[zfs-discuss] ZFS update best practices

Chris Siebenmann cks at cs.toronto.edu
Thu Feb 15 12:36:06 EST 2018

> Does anyone know how much internal testing is put on canditate
> releases?

 For a while, ZFS on Linux has had a relatively extensive test suite
(and new features almost invariably come with new tests). You can see
this in the source repository here:

As far as I know, this test suite is run against essentially all
changes to ZoL (and changes are only accepted if all the tests
still pass). These tests should insure that we're covered against
anything obviously broken, although I don't know how much they can
test against subtle and infrequent bugs like race conditions.

(Fixes for race conditions have to be made every so often. But
some of these are actually upstream bugs that eventually get
exposed and found by someone; for example, today a fix for this
one went into ZoL: https://www.illumos.org/issues/8857 )

 I don't know if the process of deciding on release candidates has
additional formal tests involved, although it may well. Also I don't
know if there are performance regression tests run before release;
as far as I know the test suite is just there to look for bugs and
things working correctly.

 A number of people run development versions of ZFS on various systems
(I do on my office workstation, for example). These people serve as
some additional checking over and above the automated test suite. I
don't know if anyone does canary deploys of release candidates into
near-production environments, though.

 The thing that probably gets the least testing in general is ZoL's
hooks into the init system to do things like bring up pools on reboot.
There are a lot of different environments for this, all somewhat subtly
different (System V init, systemd on Ubuntu, systemd on RHEL/CentOS,
etc), and various different scenarios (install from scratch, upgrade
from a recent version, upgrade from an old version), all of which
makes comprehensive testing hard. I know that there have been problems
with this in the past, and I think some of them have even made it into
released versions.

(Since I run versions from the development tree in a hacked up
environment on Fedora, I'm not sure what has made it into releases and
what happens in more standard and normal environments.)

	- cks

More information about the zfs-discuss mailing list