[libvirt] [libvirt-guests] New feature ALWAYS_START

Wed Sep 23 13:12:18 UTC 2015

On Wed, Sep 23, 2015 at 11:39:29AM +0200, Marek Lukács wrote:
>Hi Martin,
>
>thank you for your reply. Please check my comments.
>
>On Wed, Sep 23, 2015 at 10:50 AM, Martin Kletzander <mkletzan at redhat.com>
>wrote:
>
>> On Tue, Sep 22, 2015 at 07:33:30AM +0200, Marek Lukács wrote:
>>
>>> Hi,
>>>
>>> It will be nice feature to have configuration option
>>> ALWAYS_START="$uri:$name $uri:name ..." in libvirt-guests
>>> configuration file.
>>>
>>> If ON_BOOT is "start" and if ALWAYS_START is not empty, it iterates
>>> over the ALWAYS_START and starts guests with same conditions (delays
>>> etc.) before it starts guests from LISTFILE.
>>>
>>>
>> To be honest, I don't think that's _exactly_ what you want _just_ from
>> libvirt itself; let me explain.
>>
>> Benefits:
>>> - guests are started with delays
>>>
>>
>> Delays that are done due to the guests are not something we should
>> handle.  Guests and mainly the applications inside them should handle
>> this gracefully.  Just delaying the starts is still error-prone.
>>
>
>I fully understand this and fully agree with you. Production application
>has to handle this gracefully. But still I see a space for doing it in
>environments where it is not so necessary to configure application so
>gracefully, like development or testing environments. Delays are fine for
>many not production environments, where applications are not in production
>state.
>
>
>> - guests are started after host failure
>>>
>>
>> That's what libvirt-guests does already.  And if you want some domains
>> to be started on every start, there's the 'autostart' parameter for
>> domains.
>>
>
>No, libvirt-guests only starts those domains which has been running before
>"service" libvirt-guests has been stopped. In case of host failure, there
>is no LISTFILE in filesystem, as it has not been generated by
>libvirt-guests stop mechanism.
>

Oh, so here is what I have misunderstood.  Well, just misread, of
course libvirt-guests doesn't handle host failures.

>But probably I do not understand "script" libvirt-guests correctly and why
>it is in libvirt. I will be happy, if you will give me more details, why
>there is this script, even if libvirt has 'autostart' parameter for domains.
>
>- For what usage is libvirt-guests designed?

My interpretation is that it is there in order for you to be able to
shutdown and boot later without losing the machines you were running.

Autostart says that particular domain should be started every time the
daemon is started.  Basically says that particular guest should be
running on the host all the time.

>- Why it supports delays?

Because most of the time you'll want to resume from managed save and
you might cause a big load in case you're starting bunch of machines
because all of those will start loading everything from disk and so
on.

>- Why to have libvirt-guests if there is 'autostart' domain parameter?

It does two different things.  Domain with autostart will be started
every time daemon (actually not even the computer) is started, but
libvirt-guests will stop/save domains that are running when the
computer is being shut down and start/resume them when it is starting
back.

>- For who is libvirt-guests and who should use 'autostart' domain parameter?
>

Well, libvirt-guests was added by a guy who was too lazy to clean up
his machine before rebooting (sorry Jirka, I had to).  But you get the
picture.  Use it for whatever you like and whatever suits you.  I'm
not even against adding what you suggested, I just wanted to make sure
you're not relying on the script for something critical as my
understanding of it that it is not very error prone.

>
>> - guests are started in specific order (for example complex
>>> environment, when DB should be started before other guest, etc.)
>>>
>>>
>> Again, same as the first point.  This should be handled gracefully in
>> the application itself or at least worked around in the guest (not
>> starting DB-related app before DB is accessible).
>>
>> Anyway, if you *really* want this, then the easiest thing to do is
>> creating a service that starts before libvirt-guests, but after
>> libvirt, which is just something similar to "local", so it just runs a
>> script that does:
>>
>>  for i in domain_one some_other_domain database_dom
>>  do
>>    virsh start "$i"
>>    sleep 60 # or you can try connecting to make sure it started
>>  done
>>
>> or something similar.  However, you might still propose a simple patch
>> for the feature you described.
>>
>
>For me it is no problem to design my own script to handle my needs. I have
>spent some time googling, if there is already a tool for it. I found only
>similar questions, so I got feeling, that I am not the only one with
>similar requirements.
>
>Script libvirt-guests in my eyes handles very similar task. It starts
>domains with delays, it starts domains what has been running at previous
>stop, but do not handle situation in case of host failure and do not starts
>domain in specific order. I prefer and I think, that it is better to not
>create new script no one knows about, but to modify existing one everybody
>knows about. But again, maybe I do not understand why and for what
>libvirt-guests is.
>

Well, there is no reference purpose and as I said I'm not even against
adding your option there.  And if we get it into libvirt-guests,
others might use it.  Would you mind proposing it as a patch?  It
could be pretty straightforward, I guess.

Have a nice day,
Martin

>Anyway, much easier in case of testing and development environments is to
>set start sequence in /etc/rc.local and forget about script like
>libvirt-guests, if it does not have features I described.
>
>
>> Regards,
>>>
>>> Marek Lukács
>>>
>>> --
>>> libvir-list mailing list
>>> libvir-list at redhat.com
>>> https://www.redhat.com/mailman/listinfo/libvir-list
>>>
>>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/libvir-list/attachments/20150923/00dc8217/attachment-0001.sig>