[libvirt] [PATCH] libvirt-guests: emit most messages to stderr

[Martin Kletzander]

On Thu, Jan 07, 2016 at 10:30:49AM +1100, Michael Chapman wrote:
>On Wed, 6 Jan 2016, Martin Kletzander wrote:
>> On Thu, Dec 31, 2015 at 04:53:54PM +1100, Michael Chapman wrote:
>>> Some of the shell functions in this script (e.g. check_guests_shutdown)
>>> produce output that is captured and processed by other parts of the
>>> script. Any error messages from these shell functions must go to stderr,
>>> not stdout.
>>>
>>> Rather than trying to determine which messages are safe for stdout and
>>> which are not (and maintaining this safety as the script is refactored),
>>> simplify things by emitting all messages to stderr, except for the
>>> output from gueststatus(), rh_status() and usage().
>>>
>>
>> This is a good idea, but I just simply changing all outputs to stderr
>> seems wrong.  It doesn't matter for systemd systems, I guess, we output
>> everything to stdout currently and it goes to the logs anyway.  But for
>> other systems it might pollute the output with just informative
>> messages.  For example on my system I would see stderr but not stdout
>> (that goes to its logfile).  What is the reasoning behind this change?
>> Why can't we keep everything on stdout?
>
>I was rebooting a hypervisor with running guests. My libvirt-guests config
>has:
>
>ON_SHUTDOWN=shutdown
>PARALLEL_SHUTDOWN=8
>
>I noticed it was taking an awful long time to reboot. What had happened
>was that it had hit the error path in check_guests_shutdown, so
>shutdown_guests_parallel was stuck looping over:
>
>virsh domid Failed
>virsh domid to
>virsh domid determine
>virsh domid state
>virsh domid of
>virsh domid guest
>...
>
>The error message had been printed to stdout, so it got captured as the
>list of guests still needing to be shutdown.
>
>I could have just fixed that one error case (but how? Not use $(...)
>captures?), but I decided to change them all since it seemed likely that
>this problem might crop up again in the future.
>
>Can you think of an alternative approach? I hadn't considered the case
>of systems redirecting only stdout from initscripts.
>

So sorry for not responding to this for so long.  I would just say keep
all the messages to stdout and output one error message (doesn't have to
be from virsh, just 'echo $msg 2>&1' will suffice) to stderr for each
domain that we failed to stop/start.

I did not mean to hinder the fix by any means, sorry for that once
again.  I'd be more than happy to merge fixes for the script upstream.

Thanks,
Martin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://listman.redhat.com/archives/libvir-list/attachments/20160217/95c8b11f/attachment-0001.sig>