Il 27/02/2014 15:11, Ademar Reis ha scritto:
>> I only update when I have a particularly pressing itch that requires
>> me to cook a patch, or when I have to test some new guest that has
>> just been added.  Roughly it's every 2-3 months.  All I know is that
>> when "next" existed, using master just worked, while right now I
>> cannot even use stable releases because they're just as unstable as
>> master.  This is what made me worry about declining quality of
>> releases.
> 
> You're being bitten by the timing: with the split into test
> providers, this is probably the worst time ever to evaluate the
> stability of virt-test.

Not really.  The split caused absolutely zero problems to me, actually.
Look at the patches that I asked to revert or fix (or that I have local
workarounds for):

  guest-os: Change kickstart/unattended parameters passed to kernel
  Date:   Thu May 16 11:29:49 2013 -0300

  shared.unattended: include some autotest requried pkgs
**Date:   Tue Jul 30 15:09:03 2013 +0800

  virttest.remote: Update password and login prompt in handle_prompts function
  Date:   Mon Sep 23 16:01:03 2013 +0800

  Fix unattended_install.url on Fedora guests
  Date:   Thu Aug 22 18:43:18 2013 -0300

  add scsi-target-utils pkg to support migrate.after_extensive_io.iscsi
**Date:   Thu Oct 24 13:45:57 2013 +0800

  Unify Windows unattended files
**Date:   Wed Dec 18 18:05:59 2013 +0800

  run: Don't overwrite monitors when using --monitor
**Date:   Wed Jan 15 09:11:27 2014 +0100

None of them date to after the split, and I've marked those that
should have been detected by a run that is roughly "only
Windows.2008.r2, Windows.7, RHEL.5, RHEL.6, Fedora.20; only ide;
only rtl8139; only unattended_install.cdrom; only qcow2".

On top of this, the whole image-backup/restore changes are hopelessly broken
for me; I run autotest from an NFS share, and it is not fun to move 3 GB from
storage to workstation and back to storage on 100Mbit network.  I haven't yet
looked into them, so I am just disabling locally in my copy of autotest.

> In the long term, what you probably want is to use a stable
> release (there should be new releases every week or so). 5% of
> the time you should run master to submit a patch or get a recent
> fix before a release is made.

This is unfortunately not the case if a breakage is introduced once every
1-2 months, and I update once 2-3 months.  I learnt that it's simpler to
just jump to origin/master.

Paolo

_______________________________________________
Virt-test-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/virt-test-devel

Reply via email to