Re: [bug fix] pg_rewind creates corrupt WAL files, and the standby cannot catch up the primary

On Thu, Mar 01, 2018 at 07:49:06AM +0000, Tsunakawa, Takayuki wrote:
> It's a regret that Chen's patch, which limits the WAL to be copied, is
> not committed yet.  It looks good to be ready for committer.

The message I sent provides reasons about why it should not be
integrated.  Particularly since the prior last checkpoint has been
removed in v11, there is always going to be a whole in WAL segments as
you need to create a checkpoint on the ex-standby after it has been
promoted so as its control file data is correctly reflected on disk.

> > > Related to this, shouldn't pg_rewind avoid copying more files and
> > > directories like pg_basebackup?  Currently, pg_rewind doesn't copy
> > > postmaster.pid, postmaster.opts, and temporary files/directories
> > > (pg_sql_tmp/).
> > 
> > Yes, it should not copy those files.  I have a patch in the current CF to
> > do that:
> > https://commitfest.postgresql.org/17/1507/
> 
> Wow, what a great patch.  I think I should look at it.  But I'm afraid
> it won't be backpatched because it's big...

That's a new feature.  This won't get backpatch'ed anyway.

> Even with your patch and Chen's one, my small patch is probably
> necessary to avoid leaving 0-byte or half-baked files.  I'm not sure
> whether those strangely sized files would cause actual trouble, but
> maybe it would be healthy to try to clean things up as much as
> possible.  (files in pg_twophase/ might emit WARNING messages, garbage
> server log files might make the DBA worried, etc.; yes, these may be
> just FUD.) 

Yeah, I'd like to double-check what you are proposing here anyway.
Sorry but I do not have an opinion about what you have sent yet :(
The only thing I am sure of though is that for HEAD not copying files
from pg_wal from the origin is the way to do it.  For back-branches
that's another story.
--
Michael