public inbox for [email protected]  

help / color / mirror / Atom feed

From: Tomas Vondra <[email protected]>
To: Fujii Masao <[email protected]>
To: Amit Kapila <[email protected]>
Cc: PostgreSQL Hackers <[email protected]>
Subject: Re: sequences vs. synchronous replication
Date: Wed, 22 Dec 2021 13:11:54 +0100
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>
References: <[email protected]>
	<CAA4eK1KN-Pqe1y5w0uXU2Ttn8DK4WfqgwAbCRbcDt4yzQ88wRA@mail.gmail.com>
	<[email protected]>
	<[email protected]>



On 12/22/21 05:56, Fujii Masao wrote:
> 
> 
> On 2021/12/22 10:57, Tomas Vondra wrote:
>>
>>
>> On 12/19/21 04:03, Amit Kapila wrote:
>>> On Sat, Dec 18, 2021 at 7:24 AM Tomas Vondra
>>> <[email protected]> wrote:
>>>>
>>>> while working on logical decoding of sequences, I ran into an issue 
>>>> with
>>>> nextval() in a transaction that rolls back, described in [1]. But after
>>>> thinking about it a bit more (and chatting with Petr Jelinek), I think
>>>> this issue affects physical sync replication too.
>>>>
>>>> Imagine you have a primary <-> sync_replica cluster, and you do this:
>>>>
>>>>     CREATE SEQUENCE s;
>>>>
>>>>     -- shutdown the sync replica
>>>>
>>>>     BEGIN;
>>>>     SELECT nextval('s') FROM generate_series(1,50);
>>>>     ROLLBACK;
>>>>
>>>>     BEGIN;
>>>>     SELECT nextval('s');
>>>>     COMMIT;
>>>>
>>>> The natural expectation would be the COMMIT gets stuck, waiting for the
>>>> sync replica (which is not running), right? But it does not.
>>>>
>>>
>>> How about if we always WAL log the first sequence change in a 
>>> transaction?
>>>
>>
>> I've been thinking about doing something like this, but I think it 
>> would not have any significant advantages compared to using 
>> "SEQ_LOG_VALS 0". It would still have the same performance hit for 
>> plain nextval() calls, and there's no measurable impact on simple 
>> workloads that already write WAL in transactions even with 
>> SEQ_LOG_VALS 0.
> 
> Just idea; if wal_level > minimal, how about making nextval_internal() 
> (1) check whether WAL is replicated to sync standbys, up to the page lsn 
> of the sequence, and (2) forcibly emit a WAL record if not replicated 
> yet? The similar check is performed at the beginning of 
> SyncRepWaitForLSN(), so probably we can reuse that code.
> 

Interesting idea, but I think it has a couple of issues :-(

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.

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.

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.

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.)


So I don't think this would actually help much.


regards

-- 
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

view thread (78+ messages)  latest in thread

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients using the --to and --cc options:
  reply via email

  To: [email protected]
  Cc: [email protected], [email protected], [email protected], [email protected]
  Subject: Re: sequences vs. synchronous replication
  In-Reply-To: <[email protected]>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox