SQL Query to get the last change among many tables - sql

Question

Given three related tables I would like to find out for given filter criteria what was the last change date and who did these changes.
Example: Three tables, where A <1--n> B <1--n> C
CREATE TABLE table_a
(
a_id bigint not null,
some_data varchar(255),
last_changed_at timestamp not null,
last_changed_by_id bigint not null,
primary key (a_id)
);
CREATE TABLE table_b
(
b_id bigint not null,
a_id bigint not null,
some_data varchar(255),
last_changed_at timestamp not null,
last_changed_by_id bigint not null,
primary key (b_id)
);
CREATE TABLE table_c
(
c_id bigint not null,
b_id bigint not null,
a_id bigint not null,
some_data varchar(255),
last_changed_at timestamp not null,
last_changed_by_id bigint not null,
primary key (c_id)
);
alter table table_b add constraint table_a_fk foreign key (a_id) references table_a;
alter table table_c add constraint table_a_fk foreign key (a_id) references table_a;
alter table table_c add constraint table_b_fk foreign key (b_id) references table_b;
We would like to see among all three tables (table_a, table_b, table_c) what was the last range done related to a_id and by whom given conditions on table_a (e.g. WHERE table_a.some_data like 'abc%').
As we are using Hibernate Envers for all three tables we have also audit tables available defined as:
CREATE TABLE table_a_aud
(
rev bigint not null,
revtype smallint not null,
revend bigint not null,
a_id bigint not null,
some_data varchar(255),
last_changed_at timestamp not null,
last_changed_by_id bigint not null,
primary key (a_id, rev)
);
CREATE TABLE table_b_aud
(
rev bigint not null,
revtype smallint not null,
revend bigint not null,
b_id bigint not null,
a_id bigint not null,
some_data varchar(255),
last_changed_at timestamp not null,
last_changed_by_id bigint not null,
primary key (b_id, rev)
);
CREATE TABLE table_c_aud
(
rev bigint not null,
revtype smallint not null,
revend bigint not null,
c_id bigint not null,
b_id bigint not null,
a_id bigint not null,
some_data varchar(255),
last_changed_at timestamp not null,
last_changed_by_id bigint not null,
primary key (c_id, rev)
);
CREATE TABLE revinfo
(
rev bigint not null,
revtstmp bigint,
last_changed_at timestamp not null,
last_changed_by_id bigint not null,
primary key (rev)
);
alter table table_a_aud add constraint revinfo_a_fk foreign key (rev) references revinfo;
alter table table_b_aud add constraint revinfo_b_fk foreign key (rev) references revinfo;
alter table table_c_aud add constraint revinfo_c_fk foreign key (rev) references revinfo;
I have tried the following variants, but I am not sure if one of these is the best approach.
Union with all three tables, window function to pick the newest.
Union with all three revision tables (have much more data), window function to pick the newest
Create a new table with the latest changed by a_id, last_changed_at, last_changed_by_id filled with either (a) a trigger or (b) by our application.
Variant 1 may look like:
SELECT tab_a.*, lc.last_changed_at, lc.last_changed_by_id
FROM table_a tab_a
JOIN (
(SELECT a_id, last_changed_at, last_changed_by_id
FROM (
SELECT a_id,
last_changed_at,
last_changed_by_id,
ROW_NUMBER() OVER (PARTITION BY unioned.a_id ORDER BY unioned.last_changed_at DESC) AS rk
FROM (
(SELECT a_id, last_changed_at, last_changed_by_id FROM table_a)
UNION ALL
(SELECT a_id, last_changed_at, last_changed_by_id FROM table_b)
UNION ALL
(SELECT a_id, last_changed_at, last_changed_by_id FROM table_c)) unioned) unioned_with_rank
WHERE unioned_with_rank.rk = 1)) lc ON tab_a.a_id = lc.a_id
WHERE tab_a.some_data like 'abc%';
Although adding indexes on table_a, table_b and table_c Postgres is still doing a seq scan on all tables. Union and then using a window function may not be the best from a performance perspective.
Variant 2 may look like:
SELECT tab_a.*, lc.last_changed_at, lc.last_changed_by_id
FROM table_a tab_a
JOIN (
SELECT a_id, MAX(rev) rev
FROM (SELECT a_id, rev
FROM table_a_aud
UNION ALL
SELECT a_id, rev
FROM table_b_aud
UNION ALL
SELECT a_id, rev
FROM table_c_aud) unioned
GROUP BY a_id) uniongrouped ON tab_a.a_id = uniongrouped.a_id
JOIN revinfo_ lc ON uniongrouped.rev = lc.rev;
WHERE tab_a.some_data like 'abc%';
We have still a union, but we could have a sorted index on (a_id, rev DESC) which might help. But *_aud tables are containing > 10 x more data than the other tables. IMO this may still be the better approach to use this one.
Variant 3
Implemented in our application, no special database query needed. Result can be a simple join:
SELECT tab_a.*, lc.last_changed_at, lc.last_changed_by_id
FROM table_a tab_a JOIN last_change lc ON tab_a.a_id = lc.a_id
This would make the query statements much faster, but we need to distribute update code all over the application code, which I don't like. Also, I see no trigger which can be executed shortly before the transaction ends and only adding one insert/update statement instead of multiples if many entries have been added/updated on tables table_a, table_b and table_c in the same transaction.
Does anyone has done something similar?
What is the best approach to solving this requirement?
Is there a generic approach that works for PostgreSQL, SQL Server and maybe H2 and Oracle?