Re: TAP test to cover "EndOfLogTLI != replayTLI" case

On Tue, Jan 18, 2022 at 10:31 AM Kyotaro Horiguchi
<horikyota(dot)ntt(at)gmail(dot)com> wrote:
>
> At Tue, 18 Jan 2022 12:25:15 +0800, Julien Rouhaud <rjuju123(at)gmail(dot)com> wrote in
> > Hi,
> >
> > On Mon, Jan 17, 2022 at 09:33:57PM +0800, Julien Rouhaud wrote:
> > >
> > > Thanks for the updated patch! Note that thanks to Andres and Thomas work, you
> > > can now easily rely on the exact same CI than the cfbot on your own github
> > > repository, if you need to debug something on a platform you don't have access
> > > to. You can check the documentation at [1] for more detail on how to setup the
> > > CI.
> >
> > The cfbot reports that the patch still fails on Windows but also failed on
> > macos with the same error: https://cirrus-ci.com/task/5655777858813952:
> >
> > [14:20:43.950] # Failed test 'check standby content before timeline switch 0/500FAF8'
> > [14:20:43.950] # at t/003_recovery_targets.pl line 239.
> > [14:20:43.950] # got: '6000'
> > [14:20:43.950] # expected: '7000'
> > [14:20:43.950] # Looks like you failed 1 test of 10.
> >
> > I'm switching the CF entry to Waiting on Author.
>
> The most significant advantages of the local CI setup are
>
> - CI immediately responds to push.
>
> - You can dirty the code with additional logging aid as much as you
> like to see closely what is going on. It makes us hesitant to do
> the same on this ML:p
>

Indeed :)

I found the cause for the test failing on window -- is due to the
custom archive command setting which wasn't setting the correct window
archive directory path.

There is no option to choose a custom wal archive and restore patch in
the TAP test. The attach 0001 patch proposes the same, which enables
you to choose a custom WAL archive and restore directory. The only
concern I have is that $node->info() will print the wrong archive
directory path in that case, do we need to fix that? We might need to
store archive_dir like base_dir, I wasn't sure about that. Thoughts?
Comments?

Regards,
Amul