Re: In-placre persistance change of a relation - Mailing list pgsql-hackers

From Kyotaro Horiguchi
Subject Re: In-placre persistance change of a relation
Date
Msg-id 20210325.141503.245781197327278199.horikyota.ntt@gmail.com
Whole thread Raw
In response to Re: In-placre persistance change of a relation  (Kyotaro Horiguchi <horikyota.ntt@gmail.com>)
Responses RE: In-placre persistance change of a relation  (Jakub Wartak <Jakub.Wartak@tomtom.com>)
List pgsql-hackers
At Thu, 25 Mar 2021 14:08:05 +0900 (JST), Kyotaro Horiguchi <horikyota.ntt@gmail.com> wrote in 
> (I'm not sure when the subject was broken..)
> 
> At Thu, 14 Jan 2021 17:32:17 +0900 (JST), Kyotaro Horiguchi <horikyota.ntt@gmail.com> wrote in 
> > Commit bea449c635 conflicts with this on the change of the definition
> > of DropRelFileNodeBuffers. The change simplified this patch by a bit:p
> 
> In this version, I got rid of the "CLEANUP FORK"s, and added a new
> system "Smgr marks".  The mark files have the name of the
> corresponding fork file followed by ".u" (which means Uncommitted.).
> "Uncommited"-marked main fork means the same as the CLEANUP2_FORKNUM
> and uncommitted-marked init fork means the same as the CLEANUP_FORKNUM
> in the previous version.x
> 
> I noticed that the previous version of the patch still leaves an
> orphan main fork file after "BEGIN; CREATE TABLE x; ROLLBACK; (crash
> before checkpoint)" since the "mark" file (or CLEANUP2_FORKNUM) is
> revmoed at rollback.  In this version the responsibility to remove the
> mark files is moved to SyncPostCheckpoint, where main fork files are
> actually removed.

For the record, I noticed that basebackup could be confused by the
mark files but I haven't looked that yet.

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center



pgsql-hackers by date:

Previous
From: Kyotaro Horiguchi
Date:
Subject: Re: In-placre persistance change of a relation
Next
From: Ashutosh Bapat
Date:
Subject: Decoding speculative insert with toast leaks memory