Re: sequences vs. synchronous replication

From: Fujii Masao <masao(dot)fujii(at)oss(dot)nttdata(dot)com>
To: Tomas Vondra <tomas(dot)vondra(at)enterprisedb(dot)com>, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com>
Cc: PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>
Subject: Re: sequences vs. synchronous replication
Date: 2021-12-22 17:50:49
Message-ID: 5774a616-8e87-d7a5-7856-6db0ff1a78b0@oss.nttdata.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On 2021/12/22 21:11, Tomas Vondra wrote:
> Interesting idea, but I think it has a couple of issues :-(

Thanks for the review!

> 1) We'd need to know the LSN of the last WAL record for any given sequence, and we'd need to communicate that between backends somehow. Which seems rather tricky to do without affecting performance.

How about using the page lsn for the sequence? nextval_internal() already uses that to check whether it's less than or equal to checkpoint redo location.

> 2) SyncRepWaitForLSN() is used only in commit-like situations, and it's a simple wait, not a decision to write more WAL. Environments without sync replicas are affected by this too - yes, the data loss issue is not there, but the amount of WAL is still increased.

How about reusing only a part of code in SyncRepWaitForLSN()? Attached is the PoC patch that implemented what I'm thinking.

> IIRC sync_standby_names can change while a transaction is running, even just right before commit, at which point we can't just go back in time and generate WAL for sequences accessed earlier. But we still need to ensure the sequence is properly replicated.

Yes. In the PoC patch, SyncRepNeedsWait() still checks sync_standbys_defined and uses SyncRepWaitMode. But they should not be checked nor used because their values can be changed on the fly, as you pointed out. Probably SyncRepNeedsWait() will need to be changed so that it doesn't use them.

> 3) I don't think it'd actually reduce the amount of WAL records in environments with many sessions (incrementing the same sequence). In those cases the WAL (generated by in-progress xact from another session) is likely to not be flushed, so we'd generate the extra WAL record. (And if the other backends would need flush LSN of this new WAL record, which would make it more likely they have to generate WAL too.)

With the PoC patch, only when previous transaction that executed nextval() and caused WAL record is aborted, subsequent nextval() generates additional WAL record. So this approach can reduce WAL volume than other approach?

In the PoC patch, to reduce WAL volume more, it might be better to make nextval_internal() update XactLastRecEnd and assign XID rather than emitting new WAL record, when SyncRepNeedsWait() returns true.

Regards,

--
Fujii Masao
Advanced Computing Technology Center
Research and Development Headquarters
NTT DATA CORPORATION

Attachment Content-Type Size
sequence_log.patch text/plain 2.2 KB

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tomas Vondra 2021-12-22 18:49:15 Re: sequences vs. synchronous replication
Previous Message Andrey Lepikhov 2021-12-22 17:01:22 Re: Clarifying/rationalizing Vars' varno/varattno/varnoold/varoattno