Technique for querying date based log data - sql

Question

I have date based log data for financial records. Every time the record changes, a new copy of the record is made in the database.
The current method I am using, which I describe below, is both complex and poor performing. I am dealing with millions of rows and lots of log tables.
The logs are tables in my database that mimic the table we are logging with the addition of a unique log identifier and log date.
For instance, database table RecordLog looks like this:
LogId | RecordId | Log Date | Record Data
--------------------------------------------------------
1 |1 | 2019-07-02 | ...
2 |1 | 2019-05-12 | ...
3 |1 | 2019-03-22 | ...
4 |1 | 2019-01-01 | ...
5 |1 | 2018-08-01 | ...
6 |2 | 2018-01-01 | ...
7 |3 | 2019-01-01 | ...
8 |3 | 2019-02-15 | ...
9 |3 | 2018-10-15 | ...
-The LogId is the log unique id for the RecordLog table, while the RecordId references the unique identifier on the Record table.
-The Record data would mimic the rest of the Record table.
A lot of reporting|analytics occurs based on point in time. For instance, the user wants know the state of affairs at 2019-01-02
In that case we would get these rows since they are the closest recorded instances <= 2019-01-02:
LogId | RecordId | Log Date | Record Data
--------------------------------------------------------
4 |1 | 2019-01-01 | ...
6 |2 | 2018-01-01 | ...
7 |3 | 2019-01-01 | ...
In order to perform these queries now, I am utilizing an inner query.
select * from RecordLog where
...
and ...
and ...
and RecordLog.LogId in (
select max(InnerRecordLog.LogId) from RecordLog as InnerRecordLog
where InnerRecordLog.LogDate <= ?
group by InnerRecordLog.RecordId
order by InnerRecordLog.LogDate desc
)
One of the challenges is I am using HQL to write these queries which limits my access to some native db options

Answer 1

Postgres has a great extension called distinct on which is perfected suited for this:
select distinct on (lr.recordid) rl.*
from recordlog rl
where rl.logdate <= '2019-01-02'
order by lr.recordid, rl.logdate desc;
distinct on (as used here) returns one record per recordid (the keys in parentheses). The specific record is the latest logdate record -- but subject to the where conditions, of course.
In other databases, the most efficient method is usually a correlated subquery:
select rl.*
from recordlog rl
where rl.logdate = (select max(rl2.logdate)
from recordlog rl2
where rl2.recordid = rl.recordid and
rl2.logdate <= '2019-01-02'
);