Re: Recent eelpout failures on 9.x branches

On Tue, Dec 01, 2020 at 06:07:17PM -0500, Tom Lane wrote:
> Thomas Munro <thomas.munro@gmail.com> writes:
> > Unfortunately, eelpout got kicked off the nice shiny ARM server it was
> > running on so last week I moved it to a rack mounted Raspberry Pi.  It
> > seems to be totally I/O starved causing some timeouts to be reached,
> > and I'm looking into fixing that by adding fast storage.  This may
> > take a few days.  Sorry for the noise.
> 
> Ah-hah.  Now that I look, eelpout is very clearly slower overall
> than it was a couple weeks ago, so all is explained.
> 
> It might still be reasonable to raise wal_sender_timeout in the
> buildfarm environment, though.  We usually try to make sure that
> buildfarm timeouts border on ridiculous, not just because of
> underpowered critters but also for cases like CLOBBER_CACHE_ALWAYS
> animals.

My buildfarm animals override these:

    extra_config =>{
        DEFAULT => [
            "authentication_timeout = '600s'",
            "wal_receiver_timeout = '18000s'",
            "wal_sender_timeout = '18000s'",
        ],
    },
    build_env =>{
        PGCTLTIMEOUT => 18000,
    },

Each of those timeouts caused failures before I changed it.  For animals fast
enough to make them irrelevant, I've not yet encountered a disadvantage.

> I'm also wondering a bit why the issue isn't affecting the newer
> branches.  It's certainly not because we made the test shorter ...

That is peculiar.