[tarantool-patches] Re: [PATCH 0/5] Delete old WAL files if running out of disk space

Konstantin Osipov kostja at tarantool.org
Tue Oct 16 22:05:22 MSK 2018

* Vladimir Davydov <vdavydov.dev at gmail.com> [18/10/08 13:52]:
> If a replica permanently stops working for some reason, it will pin WAL
> files it would need to resume until it is deleted from the _cluster
> system space or the master is restarted. This happens in production when
> an admin drops a replica and forgets to remove it from the master, and
> this is quite annoying, because it may result in ENOSPC errors on the
> master.

I started benching this patch to check whether fallocate() introduces a 
performance regression and  discovered that there is a general 45% regression
between 1.6 and 1.10.

I hope finally once I have pointed it out and A.Lyapunov has
pointed it out, it will be addressed.

In any case we need to measure fallocate() impact very carefully
before adding it. It seems we make things unnecessarily
complicated all in order to spare the user from spurious ENOSPC,
In my opinion it's a non-goal.

If we get rid of this requirement, we don't need fallocate(), and
the patch could be made simpler in a couple more dimensions.

Please consider making a trivial patch which follows the steps of
the patch by @belyak


Konstantin Osipov, Moscow, Russia, +7 903 626 22 32
http://tarantool.io - www.twitter.com/kostja_osipov

More information about the Tarantool-patches mailing list