|From:||Bruce Momjian <bruce(at)momjian(dot)us>|
|To:||Andres Freund <andres(at)anarazel(dot)de>|
|Cc:||Marti Raudsepp <marti(at)juffo(dot)org>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org>|
|Subject:||Re: [PATCH] pg_upgrade fails when postgres/template1 isn't in default tablespace|
|Views:||Raw Message | Whole Thread | Download mbox | Resend email|
On Thu, Sep 3, 2015 at 01:03:30PM +0200, Andres Freund wrote:
> Hi Bruce,
> Are you actually waiting for review in this thread? Or is the CF entry
> more of a reminder?
Oh, I have a commit-fest entry for this? Well, whoever did that was
doing the right thing so things are not forgotten. Yeah! :-)
Anyway, I was going to work on this once I had read my email backlog,
but because of the commit-fest entry, I worked on it today instead.
I have developed the attached patch which fixes the pg_upgrade problem,
and the pg_dumpall problem with postgres and template1 in non-default
tablespaces. I modified the pg_dumpall patch with the following changes:
* It is a general pg_dumpall bug, not specific to binary upgrade mode,
so the new code will be used in non-binary upgrade mode too.
* I hard-coded the "connect template1" string, as it is a constant (no
need for %s and fmtId()).
* You can't mix fprintf() and appendPQExpBuffer() and get the output in
the specified order, i.e. fprintf will come out first. Now, in your
case, it didn't matter, but it isn't clean either.
* I wanted the original database connection to be restored right after
The new output looks like this:
ALTER DATABASE template1 SET TABLESPACE tt;
Based on our previous policy, these are both bugs and cause either
errors or inaccurate restores, so I plan to apply the attached patch to
all back branches.
+ Everyone has their own god. +
|Next Message||Bruce Momjian||2015-09-04 00:11:40||Re: Freeze avoidance of very large table.|
|Previous Message||Tatsuo Ishii||2015-09-03 23:28:19||Re: BRIN INDEX value|