Thread: If SyncRepWaitForLSN() fails, would the postgres backend do a roll-back?

If SyncRepWaitForLSN() fails, would the postgres backend do a roll-back?

From
Rui Hai Jiang
Date:
Hi,

I'm researching the synchronous replication. I see the backend of the Primary Server calls SyncRepWaitForLSN() to wait for the Standby Server to write the WAL records.

If some thing happens, such as network failure or disk failure, causes the Standby Server fail to receive WAL records or fail to write WAL records, would the backend of the Primary Server catch and handle this failure? Would the backend roll-back the transactions? Would the backend continue to work?

If the transactions at the Primary Server need rollback, I think SyncRepWaitForLSN() should return some information to the backend.

Thank you,
Rui Hai 
> Hi,
> 
> I'm researching the synchronous replication. I see the backend of the
> Primary Server calls SyncRepWaitForLSN() to wait for the Standby Server to
> write the WAL records.
> 
> If some thing happens, such as network failure or disk failure, causes the
> Standby Server fail to receive WAL records or fail to write WAL records,
> would the backend of the Primary Server catch and handle this failure?
> Would the backend roll-back the transactions? Would the backend continue to
> work?

I don't think the backend would roll back the transaction because when
SyncRepWaitForLSN() is called, the transaction has been already
committed. There's no way to roll back a transaction which has been
committed.

Best regards,
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese:http://www.sraoss.co.jp



Re: If SyncRepWaitForLSN() fails, would the postgres backend do a roll-back?

From
Michael Paquier
Date:
On Wed, Jun 8, 2016 at 7:43 PM, Tatsuo Ishii <ishii@postgresql.org> wrote:
>> Hi,
>>
>> I'm researching the synchronous replication. I see the backend of the
>> Primary Server calls SyncRepWaitForLSN() to wait for the Standby Server to
>> write the WAL records.
>>
>> If some thing happens, such as network failure or disk failure, causes the
>> Standby Server fail to receive WAL records or fail to write WAL records,
>> would the backend of the Primary Server catch and handle this failure?
>> Would the backend roll-back the transactions? Would the backend continue to
>> work?
>
> I don't think the backend would roll back the transaction because when
> SyncRepWaitForLSN() is called, the transaction has been already
> committed. There's no way to roll back a transaction which has been
> committed.

Confirmed. The transaction is not rolled back and is considered
committed locally. Don't count of course on statement_timeout to stop
the transaction from being stuck.
-- 
Michael