on all checksum + -- columns; that index would not be significantly smaller than the content + -- table itself, and therefore won't be used. Therefore we union together + -- all contents that differ on at least one column from what is already + -- available. If there is a collision on some (but not all) columns, the + -- relevant tmp_content entry will be included in the set of content to be + -- added, causing a downstream violation of unicity constraint. return query - select sha1, sha1_git, sha256 from tmp_content - except - select sha1, sha1_git, sha256 from content; + (select sha1, sha1_git, sha256 from tmp_content as tmp + where not exists + (select 1 from content as c where c.sha1 = tmp.sha1)) + union + (select sha1, sha1_git, sha256 from tmp_content as tmp + where not exists + (select 1 from content as c where c.sha1_git = tmp.sha1_git)) + union + (select sha1, sha1_git, sha256 from tmp_content as tmp + where not exists + (select 1 from content as c where c.sha256 = tmp.sha256)); return; end $$; -- 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 $$ -begin - insert into directory_entry_file (target, name, perms, atime, mtime, ctime) - select distinct t.target, t.name, t.perms, t.atime, t.mtime, t.ctime - from tmp_directory_entry_file t - where not exists ( - select 1 - from directory_entry_file i - where t.target = i.target and t.name = i.name and t.perms = i.perms and - t.atime is not distinct from i.atime and - t.mtime is not distinct from i.mtime and - t.ctime is not distinct from i.ctime); - with new_entries as ( - select t.dir_id, array_agg(i.id) as entries - from tmp_directory_entry_file t - inner join directory_entry_file i - on t.target = i.target and t.name = i.name and t.perms = i.perms and - t.atime is not distinct from i.atime and - t.mtime is not distinct from i.mtime and - t.ctime is not distinct from i.ctime - group by t.dir_id - ) - update directory as d - set file_entries = new_entries.entries - from new_entries - where d.id = new_entries.dir_id; +create type directory_entry_type as enum('file', 'dir', 'rev'); - return; -end -$$; --- 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 + language sql as $$ -begin - return query - with dir as ( - select id as dir_id, dir_entries, file_entries, rev_entries - from directory - where id = walked_dir_id), - ls_d as (select dir_id, unnest(dir_entries) as entry_id from dir), - ls_f as (select dir_id, unnest(file_entries) as entry_id from dir), - ls_r as (select dir_id, unnest(rev_entries) as entry_id from dir) - (select dir_id, 'dir'::directory_entry_type as type, - target, name, perms, atime, mtime, ctime - from ls_d - left join directory_entry_dir d on ls_d.entry_id = d.id) - union - (select dir_id, 'file'::directory_entry_type as type, - target, name, perms, atime, mtime, ctime - from ls_f - left join directory_entry_file d on ls_f.entry_id = d.id) - union - (select dir_id, 'rev'::directory_entry_type as type, - target, name, perms, atime, mtime, ctime - from ls_r - left join directory_entry_rev d on ls_r.entry_id = d.id) - order by name; - return; -end + with dir as ( + select id as dir_id, dir_entries, file_entries, rev_entries + from directory + where id = walked_dir_id), + ls_d as (select dir_id, unnest(dir_entries) as entry_id from dir), + ls_f as (select dir_id, unnest(file_entries) as entry_id from dir), + ls_r as (select dir_id, unnest(rev_entries) as entry_id from dir) + (select dir_id, 'dir'::directory_entry_type as type, + target, name, perms, atime, mtime, ctime + from ls_d + left join directory_entry_dir d on ls_d.entry_id = d.id) + union + (select dir_id, 'file'::directory_entry_type as type, + target, name, perms, atime, mtime, ctime + from ls_f + left join directory_entry_file d on ls_f.entry_id = d.id) + union + (select dir_id, 'rev'::directory_entry_type as type, + target, name, perms, atime, mtime, ctime + from ls_r + left join directory_entry_rev d on ls_r.entry_id = d.id) + order by name; $$; + -- 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 + language sql as $$ -begin - return query - with recursive rev_list(id) as ( - (select id from revision where id = root_revision) - union - (select parent_id - from revision_history as h - join rev_list on h.id = rev_list.id) - ) - select * from rev_list; - return; -end + with recursive rev_list(id) as ( + (select id from revision where id = root_revision) + union + (select parent_id + from revision_history as h + join rev_list on h.id = rev_list.id) + ) + select * from rev_list; $$; + -- 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 + language sql as $$ -begin - return query - select revision.id, date, date_offset, - committer_date, committer_date_offset, - type, directory, message, - author.name as author_name, author.email as author_email, - committer.name as committer_name, committer.email as committer_email - from swh_revision_list(root_revision) as rev_list - join revision on revision.id = rev_list - join person as author on revision.author = author.id - join person as committer on revision.committer = committer.id; - return; -end + select revision.id, date, date_offset, + committer_date, committer_date_offset, + type, directory, message, + author.name as author_name, author.email as author_email, + committer.name as committer_name, committer.email as committer_email + from swh_revision_list(root_revision) as rev_list + join revision on revision.id = rev_list + join person as author on revision.author = author.id + join person as committer on revision.committer = committer.id; $$; + -- 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. +-- directory root, reach down to a file with the desired content. Return NULL +-- if no match is found. -- -- 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 + language sql as $$ -declare - d content_dir; -begin with recursive path as ( -- Recursively build a path from the requested content to a root -- directory. Each iteration returns a pair (dir_id, filename) where -- filename is relative to dir_id. Stops when no parent directory can -- be found. (select dir.id as dir_id, dir_entry_f.name as name, 0 as depth from directory_entry_file as dir_entry_f join content on content.sha1_git = dir_entry_f.target join directory as dir on dir.file_entries @> array[dir_entry_f.id] where content.sha1 = content_id limit 1) union all (select dir.id as dir_id, (dir_entry_d.name || '/' || path.name)::unix_path as name, path.depth + 1 from path join directory_entry_dir as dir_entry_d on dir_entry_d.target = path.dir_id join directory as dir on dir.dir_entries @> array[dir_entry_d.id] limit 1) ) - select dir_id, name from path order by depth desc limit 1 - into strict d; - - return d; -end + select dir_id, name from path order by depth desc limit 1; $$; + -- Walk the revision history starting from a given revision, until a matching --- occurrence is found. Return all occurrence information. +-- occurrence is found. Return all occurrence information if one is found, NULL +-- otherwise. create or replace function swh_revision_find_occurrence(revision_id sha1_git) returns occurrence language plpgsql as $$ declare occ occurrence%ROWTYPE; rev sha1_git; begin -- first check to see if revision_id is already pointed by an occurrence select origin, branch, revision from occurrence_history as occ_hist where occ_hist.revision = revision_id order by upper(occ_hist.validity) -- TODO filter by authority? limit 1 into occ; -- no occurrence point to revision_id, walk up the history if not found then -- recursively walk the history, stopping immediately before a revision -- pointed to by an occurrence. -- TODO find a nicer way to stop at, but *including*, that revision with recursive revlog as ( (select revision_id as rev_id, 0 as depth) union all (select hist.parent_id as rev_id, revlog.depth + 1 from revlog join revision_history as hist on hist.id = revlog.rev_id and not exists(select 1 from occurrence_history where revision = hist.parent_id) limit 1) ) select rev_id from revlog order by depth desc limit 1 - into strict rev; + into rev; + if not found then return null; end if; -- as we stopped before a pointed by revision, look it up again and -- return its data select origin, branch, revision from revision_history as rev_hist, occurrence_history as occ_hist where rev_hist.id = rev and occ_hist.revision = rev_hist.parent_id order by upper(occ_hist.validity) -- TODO filter by authority? limit 1 - into strict occ; -- will fail if no occurrence is found, and that's OK + into occ; end if; - return occ; + return occ; -- might be NULL end $$; + -- 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. +-- Return information about an arbitrary occurrence/revision/tree if one is +-- found, NULL otherwise. create or replace function swh_content_find_occurrence(content_id sha1) returns content_occurrence language plpgsql as $$ declare dir content_dir; rev sha1_git; occ occurrence%ROWTYPE; coc content_occurrence; begin -- each step could fail if no results are found, and that's OK select * from swh_content_find_directory(content_id) -- look up directory - into strict dir; + into dir; + if not found then return null; end if; + select id from revision where directory = dir.directory -- look up revision limit 1 - into strict rev; + into rev; + if not found then return null; end if; + select * from swh_revision_find_occurrence(rev) -- look up occurrence - into strict occ; + into occ; + if not found then return null; end if; select origin.type, origin.url, occ.branch, rev, dir.path from origin where origin.id = occ.origin - into strict coc; + into coc; - return coc; + return coc; -- might be NULL end $$; diff --git a/sql/swh-schema.sql b/sql/swh-schema.sql index 6227e28d..163d2f78 100644 --- a/sql/swh-schema.sql +++ b/sql/swh-schema.sql @@ -1,340 +1,340 @@ --- --- 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(16, now(), 'Work In Progress'); + values(17, 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.) create domain unix_path as bytea; -- 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'); -- 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 ); -- 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); -- Content we have seen but skipped for some reason. This table is -- separate from the content table as we might not have the sha1 -- checksum of that data (for instance when we inject git -- repositories, objects that are too big will be skipped here, and we -- will only know their sha1_git). 'reason' contains the reason the -- content was skipped. origin is a nullable column allowing to find -- out which origin contains that skipped content. create table skipped_content ( sha1 sha1, sha1_git sha1_git, sha256 sha256, length bigint not null, ctime timestamptz not null default now(), status content_status not null default 'absent', reason text not null, origin bigint references origin(id), unique (sha1, sha1_git, sha256) ); -- those indexes support multiple NULL values. create unique index on skipped_content(sha1); create unique index on skipped_content(sha1_git); create unique index on skipped_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 ); -- 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 array dir_entries -- 2. list the contained directory_entry_file using array file_entries -- 3. list the contained directory_entry_rev using array rev_entries -- 4. UNION -- -- Synonyms/mappings: -- * git: tree create table directory ( id sha1_git primary key, dir_entries bigint[], -- sub-directories, reference directory_entry_dir file_entries bigint[], -- contained files, reference directory_entry_file rev_entries bigint[] -- mounted revisions, reference directory_entry_rev ); create index on directory using gin (dir_entries); create index on directory using gin (file_entries); create index on directory using gin (rev_entries); -- 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); create unique index on directory_entry_dir(target, name, perms) where atime is null and mtime is null and ctime is null; -- 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); create unique index on directory_entry_file(target, name, perms) where atime is null and mtime is null and ctime is null; -- 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); create unique index on directory_entry_rev(target, name, perms) where atime is null and mtime is null and ctime is null; 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, date_offset smallint, committer_date timestamptz, committer_date_offset smallint, type revision_type not null, directory sha1_git, -- file-system tree message bytea, 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_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, -- 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, 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, date timestamptz, date_offset smallint, name text, comment bytea, author bigint references person(id) ); diff --git a/sql/upgrades/017.sql b/sql/upgrades/017.sql new file mode 100644 index 00000000..d7dcdbd0 --- /dev/null +++ b/sql/upgrades/017.sql @@ -0,0 +1,208 @@ +-- SWH DB schema upgrade +-- from_version: 16 +-- to_version: 17 +-- description: improve swh_content_missing, generic shw_directory_entry_add, +-- LANGUAGE sql where possible, better API for swh_content_find_* + +create or replace function swh_content_missing() + returns setof content_signature + language plpgsql +as $$ +begin + return query + (select sha1, sha1_git, sha256 from tmp_content as tmp + where not exists + (select 1 from content as c where c.sha1 = tmp.sha1)) + union + (select sha1, sha1_git, sha256 from tmp_content as tmp + where not exists + (select 1 from content as c where c.sha1_git = tmp.sha1_git)) + union + (select sha1, sha1_git, sha256 from tmp_content as tmp + where not exists + (select 1 from content as c where c.sha256 = tmp.sha256)); + return; +end +$$; + +create or replace function swh_directory_entry_add(typ directory_entry_type) + returns void + language plpgsql +as $$ +begin + execute format(' + insert into directory_entry_%1$s (target, name, perms, atime, mtime, ctime) + select distinct t.target, t.name, t.perms, t.atime, t.mtime, t.ctime + from tmp_directory_entry_%1$s t + where not exists ( + select 1 + from directory_entry_%1$s i + where t.target = i.target and t.name = i.name and t.perms = i.perms and + t.atime is not distinct from i.atime and + t.mtime is not distinct from i.mtime and + t.ctime is not distinct from i.ctime) + ', typ); + + execute format(' + with new_entries as ( + select t.dir_id, array_agg(i.id) as entries + from tmp_directory_entry_%1$s t + inner join directory_entry_%1$s i + on t.target = i.target and t.name = i.name and t.perms = i.perms and + t.atime is not distinct from i.atime and + t.mtime is not distinct from i.mtime and + t.ctime is not distinct from i.ctime + group by t.dir_id + ) + update directory as d + set %1$s_entries = new_entries.entries + from new_entries + where d.id = new_entries.dir_id + ', typ); + + return; +end +$$; + +drop function swh_directory_entry_dir_add (); +drop function swh_directory_entry_file_add (); +drop function swh_directory_entry_rev_add (); + +create or replace function swh_mktemp_revision() + returns void + language sql +as $$ + create temporary table tmp_revision ( + like revision including defaults, + author_name text not null default '', + author_email text not null default '', + committer_name text not null default '', + committer_email text not null default '' + ) on commit drop; + alter table tmp_revision drop column author; + alter table tmp_revision drop column committer; +$$; + +create or replace function swh_mktemp_release() + returns void + language sql +as $$ + create temporary table tmp_release ( + like release including defaults, + author_name text not null default '', + author_email text not null default '' + ) on commit drop; + alter table tmp_release drop column author; +$$; + +create or replace function swh_directory_walk_one(walked_dir_id sha1_git) + returns setof directory_entry + language sql +as $$ + with dir as ( + select id as dir_id, dir_entries, file_entries, rev_entries + from directory + where id = walked_dir_id), + ls_d as (select dir_id, unnest(dir_entries) as entry_id from dir), + ls_f as (select dir_id, unnest(file_entries) as entry_id from dir), + ls_r as (select dir_id, unnest(rev_entries) as entry_id from dir) + (select dir_id, 'dir'::directory_entry_type as type, + target, name, perms, atime, mtime, ctime + from ls_d + left join directory_entry_dir d on ls_d.entry_id = d.id) + union + (select dir_id, 'file'::directory_entry_type as type, + target, name, perms, atime, mtime, ctime + from ls_f + left join directory_entry_file d on ls_f.entry_id = d.id) + union + (select dir_id, 'rev'::directory_entry_type as type, + target, name, perms, atime, mtime, ctime + from ls_r + left join directory_entry_rev d on ls_r.entry_id = d.id) + order by name; +$$; + +create or replace function swh_revision_list(root_revision sha1_git) + returns setof sha1_git + language sql +as $$ + with recursive rev_list(id) as ( + (select id from revision where id = root_revision) + union + (select parent_id + from revision_history as h + join rev_list on h.id = rev_list.id) + ) + select * from rev_list; +$$; + +create or replace function swh_revision_log(root_revision sha1_git) + returns setof revision_log_entry + language sql +as $$ + select revision.id, date, date_offset, + committer_date, committer_date_offset, + type, directory, message, + author.name as author_name, author.email as author_email, + committer.name as committer_name, committer.email as committer_email + from swh_revision_list(root_revision) as rev_list + join revision on revision.id = rev_list + join person as author on revision.author = author.id + join person as committer on revision.committer = committer.id; +$$; + +create or replace function swh_content_find_directory(content_id sha1) + returns content_dir + language sql +as $$ + with recursive path as ( + (select dir.id as dir_id, dir_entry_f.name as name, 0 as depth + from directory_entry_file as dir_entry_f + join content on content.sha1_git = dir_entry_f.target + join directory as dir on dir.file_entries @> array[dir_entry_f.id] + where content.sha1 = content_id + limit 1) + union all + (select dir.id as dir_id, + (dir_entry_d.name || '/' || path.name)::unix_path as name, + path.depth + 1 + from path + join directory_entry_dir as dir_entry_d on dir_entry_d.target = path.dir_id + join directory as dir on dir.dir_entries @> array[dir_entry_d.id] + limit 1) + ) + select dir_id, name from path order by depth desc limit 1; +$$; + +create or replace function swh_content_find_occurrence(content_id sha1) + returns content_occurrence + language plpgsql +as $$ +declare + dir content_dir; + rev sha1_git; + occ occurrence%ROWTYPE; + coc content_occurrence; +begin + select * from swh_content_find_directory(content_id) + into dir; + if not found then return null; end if; + + select id from revision where directory = dir.directory + limit 1 + into rev; + if not found then return null; end if; + + select * from swh_revision_find_occurrence(rev) + into occ; + if not found then return null; end if; + + select origin.type, origin.url, occ.branch, rev, dir.path + from origin + where origin.id = occ.origin + into coc; + + return coc; +end +$$; diff --git a/swh.storage.egg-info/PKG-INFO b/swh.storage.egg-info/PKG-INFO index bf733ef5..ab1ab588 100644 --- a/swh.storage.egg-info/PKG-INFO +++ b/swh.storage.egg-info/PKG-INFO @@ -1,10 +1,10 @@ Metadata-Version: 1.0 Name: swh.storage -Version: 0.0.7 +Version: 0.0.8 Summary: Software Heritage storage manager Home-page: https://forge.softwareheritage.org/diffusion/DSTO/ Author: Software Heritage developers Author-email: swh-devel@inria.fr License: UNKNOWN Description: UNKNOWN Platform: UNKNOWN diff --git a/swh.storage.egg-info/SOURCES.txt b/swh.storage.egg-info/SOURCES.txt index 3d015733..242c484f 100644 --- a/swh.storage.egg-info/SOURCES.txt +++ b/swh.storage.egg-info/SOURCES.txt @@ -1,46 +1,46 @@ .gitignore MANIFEST.in Makefile Makefile.local README.db_testing README.dev requirements.txt setup.py version.txt bin/swh-objstorage-add-dir bin/swh-objstorage-fsck bin/swh-storage-add-dir debian/changelog debian/compat debian/control debian/copyright debian/rules debian/source/format sql/.gitignore sql/Makefile sql/TODO sql/clusters.dot sql/swh-data.sql sql/swh-func.sql sql/swh-init.sql sql/swh-schema.sql sql/bin/dot_add_content sql/upgrades/015.sql sql/upgrades/016.sql +sql/upgrades/017.sql swh.storage.egg-info/PKG-INFO swh.storage.egg-info/SOURCES.txt swh.storage.egg-info/dependency_links.txt swh.storage.egg-info/requires.txt swh.storage.egg-info/top_level.txt swh/storage/__init__.py swh/storage/db.py swh/storage/objstorage.py swh/storage/storage.py swh/storage/api/__init__.py swh/storage/api/client.py swh/storage/api/server.py -swh/storage/tests/__init__.py swh/storage/tests/test_api_client.py swh/storage/tests/test_db.py swh/storage/tests/test_objstorage.py swh/storage/tests/test_storage.py \ No newline at end of file diff --git a/swh.storage.egg-info/requires.txt b/swh.storage.egg-info/requires.txt index 758a0f78..38ba03a8 100644 --- a/swh.storage.egg-info/requires.txt +++ b/swh.storage.egg-info/requires.txt @@ -1,5 +1,5 @@ flask psycopg2 requests -swh.core>=0.0.6 +swh.core>=0.0.7 vcversioner diff --git a/swh/storage/api/client.py b/swh/storage/api/client.py index 8f9279a7..78da0ee7 100644 --- a/swh/storage/api/client.py +++ b/swh/storage/api/client.py @@ -1,108 +1,111 @@ # 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 pickle import requests from swh.core.serializers import msgpack_dumps, msgpack_loads, SWHJSONDecoder def encode_data(data): return msgpack_dumps(data) def decode_response(response): content_type = response.headers['content-type'] if content_type.startswith('application/x-msgpack'): r = msgpack_loads(response.content) elif content_type.startswith('application/json'): r = response.json(cls=SWHJSONDecoder) else: raise ValueError('Wrong content type `%s` for API response' % content_type) return r class RemoteStorage(): """Proxy to a remote storage API""" def __init__(self, base_url): self.base_url = base_url def url(self, endpoint): return '%s%s' % (self.base_url, endpoint) def post(self, endpoint, data): response = requests.post( self.url(endpoint), data=encode_data(data), headers={'content-type': 'application/x-msgpack'}, ) # XXX: this breaks language-independence and should be # replaced by proper unserialization if response.status_code == 400: raise pickle.loads(decode_response(response)) return decode_response(response) def get(self, endpoint, data): response = requests.get( self.url(endpoint), params=data, ) if response.status_code == 404: return None else: return decode_response(response) def content_add(self, content): return self.post('content/add', {'content': content}) def content_missing(self, content, key_hash='sha1'): return self.post('content/missing', {'content': content, 'key_hash': key_hash}) - def content_find(self, content): + def content_exist(self, content): return self.post('content/present', {'content': content}) + def content_find_occurrence(self, content): + return self.post('content/occurrence', {'content': content}) + def directory_add(self, directories): return self.post('directory/add', {'directories': directories}) def directory_missing(self, directories): return self.post('directory/missing', {'directories': directories}) def directory_get(self, directory): return [tuple(entry) for entry in self.get('directory', {'directory': directory})] def revision_add(self, revisions): return self.post('revision/add', {'revisions': revisions}) def revision_missing(self, revisions): return self.post('revision/missing', {'revisions': revisions}) def release_add(self, releases): return self.post('release/add', {'releases': releases}) def release_missing(self, releases): return self.post('release/missing', {'releases': releases}) def occurrence_add(self, occurrences): return self.post('occurrence/add', {'occurrences': occurrences}) def origin_get(self, origin): origin = self.get('origin', origin) if not origin: return None else: return origin['id'] def origin_add_one(self, origin): return self.post('origin', {'origin': origin}) diff --git a/swh/storage/api/server.py b/swh/storage/api/server.py index 484fa24d..0c23a2e4 100644 --- a/swh/storage/api/server.py +++ b/swh/storage/api/server.py @@ -1,167 +1,173 @@ # 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 json import logging import pickle from flask import Flask, Request, Response, abort, g, request from swh.core import config from swh.core.serializers import msgpack_dumps, msgpack_loads, SWHJSONDecoder from swh.storage import Storage DEFAULT_CONFIG = { 'db': ('str', 'dbname=softwareheritage-dev'), 'storage_base': ('str', '/tmp/swh-storage/test'), } class BytesRequest(Request): """Request with proper escaping of arbitrary byte sequences.""" encoding = 'utf-8' encoding_errors = 'surrogateescape' app = Flask(__name__) app.request_class = BytesRequest def encode_data(data): return Response( msgpack_dumps(data), mimetype='application/x-msgpack', ) def decode_request(request): content_type = request.mimetype data = request.get_data() if content_type == 'application/x-msgpack': r = msgpack_loads(data) elif content_type == 'application/json': r = json.loads(data, cls=SWHJSONDecoder) else: raise ValueError('Wrong content type `%s` for API request' % content_type) return r @app.errorhandler(Exception) def error_handler(exception): # XXX: this breaks language-independence and should be # replaced by proper serialization of errors response = encode_data(pickle.dumps(exception)) response.status_code = 400 return response @app.before_request def before_request(): g.storage = Storage(app.config['db'], app.config['storage_base']) @app.route('/') def index(): return 'Hello' @app.route('/content/missing', methods=['POST']) def content_missing(): return encode_data(g.storage.content_missing(**decode_request(request))) @app.route('/content/present', methods=['POST']) -def content_find(): - return encode_data(g.storage.content_find(**decode_request(request))) +def content_exist(): + return encode_data(g.storage.content_exist(**decode_request(request))) + + +@app.route('/content/occurrence', methods=['POST']) +def content_find_occurrence(): + res = g.storage.content_find_occurrence(**decode_request(request)) + return encode_data(res) @app.route('/content/add', methods=['POST']) def content_add(): return encode_data(g.storage.content_add(**decode_request(request))) @app.route('/directory/missing', methods=['POST']) def directory_missing(): return encode_data(g.storage.directory_missing(**decode_request(request))) @app.route('/directory/add', methods=['POST']) def directory_add(): return encode_data(g.storage.directory_add(**decode_request(request))) @app.route('/directory', methods=['GET']) def directory_get(): dir = request.args['directory'].encode('utf-8', 'surrogateescape') return encode_data(g.storage.directory_get(dir)) @app.route('/revision/add', methods=['POST']) def revision_add(): return encode_data(g.storage.revision_add(**decode_request(request))) @app.route('/revision/missing', methods=['POST']) def revision_missing(): return encode_data(g.storage.revision_missing(**decode_request(request))) @app.route('/release/add', methods=['POST']) def release_add(): return encode_data(g.storage.release_add(**decode_request(request))) @app.route('/release/missing', methods=['POST']) def release_missing(): return encode_data(g.storage.release_missing(**decode_request(request))) @app.route('/occurrence/add', methods=['POST']) def occurrence_add(): return encode_data(g.storage.occurrence_add(**decode_request(request))) @app.route('/origin', methods=['GET']) def origin_get(): origin = { 'type': request.args['type'], 'url': request.args['url'], } id = g.storage.origin_get(origin) if not id: abort(404) else: origin['id'] = id return encode_data(origin) @app.route('/origin', methods=['POST']) def origin_add_one(): return encode_data(g.storage.origin_add_one(**decode_request(request))) def run_from_webserver(environ, start_response): """Run the WSGI app from the webserver, loading the configuration.""" config_path = '/etc/softwareheritage/storage/storage.ini' app.config.update(config.read(config_path, DEFAULT_CONFIG)) handler = logging.StreamHandler() app.logger.addHandler(handler) return app(environ, start_response) if __name__ == '__main__': import sys app.config.update(config.read(sys.argv[1], DEFAULT_CONFIG)) app.run(debug=True) diff --git a/swh/storage/db.py b/swh/storage/db.py index 07200db1..0ee3e5d8 100644 --- a/swh/storage/db.py +++ b/swh/storage/db.py @@ -1,218 +1,239 @@ # 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 def entry_to_bytes(entry): """Convert an entry coming from the database to bytes""" if isinstance(entry, memoryview): return entry.tobytes() return entry def line_to_bytes(line): """Convert a line coming from the database to bytes""" return line.__class__(entry_to_bytes(entry) for entry in line) def cursor_to_bytes(cursor): """Yield all the data from a cursor as bytes""" yield from (line_to_bytes(line) for line in cursor) 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),)) @stored_procedure('swh_mktemp_revision') def mktemp_revision(self, cur=None): pass @stored_procedure('swh_mktemp_release') def mktemp_release(self, cur=None): pass def copy_to(self, items, tblname, columns, cur=None, item_cb=None): def escape(data): if data is None: return '' if isinstance(data, bytes): return '\\x%s' % binascii.hexlify(data).decode('ascii') elif isinstance(data, str): return '"%s"' % data.replace('"', '""') else: return str(data) with tempfile.TemporaryFile('w+') as f: for d in items: if item_cb is not None: item_cb(d) line = [escape(d.get(k)) for k in columns] f.write(','.join(line)) f.write('\n') f.seek(0) self._cursor(cur).copy_expert('COPY %s (%s) FROM STDIN CSV' % ( tblname, ', '.join(columns)), f) @stored_procedure('swh_content_add') def content_add_from_temp(self, cur=None): pass @stored_procedure('swh_skipped_content_add') def skipped_content_add_from_temp(self, cur=None): pass @stored_procedure('swh_revision_add') def revision_add_from_temp(self, cur=None): pass @stored_procedure('swh_release_add') def release_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 cursor_to_bytes(cur) def skipped_content_missing_from_temp(self, cur=None): cur = self._cursor(cur) cur.execute("""SELECT sha1, sha1_git, sha256 FROM swh_skipped_content_missing()""") yield from cursor_to_bytes(cur) def content_find(self, sha1=None, sha1_git=None, sha256=None, cur=None): """Find the content optionally on a combination of the following checksums sha1, sha1_git or sha256. Args: sha1: sha1 content git_sha1: the sha1 computed `a la git` sha1 of the content sha256: sha256 content Returns: - The content if found or null. + The triplet (sha1, sha1_git, sha256) if found or None. """ cur = self._cursor(cur) cur.execute("""SELECT sha1, sha1_git, sha256 FROM swh_content_find(%s, %s, %s) LIMIT 1""", (sha1, sha1_git, sha256)) - yield from cursor_to_bytes(cur) + content = line_to_bytes(cur.fetchone()) + return None if content == (None, None, None) else content + + def content_find_occurrence(self, sha1, cur=None): + """Find one content's occurrence. + + Args: + sha1: sha1 content + cur: cursor to use + + Returns: + One occurrence for that particular sha1 + + """ + cur = self._cursor(cur) + + cur.execute("""SELECT * + FROM swh_content_find_occurrence(%s) + LIMIT 1""", + (sha1, )) + + return line_to_bytes(cur.fetchone()) def directory_missing_from_temp(self, cur=None): cur = self._cursor(cur) cur.execute('SELECT * FROM swh_directory_missing()') yield from cursor_to_bytes(cur) def directory_walk_one(self, directory, cur=None): cur = self._cursor(cur) cur.execute('SELECT * FROM swh_directory_walk_one(%s)', (directory,)) yield from cursor_to_bytes(cur) def revision_missing_from_temp(self, cur=None): cur = self._cursor(cur) cur.execute('SELECT id FROM swh_revision_missing() as r(id)') yield from cursor_to_bytes(cur) def release_missing_from_temp(self, cur=None): cur = self._cursor(cur) cur.execute('SELECT id FROM swh_release_missing() as r(id)') yield from cursor_to_bytes(cur) diff --git a/swh/storage/objstorage.py b/swh/storage/objstorage.py index d843f0b8..bf03ed70 100644 --- a/swh/storage/objstorage.py +++ b/swh/storage/objstorage.py @@ -1,388 +1,391 @@ # 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 gzip import os import shutil import tempfile from contextlib import contextmanager from swh.core import hashutil ID_HASH_ALGO = 'sha1' # ID_HASH_ALGO = 'sha1_git' GZIP_BUFSIZ = 1048576 class Error(Exception): def __str__(self): return 'storage error on object: %s' % self.args class ObjNotFoundError(Error): def __str__(self): return 'object not found: %s' % self.args def _obj_dir(hex_obj_id, root_dir, depth): """compute the storage directory of an object Args: hex_obj_id: object id as hexlified string root_dir: object storage root directory depth: slicing depth of object IDs in the storage see also: `_obj_path` """ if len(hex_obj_id) < depth * 2: raise ValueError('object id "%s" is too short for slicing at depth %d' % (hex_obj_id, depth)) # compute [depth] substrings of [obj_id], each of length 2, starting from # the beginning id_steps = [hex_obj_id[i*2:i*2+2] for i in range(0, depth)] steps = [root_dir] + id_steps return os.path.join(*steps) def _obj_path(hex_obj_id, root_dir, depth): """similar to `obj_dir`, but also include the actual object file name in the returned path """ return os.path.join(_obj_dir(hex_obj_id, root_dir, depth), hex_obj_id) @contextmanager def _write_obj_file(hex_obj_id, root_dir, depth): """context manager for writing object files to the object storage During writing data are written to a temporary file, which is atomically renamed to the right file name after closing. This context manager also takes care of (gzip) compressing the data on the fly. Yields: a file-like object open for writing bytes Sample usage: with _write_obj_file(hex_obj_id, root_dir, depth) as f: f.write(obj_data) """ dir = _obj_dir(hex_obj_id, root_dir, depth) if not os.path.isdir(dir): os.makedirs(dir) path = os.path.join(dir, hex_obj_id) - tmp_path = path + '.tmp' - with gzip.GzipFile(tmp_path, 'wb') as f: + (tmp, tmp_path) = tempfile.mkstemp(suffix='.tmp', prefix='hex_obj_id.', + dir=dir) + tmp_f = os.fdopen(tmp, 'wb') + with gzip.GzipFile(filename=tmp_path, fileobj=tmp_f) as f: yield f + tmp_f.close() os.rename(tmp_path, path) class ObjStorage: """high-level API to manipulate the Software Heritage object storage Conceptually, the object storage offers 4 methods: - add() add a new object, returning an object id - __contains__() check if an object is present, by object id - get() retrieve the content of an object, by object id - check() check the integrity of an object, by object id Variants of the above methods are implemented by this class, depending on how the content of an object is specified (bytes, file-like object, etc.). On disk, an object storage is a directory tree containing files named after their object IDs. An object ID is a checksum of its content, depending on the value of the ID_HASH_ALGO constant (see hashutil for its meaning). To avoid directories that contain too many files, the object storage has a given depth (default: 3). Each depth level consumes two characters of the object id. So for instance a file with (git) SHA1 of 34973274ccef6ab4dfaaf86599792fa9c3fe4689 will be stored in an object storage configured at depth 3 at 34/97/32/34973274ccef6ab4dfaaf86599792fa9c3fe4689. The actual files in the storage are stored in gzipped compressed format. Each file can hence be self-verified (on the shell) with something like: actual_id=34973274ccef6ab4dfaaf86599792fa9c3fe4689 expected_id=$(zcat $filename | sha1sum | cut -f 1 -d' ') if [ $actual_id != $expected_id ] ; then echo "AYEEE, invalid object $actual_id /o\" fi """ def __init__(self, root, depth=3): """create a proxy object to the object storage Args: root: object storage root directory depth: slicing depth of object IDs in the storage """ if not os.path.isdir(root): raise ValueError('obj storage root "%s" is not a directory' % root) self._root_dir = root self._depth = depth self._temp_dir = os.path.join(root, 'tmp') if not os.path.isdir(self._temp_dir): os.makedirs(self._temp_dir) def __obj_dir(self, hex_obj_id): """_obj_dir wrapper using this storage configuration""" return _obj_dir(hex_obj_id, self._root_dir, self._depth) def __obj_path(self, hex_obj_id): """_obj_path wrapper using this storage configuration""" return _obj_path(hex_obj_id, self._root_dir, self._depth) def __contains__(self, obj_id): """check whether a given object id is present in the storage or not Return: True iff the object id is present in the storage """ hex_obj_id = hashutil.hash_to_hex(obj_id) return os.path.exists(_obj_path(hex_obj_id, self._root_dir, self._depth)) def add_bytes(self, bytes, obj_id=None): """add a new object to the object storage Args: bytes: content of the object to be added to the storage obj_id: checksums of `bytes` as computed by ID_HASH_ALGO. When given, obj_id will be trusted to match bytes. If missing, obj_id will be computed on the fly. """ if obj_id is None: # missing checksum, compute it in memory and write to file h = hashutil._new_hash(ID_HASH_ALGO, len(bytes)) h.update(bytes) obj_id = h.digest() if obj_id in self: return obj_id hex_obj_id = hashutil.hash_to_hex(obj_id) # object is either absent, or present but overwrite is requested with _write_obj_file(hex_obj_id, root_dir=self._root_dir, depth=self._depth) as f: f.write(bytes) return obj_id def add_file(self, f, length, obj_id=None): """similar to `add_bytes`, but add the content of file-like object f to the object storage add_file will read the file content only once, and avoid storing all of it in memory """ if obj_id is None: # unknkown object id: work on temp file, compute checksum as we go, # mv temp file into place (tmp, tmp_path) = tempfile.mkstemp(dir=self._temp_dir) try: t = os.fdopen(tmp, 'wb') tz = gzip.GzipFile(fileobj=t) sums = hashutil._hash_file_obj(f, length, algorithms=[ID_HASH_ALGO], chunk_cb=lambda b: tz.write(b)) tz.close() t.close() obj_id = sums[ID_HASH_ALGO] if obj_id in self: return obj_id hex_obj_id = hashutil.hash_to_hex(obj_id) dir = self.__obj_dir(hex_obj_id) if not os.path.isdir(dir): os.makedirs(dir) path = os.path.join(dir, hex_obj_id) os.rename(tmp_path, path) finally: if os.path.exists(tmp_path): os.unlink(tmp_path) else: # known object id: write to .new file, rename if obj_id in self: return obj_id hex_obj_id = hashutil.hash_to_hex(obj_id) with _write_obj_file(hex_obj_id, root_dir=self._root_dir, depth=self._depth) as obj: shutil.copyfileobj(f, obj) return obj_id @contextmanager def get_file_obj(self, obj_id): """context manager to read the content of an object Args: obj_id: object id Yields: a file-like object open for reading (bytes) Raises: ObjNotFoundError: if the requested object is missing Sample usage: with objstorage.get_file_obj(obj_id) as f: do_something(f.read()) """ if obj_id not in self: raise ObjNotFoundError(obj_id) hex_obj_id = hashutil.hash_to_hex(obj_id) path = self.__obj_path(hex_obj_id) with gzip.GzipFile(path, 'rb') as f: yield f def get_bytes(self, obj_id): """retrieve the content of a given object Args: obj_id: object id Returns: the content of the requested objects as bytes Raises: ObjNotFoundError: if the requested object is missing """ with self.get_file_obj(obj_id) as f: return f.read() def _get_file_path(self, obj_id): """retrieve the path of a given object in the objects storage Note that the path point to a gzip-compressed file, so you need gzip.open() or equivalent to get the actual object content. Args: obj_id: object id Returns: a file path pointing into the object storage Raises: ObjNotFoundError: if the requested object is missing """ if obj_id not in self: raise ObjNotFoundError(obj_id) hex_obj_id = hashutil.hash_to_hex(obj_id) return self.__obj_path(hex_obj_id) def check(self, obj_id): """integrity check for a given object verify that the file object is in place, and that the gzipped content matches the object id Args: obj_id: object id Raises: ObjNotFoundError: if the requested object is missing Error: if the requested object is corrupt """ if obj_id not in self: raise ObjNotFoundError(obj_id) hex_obj_id = hashutil.hash_to_hex(obj_id) try: with gzip.open(self.__obj_path(hex_obj_id)) as f: length = None if ID_HASH_ALGO.endswith('_git'): # if the hashing algorithm is git-like, we need to know the # content size to hash on the fly. Do a first pass here to # compute the size length = 0 while True: chunk = f.read(GZIP_BUFSIZ) length += len(chunk) if not chunk: break f.rewind() checksums = hashutil._hash_file_obj(f, length, algorithms=[ID_HASH_ALGO]) actual_obj_id = checksums[ID_HASH_ALGO] if obj_id != actual_obj_id: raise Error('corrupt object %s should have id %s' % (obj_id, actual_obj_id)) except (OSError, IOError): # IOError is for compatibility with older python versions raise Error('corrupt object %s is not a gzip file' % obj_id) def __iter__(self): """iterate over the object identifiers currently available in the storage Warning: with the current implementation of the object storage, this method will walk the filesystem to list objects, meaning that listing all objects will be very slow for large storages. You almost certainly don't want to use this method in production. Return: iterator over object IDs """ def obj_iterator(): # XXX hackish: it does not verify that the depth of found files # matches the slicing depth of the storage for root, _dirs, files in os.walk(self._root_dir): for f in files: yield bytes.fromhex(f) return obj_iterator() def __len__(self): """compute the number of objects available in the storage Warning: this currently uses `__iter__`, its warning about bad performances applies Return: number of objects contained in the storage """ return sum(1 for i in self) diff --git a/swh/storage/storage.py b/swh/storage/storage.py index 8b4aded2..1f770275 100644 --- a/swh/storage/storage.py +++ b/swh/storage/storage.py @@ -1,526 +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 operator import itemgetter 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): - """Predicate to check the presence of a content's hashes. + """Find a content hash in db. Args: - hashes: iterable of dictionaries representing individual pieces of - hash. Each dictionary has the following keys: - - a key for each checksum algorithm in swh.core.hashutil.ALGORITHMS - mapped to the corresponding checksum + 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 + 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 nor sha256 + ValueError in case the key of the dictionary is not sha1, sha1_git + nor sha256. """ db = self.db - # filter out the checksums - keys = ['sha1', 'sha1_git', 'sha256'] - if content == {}: raise ValueError('Key must be one of sha1, git_sha1, sha256.') for key in content.keys(): - if key not in keys: + if key not in SWH_HASH_KEYS: raise ValueError('Key must be one of sha1, git_sha1, sha256.') # format the output - found_hashes = db.content_find(sha1=content.get('sha1'), - sha1_git=content.get('sha1_git'), - sha256=content.get('sha256'), - cur=cur) + found_hash = db.content_find(sha1=content.get('sha1'), + sha1_git=content.get('sha1_git'), + sha256=content.get('sha256'), + cur=cur) + + return found_hash - hashes = list(found_hashes) - if len(hashes) > 0: - return hashes[0] != (None, None, None) - return False + @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] diff --git a/swh/storage/tests/__init__.py b/swh/storage/tests/__init__.py deleted file mode 100644 index fdffa2a0..00000000 --- a/swh/storage/tests/__init__.py +++ /dev/null @@ -1 +0,0 @@ -# placeholder diff --git a/swh/storage/tests/test_api_client.py b/swh/storage/tests/test_api_client.py index eda319b9..2af6b875 100644 --- a/swh/storage/tests/test_api_client.py +++ b/swh/storage/tests/test_api_client.py @@ -1,78 +1,77 @@ # 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 multiprocessing import socket import time import unittest from urllib.request import urlopen -from .test_storage import AbstractTestStorage - +from swh.storage.tests.test_storage import AbstractTestStorage from swh.storage.api.client import RemoteStorage from swh.storage.api.server import app class TestRemoteStorage(AbstractTestStorage, unittest.TestCase): """Test the remote storage API. This class doesn't define any tests as we want identical functionality between local and remote storage. All the tests are therefore defined in AbstractTestStorage. """ def setUp(self): super().setUp() self.start_server() self.storage = RemoteStorage(self.url()) def tearDown(self): self.stop_server() super().tearDown() def url(self): return 'http://127.0.0.1:%d/' % self.port def start_server(self): """Spawn the API server using multiprocessing""" self.process = None # WSGI app configuration self.app = app self.app.config['db'] = 'dbname=%s' % self.dbname self.app.config['storage_base'] = self.objroot # Get an available port number sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM) sock.bind(('127.0.0.1', 0)) self.port = sock.getsockname()[1] sock.close() # We need a worker function for multiprocessing def worker(app, port): return app.run(port=port, use_reloader=False) self.process = multiprocessing.Process( target=worker, args=(self.app, self.port) ) self.process.start() # Wait max. 5 seconds for server to spawn i = 0 while i < 20: try: urlopen(self.url()) except Exception: i += 1 time.sleep(0.25) else: break def stop_server(self): """Terminate the API server""" if self.process: self.process.terminate() diff --git a/swh/storage/tests/test_db.py b/swh/storage/tests/test_db.py index ca10f8cf..2cfcbc3d 100644 --- a/swh/storage/tests/test_db.py +++ b/swh/storage/tests/test_db.py @@ -1,46 +1,53 @@ # 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 os import unittest from nose.tools import istest from nose.plugins.attrib import attr from swh.core.tests.db_testing import DbTestFixture from swh.core.hashutil import hex_to_hash from swh.storage.db import Db +TEST_DIR = os.path.dirname(os.path.abspath(__file__)) +TEST_DATA_DIR = os.path.join(TEST_DIR, '../../../../swh-storage-testdata') + + @attr('db') class TestDb(DbTestFixture, unittest.TestCase): + TEST_DB_DUMP = os.path.join(TEST_DATA_DIR, 'dumps/swh.dump') + def setUp(self): super().setUp() self.db = Db(self.conn) def tearDown(self): self.db.conn.close() super().tearDown() @istest def add_content(self): cur = self.cursor sha1 = hex_to_hash('34973274ccef6ab4dfaaf86599792fa9c3fe4689') self.db.mktemp('content', cur) self.db.copy_to([{ 'sha1': sha1, 'sha1_git': hex_to_hash( 'd81cc0710eb6cf9efd5b920a8453e1e07157b6cd'), 'sha256': hex_to_hash( '673650f936cb3b0a2f93ce09d81be107' '48b1b203c19e8176b4eefc1964a0cf3a'), 'length': 3}], 'tmp_content', ['sha1', 'sha1_git', 'sha256', 'length'], cur) self.db.content_add_from_temp(cur) self.cursor.execute('SELECT sha1 FROM content WHERE sha1 = %s', (sha1,)) self.assertEqual(self.cursor.fetchone()[0].tobytes(), sha1) diff --git a/swh/storage/tests/test_storage.py b/swh/storage/tests/test_storage.py index e3dc189c..c1fcde4e 100644 --- a/swh/storage/tests/test_storage.py +++ b/swh/storage/tests/test_storage.py @@ -1,304 +1,440 @@ # 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 os +import psycopg2 import shutil import tempfile import unittest from nose.tools import istest from nose.plugins.attrib import attr from swh.core.tests.db_testing import DbTestFixture from swh.core.hashutil import hex_to_hash from swh.storage import Storage +TEST_DIR = os.path.dirname(os.path.abspath(__file__)) +TEST_DATA_DIR = os.path.join(TEST_DIR, '../../../../swh-storage-testdata') + + @attr('db') class AbstractTestStorage(DbTestFixture): """Base class for Storage testing. This class is used as-is to test local storage (see TestStorage below) and remote storage (see TestRemoteStorage in test_remote_storage.py. We need to have the two classes inherit from this base class separately to avoid nosetests running the tests from the base class twice. """ + TEST_DB_DUMP = os.path.join(TEST_DATA_DIR, 'dumps/swh.dump') + 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.skipped_cont = { 'length': 1024 * 1024 * 200, 'sha1_git': hex_to_hash( '33e45d56f88993aae6a0198013efa80716fd8920'), 'reason': 'Content too long', 'status': 'absent', } self.dir = { 'id': b'4\x013\x422\x531\x000\xf51\xe62\xa73\xff7\xc3\xa90', 'entries': [ { 'name': b'foo', 'type': 'file', 'target': self.cont['sha1_git'], 'perms': 0o644, 'atime': None, 'ctime': None, 'mtime': None, }, { 'name': b'bar\xc3', 'type': 'dir', 'target': b'12345678901234567890', 'perms': 0o2000, 'atime': None, 'ctime': None, 'mtime': None, }, ], } + self.dir2 = { + 'id': b'4\x013\x422\x531\x000\xf51\xe62\xa73\xff7\xc3\xa95', + 'entries': [ + { + 'name': b'oof', + 'type': 'file', + 'target': self.cont2['sha1_git'], + 'perms': 0o644, + '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'], } + self.revision2 = { + 'id': b'87659012345678904321', + 'message': 'hello', + 'author_name': 'Roberto Dicosmo', + 'author_email': 'roberto@example.com', + 'committer_name': 'tony', + 'committer_email': 'ar@dumont.fr', + '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.dir2['id'], + } + self.origin = { 'url': 'file:///dev/null', 'type': 'git', } self.origin2 = { 'url': 'file:///dev/zero', 'type': 'git', } self.occurrence = { 'branch': 'master', 'revision': b'67890123456789012345', 'authority': 1, 'validity': datetime.datetime(2015, 1, 1, 23, 0, 0), } + self.occurrence2 = { + 'branch': 'master', + 'revision': self.revision2['id'], + 'authority': 1, + 'validity': datetime.datetime(2015, 1, 1, 23, 0, 0), + } + def tearDown(self): shutil.rmtree(self.objroot) super().tearDown() @istest def content_add(self): cont = self.cont self.storage.content_add([cont]) if hasattr(self.storage, 'objstorage'): 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_add_collision(self): + cont1 = self.cont + + # create (corrupted) content with same sha1{,_git} but != sha256 + cont1b = cont1.copy() + sha256_array = bytearray(cont1b['sha256']) + sha256_array[0] += 1 + cont1b['sha256'] = bytes(sha256_array) + + with self.assertRaises(psycopg2.IntegrityError): + self.storage.content_add([cont1, cont1b]) + @istest def skipped_content_add(self): cont = self.skipped_cont self.storage.content_add([self.skipped_cont]) self.cursor.execute('SELECT sha1, sha1_git, sha256, length, status,' 'reason FROM skipped_content WHERE sha1_git = %s', (cont['sha1_git'],)) datum = self.cursor.fetchone() self.assertEqual( (datum[0], datum[1].tobytes(), datum[2], datum[3], datum[4], datum[5]), (None, cont['sha1_git'], None, cont['length'], 'absent', 'Content too long')) @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 content_find_with_present_content(self): + def content_exist_with_present_content(self): # 1. with something to find cont = self.cont self.storage.content_add([cont]) - actually_present = self.storage.content_find({'sha1': cont['sha1']}) + actually_present = self.storage.content_exist({'sha1': cont['sha1']}) self.assertEquals(actually_present, True, "Should be present") # 2. with something to find - actually_present = self.storage.content_find( + actually_present = self.storage.content_exist( {'sha1_git': cont['sha1_git']}) self.assertEquals(actually_present, True, "Should be present") # 3. with something to find - actually_present = self.storage.content_find( + actually_present = self.storage.content_exist( {'sha256': cont['sha256']}) self.assertEquals(actually_present, True, "Should be present") # 4. with something to find - actually_present = self.storage.content_find( + actually_present = self.storage.content_exist( {'sha1': cont['sha1'], 'sha1_git': cont['sha1_git'], 'sha256': cont['sha256']}) self.assertEquals(actually_present, True, "Should be present") @istest - def content_find_with_non_present_content(self): + def content_exist_with_non_present_content(self): # 1. with something that does not exist missing_cont = self.missing_cont - actually_present = self.storage.content_find( + actually_present = self.storage.content_exist( {'sha1': missing_cont['sha1']}) self.assertEquals(actually_present, False, "Should be missing") # 2. with something that does not exist - actually_present = self.storage.content_find( + actually_present = self.storage.content_exist( {'sha1_git': missing_cont['sha1_git']}) self.assertEquals(actually_present, False, "Should be missing") # 3. with something that does not exist - actually_present = self.storage.content_find( + actually_present = self.storage.content_exist( {'sha256': missing_cont['sha256']}) self.assertEquals(actually_present, False, "Should be missing") @istest - def content_find_bad_input(self): + def content_exist_bad_input(self): # 1. with bad input with self.assertRaises(ValueError) as cm: - self.storage.content_find({}) # empty is bad + self.storage.content_exist({}) # empty is bad self.assertEqual(cm.exception.args, ('Key must be one of sha1, git_sha1, sha256.',)) # 2. with bad input with self.assertRaises(ValueError) as cm: - self.storage.content_find( + self.storage.content_exist( {'unknown-sha1': 'something'}) # not the right key self.assertEqual(cm.exception.args, ('Key must be one of sha1, git_sha1, sha256.',)) @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)) @istest def origin_add(self): self.assertIsNone(self.storage.origin_get(self.origin)) id = self.storage.origin_add_one(self.origin) self.assertEqual(self.storage.origin_get(self.origin), id) @istest def occurrence_add(self): origin_id = self.storage.origin_add_one(self.origin2) revision = self.revision.copy() revision['id'] = self.occurrence['revision'] self.storage.revision_add([revision]) self.occurrence['origin'] = origin_id self.storage.occurrence_add([self.occurrence]) + @istest + def content_find_occurrence_with_present_content(self): + # 1. with something to find + # given + self.storage.content_add([self.cont2]) + self.storage.directory_add([self.dir2]) # point to self.cont + self.storage.revision_add([self.revision2]) # points to self.dir + origin_id = self.storage.origin_add_one(self.origin2) + occurrence = self.occurrence2 + occurrence.update({'origin': origin_id}) + self.storage.occurrence_add([occurrence]) + + # when + occ = self.storage.content_find_occurrence( + {'sha1': self.cont2['sha1']}) + + # then + self.assertEquals(occ['origin_type'], self.origin2['type']) + self.assertEquals(occ['origin_url'], self.origin2['url']) + self.assertEquals(occ['branch'], self.occurrence2['branch']) + self.assertEquals(occ['revision'], self.revision2['id']) + self.assertEquals(occ['path'], self.dir2['entries'][0]['name']) + + occ2 = self.storage.content_find_occurrence( + {'sha1_git': self.cont2['sha1_git']}) + + self.assertEquals(occ2['origin_type'], self.origin2['type']) + self.assertEquals(occ2['origin_url'], self.origin2['url']) + self.assertEquals(occ2['branch'], self.occurrence2['branch']) + self.assertEquals(occ2['revision'], self.revision2['id']) + self.assertEquals(occ2['path'], self.dir2['entries'][0]['name']) + + occ3 = self.storage.content_find_occurrence( + {'sha256': self.cont2['sha256']}) + + self.assertEquals(occ3['origin_type'], self.origin2['type']) + self.assertEquals(occ3['origin_url'], self.origin2['url']) + self.assertEquals(occ3['branch'], self.occurrence2['branch']) + self.assertEquals(occ3['revision'], self.revision2['id']) + self.assertEquals(occ3['path'], self.dir2['entries'][0]['name']) + + @istest + def content_find_occurrence_with_non_present_content(self): + # 1. with something that does not exist + missing_cont = self.missing_cont + + occ = self.storage.content_find_occurrence( + {'sha1': missing_cont['sha1']}) + + self.assertEquals(occ, None, + "Content does not exist so no occurrence") + + # 2. with something that does not exist + occ = self.storage.content_find_occurrence( + {'sha1_git': missing_cont['sha1_git']}) + + self.assertEquals(occ, None, + "Content does not exist so no occurrence") + + # 3. with something that does not exist + occ = self.storage.content_find_occurrence( + {'sha256': missing_cont['sha256']}) + + self.assertEquals(occ, None, + "Content does not exist so no occurrence") + + @istest + def content_find_occurrence_bad_input(self): + # 1. with bad input + with self.assertRaises(ValueError): + self.storage.content_find_occurrence({}) # empty is bad + + # 2. with bad input + with self.assertRaises(ValueError): + self.storage.content_find_occurrence( + {'unknown-sha1': 'something'}) # not the right key + class TestStorage(AbstractTestStorage, unittest.TestCase): """Test the local storage""" pass diff --git a/version.txt b/version.txt index c29628fc..bbd24487 100644 --- a/version.txt +++ b/version.txt @@ -1 +1 @@ -v0.0.7-0-gf88331c \ No newline at end of file +v0.0.8-0-g43f5bf9 \ No newline at end of file