Re: Replication/backup defaults

On Mon, Jan 2, 2017 at 10:48 AM, Simon Riggs <simon(at)2ndquadrant(dot)com> wrote:

> On 2 January 2017 at 09:39, Magnus Hagander <magnus(at)hagander(dot)net> wrote:
>
> > Please do submit a patch for it.
>
> The way this is supposed to go is someone submits a patch and they
> receive feedback, then act on that feedback. If I was able to get away
> with deflecting all review comments with a simple "you fix it if you
> don't like it" there would be considerably more patches with my name
> on it accepted, but probably no further forward in real terms because
> of the loose ends it creates.
>

Fair enough.

It's just that people keep saying that this is easy, and have said so for a
long time, but nobody has written a patch for it.

> In this case, simply changing the default will remove a whole class of
> performance optimization that we have educated people to expect. I'm
> sorry to point this out but removing that will cause many real changes
> for people's systems that we will not be thanked for, even though I
> understand your reasoning and wish the same goals to be achieved.
>

My claim here is that a lot *fewer* people have come to expect this
performance optimization, than would (quite reasonably) expect that backups
should work on a system without taking it down for restart to reconfigure
it to support that.

I run into that all the time. I hear complaints about that all the time. I
have not heard a single user complain about performance loss after enabling
backups.

And how many people that rely on this optimization don't do any *other*
optimization on their system *anyway*, that would cause them to require a
restart anyway? It's not like we're taking away their ability to enable the
optimization, it's just not on by default.

> I'm willing to assist in a project to allow changing wal_level online
> in this release. Please let's follow that path.
>

Sure thing, I will be happy to help test and review such a patch.

I will still argue that the *default* should be wal_level=replica though.
Because once we have such a patch, it's trivial to re-enable this
performance optimization (at the cost of backups and replication).

//Magnus