RE: SI messages sent when excuting ROLLBACK PREPARED command
> On Wed, Aug 11, 2021 at 03:14:11PM +0900, Michael Paquier wrote: > > I would just tweak the comment block at the top of what's being > > changed, as per the attached. Please let me know if there are any > > objections. > > And applied as of 710796f. Thanks for your comment and commit. I've changed the patch's commit fest status to 'committed'. https://commitfest.postgresql.org/34/3257/ Regards, Liu
Re: SI messages sent when excuting ROLLBACK PREPARED command
On Wed, Aug 11, 2021 at 03:14:11PM +0900, Michael Paquier wrote: > I would just tweak the comment block at the top of what's being > changed, as per the attached. Please let me know if there are any > objections. And applied as of 710796f. -- Michael signature.asc Description: PGP signature
Re: SI messages sent when excuting ROLLBACK PREPARED command
On Tue, Aug 03, 2021 at 09:29:48AM +, liuhuail...@fujitsu.com wrote: > There was a problem with the before patch when testing. > So resubmit it. FWIW, I see no problems with patch version 1 or 2, as long as you apply patch version 1 with a command like patch -p2. One thing of patch 2 is that git diff --check complains because of a whitespace. Anyway, I also think that you are right here and that there is no need to run this code path with ROLLBACK PREPARED. It is worth noting that the point of Tom about local invalidation messages in PREPARE comes from PostPrepare_Inval(). I would just tweak the comment block at the top of what's being changed, as per the attached. Please let me know if there are any objections. -- Michael diff --git a/src/backend/access/transam/twophase.c b/src/backend/access/transam/twophase.c index 6d3efb49a4..2156de187c 100644 --- a/src/backend/access/transam/twophase.c +++ b/src/backend/access/transam/twophase.c @@ -1520,13 +1520,17 @@ FinishPreparedTransaction(const char *gid, bool isCommit) * Handle cache invalidation messages. * * Relcache init file invalidation requires processing both before and - * after we send the SI messages. See AtEOXact_Inval() + * after we send the SI messages, only when committing. See + * AtEOXact_Inval(). */ - if (hdr->initfileinval) - RelationCacheInitFilePreInvalidate(); - SendSharedInvalidMessages(invalmsgs, hdr->ninvalmsgs); - if (hdr->initfileinval) - RelationCacheInitFilePostInvalidate(); + if (isCommit) + { + if (hdr->initfileinval) + RelationCacheInitFilePreInvalidate(); + SendSharedInvalidMessages(invalmsgs, hdr->ninvalmsgs); + if (hdr->initfileinval) + RelationCacheInitFilePostInvalidate(); + } /* * Acquire the two-phase lock. We want to work on the two-phase callbacks signature.asc Description: PGP signature
RE: SI messages sent when excuting ROLLBACK PREPARED command
Hi, tom > >Hmmm, yeah, I think you're right. It probably doesn't make a big difference > >in > the real world --- anyone who's dependent on the performance of 2PC rollbaxks > is Doing It Wrong. > > But we'd have already done LocalExecuteInvalidationMessage when getting > out of the prepared transaction, so no other SI invals should be needed. > Yes, it does not make any error. > > But for the beginner, when understanding the code, it may make confused. > And for the developer, when developing based on this code, it may make > unnecessary handling added. > > So, I think it is better to optimize the code. > > Here is the patch. There was a problem with the before patch when testing. So resubmit it. Regards, Liu Huailing 0001-Disallow-sending-SI-messages-when-excuting-ROLLBACK.patch Description: 0001-Disallow-sending-SI-messages-when-excuting-ROLLBACK.patch
RE: SI messages sent when excuting ROLLBACK PREPARED command
Hi, tom Thanks for your reply. >Hmmm, yeah, I think you're right. It probably doesn't make a big difference >in the real world --- anyone who's dependent on the performance of 2PC >rollbaxks is Doing It Wrong. > But we'd have already done LocalExecuteInvalidationMessage when getting out > of the prepared transaction, so no other SI invals should be needed. Yes, it does not make any error. But for the beginner, when understanding the code, it may make confused. And for the developer, when developing based on this code, it may make unnecessary handling added. So, I think it is better to optimize the code. Here is the patch. Regards, liuhl -Original Message- From: Tom Lane Sent: Thursday, July 15, 2021 1:36 AM To: Liu, Huailing/刘 怀玲 Cc: pgsql-hack...@postgresql.org Subject: Re: SI messages sent when excuting ROLLBACK PREPARED command "liuhuail...@fujitsu.com" writes: > So, I think we needn't send SI messags when rollbacking the two-phase > transaction. > Or Does it has something special because of two-phase transaction? Hmmm, yeah, I think you're right. It probably doesn't make a big difference in the real world --- anyone who's dependent on the performance of 2PC rollbaxks is Doing It Wrong. But we'd have already done LocalExecuteInvalidationMessage when getting out of the prepared transaction, so no other SI invals should be needed. regards, tom lane twophase.patch Description: twophase.patch
Re: SI messages sent when excuting ROLLBACK PREPARED command
"liuhuail...@fujitsu.com" writes: > So, I think we needn't send SI messags when rollbacking the two-phase > transaction. > Or Does it has something special because of two-phase transaction? Hmmm, yeah, I think you're right. It probably doesn't make a big difference in the real world --- anyone who's dependent on the performance of 2PC rollbaxks is Doing It Wrong. But we'd have already done LocalExecuteInvalidationMessage when getting out of the prepared transaction, so no other SI invals should be needed. regards, tom lane
SI messages sent when excuting ROLLBACK PREPARED command
Hi When reading the code FinishPreparedTransaction, I found that SI messages are sent when executing ROLLBACK PREPARED command. But according to AtEOXact_Inval function, we send the SI messages only when committing the transaction . So, I think we needn't send SI messags when rollbacking the two-phase transaction. Or Does it has something special because of two-phase transaction? Regards