Обсуждение: Bug report: Dramatic increase in conflict with recovery afterupgrading 10.2->10.5
Bug report: Dramatic increase in conflict with recovery afterupgrading 10.2->10.5
От
Chris Travers
Дата:
I haven't done significant debugging on this yet but if anyone else has seen this problem and can point me at anything this would be good. Otherwise I believe we will expect to debug and try to fix this problem soon.
After upgrading to PostgreSQL 10.5 from 10.2, we went from one type of query having about one recovery conflict per month to about three or so every two days. The queries in question should usually complete very fast and should never hit the 30 sec max standby delay.
At present I believe this to likely be a regression. But if nobody else knows otherwise, I should know more in a couple days.
--
Best Regards,
Chris Travers
Head of Database
Saarbrücker Straße 37a, 10405 Berlin
Re: Bug report: Dramatic increase in conflict with recovery afterupgrading 10.2->10.5
От
Justin Pryzby
Дата:
On Mon, Sep 10, 2018 at 11:43:47AM +0200, Chris Travers wrote: > At present I believe this to likely be a regression. But if nobody else > knows otherwise, I should know more in a couple days. Do you have query logs or can you send details of the query ? We're not using replication, but I can't help but think of this: https://www.postgresql.org/message-id/flat/20180829140149.GO23024%40telsasoft.com ..since it's effectively a regression WRT reliability (at least if you reindex pg_class). Tom has a patch in HEAD to avoid the issue, but I don't know if there's any plan to release 10.6 until november. See related, earlier thread: https://www.postgresql.org/message-id/flat/12259.1532117714%40sss.pgh.pa.us You could compile your own binaries with Tom's patch applied (f868a81). As you probably know, it's maybe not safe to install PG10.4 binaries on a data dir where you've already upgraded to 10.5 (I believe because data file content might be written which may not be handled correctly by earlier minor release). Justin