| From: | Thomas Kellerer <shammat(at)gmx(dot)net> |
|---|---|
| To: | pgsql-hackers(at)lists(dot)postgresql(dot)org |
| Subject: | Re: Persist MVCC forever - retain history |
| Date: | 2020-07-05 19:08:07 |
| Message-ID: | 80b49c4c-6330-cba0-d7f0-ae83e9f0a790@gmx.net |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
Konstantin Knizhnik schrieb am 05.07.2020 um 19:31:
>> I am surprised that you are saying you didn't feel big interest. My
>> reading of the thread is the opposite, that there was quite some
>> interest, but that there are technical challenges to overcome. So you
>> gave up on that work?
> No, I have not gave up.
> But...
> There are well known problems of proposed approach:
> 1. Not supporting schema changes
> 2. Not compatible with DROP/TRUNCATE
> 3. Presence of large number of aborted transaction can slow down data access.
> 4. Semantic of join of tables with different timestamp is obscure.
Oracle partially solved this (at least 1,3 and 4 - don't know about 3) by storing the old versions in a separate table that is automatically managed if you enable the feature. If a query uses the AS OF to go "back in time", it's rewritten to access the history table.
Thomas
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Paul A Jungwirth | 2020-07-05 19:11:15 | Re: range_agg |
| Previous Message | Stephen Frost | 2020-07-05 17:48:17 | Re: Ideas about a better API for postgres_fdw remote estimates |