On 25.12.2019 13:45, Andrey Borodin wrote:
>> 25 дек. 2019 г., в 15:28, Maksim Milyutin <milyutinma@gmail.com> написал(а):
>>
>>> Synchronous replication
>>> does not guarantee that a committed write is actually on any replica,
>>> but it does in general guarantee that a commit has been replicated
>>> before sending a response to the client. That's arguably more
>>> important because the rest of what the application might depend on the
>>> transaction completing and replicating successfully. I don't know of
>>> cases other than cancellation in which a response is sent to the
>>> client without replication when synchronous replication is enabled.
>>
>> Yes, at query canceling (e.g. by timeout from client driver) client receives response about completed transaction
(thoughwith warning which not all client drivers can handle properly) and the guarantee about successfully replicated
transaction*violates*.
> We obviously need a design discussion here to address the issue. But the immediate question is should we add this
topicto January CF items?
+1 on posting this topic to January CF.
Andrey, some fixes from me:
1) pulled out the cancelling of QueryCancelPending from internal branch
where synchronous_commit_cancelation is set so that to avoid dummy
iterations with printing message "canceling the wait for ..."
2) rewrote errdetail message under cancelling query: I hold in this case
we cannot assert that transaction committed locally because its changes
are not visible as yet so I propose to write about locally flushed
commit wal record.
Updated patch is attached.
--
Best regards,
Maksim Milyutin