Plan for development of LXD database

[Oliver Schad]

Hi everybody,

we had that several times, that the LXD database broke cause of full
disk. The restore procedure is a bit hard, we setup everything from
scratch.

So my question is if others have an easy to use solution?

Are there plans around to make the database immune against
corruption in case of full disk?

Best Regards
Oli

--
Automatic-Server AG •••••
Oliver Schad
Geschäftsführer
Turnerstrasse 2
9000 St. Gallen | Schweiz

www.automatic-server.com | oliver...@automatic-server.com
Tel: +41 71 511 31 11 | Mobile: +41 76 330 03 47

[Stéphane Graber]

We actually have automated tests to validate recovery from such
situations but that's focused on single node usage.
If you are using it in a cluster, then there are still some gaps that
we're looking to fill and improve testing on.

Mathieu on my team who drives dqlite development (the database used by
LXD) has been quite busy hunting down bugs recently, making things
significantly more reliable.

We may be able to get a better idea of what may be affecting you if we knew:
- What version of LXD it is
- Whether it's a single node or cluster
- What the host filesystem is
- Whether this is using the snap package or some other type of
packaging (mostly to figure out what version of raft/dqlite you're
running)

It's also worth noting that we merged support for compression of the
dqlite snapshots recently. In our tests, we've seen up to a 90%
reduction in size for the on-disk database snapshots. That's not
really likely to help you much if the LXD database is already tiny
(and it often is) though.

Stéphane

> --
> You received this message because you are subscribed to the Google Groups "lxc-devel" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to lxc-devel+...@lists.linuxcontainers.org.
> To view this discussion on the web visit https://groups.google.com/a/lists.linuxcontainers.org/d/msgid/lxc-devel/20210604114734.33b19e1f%40flunder.