[systemd-devel] [IDEA] systemd as basis for HA clusters

[Colin Guthrie]

'Twas brillig, and Lennart Poettering at 25/07/13 20:59 did gyre and gimble:
> On Thu, 25.07.13 21:21, Tomasz Torcz (tomek at pipebreaker.pl) wrote:
> 
>> On Thu, Jul 25, 2013 at 06:51:21PM +0200, Lennart Poettering wrote:
>>> On Fri, 19.07.13 21:05, Pablo Nehab Hess (pablo at hess.net.br) wrote:
>>>
>>>> Hi all,
>>>>
>>>> I was wondering how much systemd could add to current high
>>>> availability cluster setups.
>>>>
>>>> Today systemd is used on HA clusters as just an init replacement.
>>>> However, there are systemd features that might come in handy and
>>>> improve the overall performance and even reliability of such clusters:
>>>>
>>>> * watchdog functionality as in
>>>> <http://0pointer.de/blog/projects/watchdog.html> is the most evident
>>>> feature;
>>>> * tcp-based dbus communication could be used to exchange information
>>>> between cluster members;
>>>>
>>>> Also, I believe systemd functionality could be extended so it would
>>>> take into consideration other nodes' systemd instances in order to
>>>> make sure each service is always alive somewhere -- call it "floating
>>>> units" if you will. :-)
>>>>
>>>> Does this idea even make sense? Is it too "one systemd to rule them all"?
>>>
>>> Well, I don't really know what exactly HA clusters would need. However,
>>> note that we actually do try to draw the line somewhere where systemd
>>> ends... I have the suspicion the HA cluster stuff something which could
>>> make great use of systemd's comprehensive bus interfaces, but I am not
>>> convinced such a project should sit in systemd itself.
>>
>>   The RH Cluster suite cares about running services, and restarting services
>> when they fail.  Just like systemd.  Main difference is that you can select
>> on which host to run this service.  
> 
> Well we have "-H" already. What more do you need?
> 
>> It could be implemented in some daemon
>> synchronising state between remote systemd's.
>>   Every time I use ”clustat” I feel like I'm should be looking at
>> ”systemctl status” listing.  When I enable services using ”clusvcadm -e foo”
>> I feel like it should be ”systemctl enable foo”.  There's a quite big
>> overlap between cluster suite and systemctl.
> 
> So these tools will walk the cluster and get the status of the
> respective service on all machines?

Not sure if it's a walk as such, but the outcome is exactly that yes.

I've not used clustat and such for a while tho', so am a bit rusty.

Col


-- 

Colin Guthrie
gmane(at)colin.guthr.ie
http://colin.guthr.ie/

Day Job:
  Tribalogic Limited http://www.tribalogic.net/
Open Source:
  Mageia Contributor http://www.mageia.org/
  PulseAudio Hacker http://www.pulseaudio.org/
  Trac Hacker http://trac.edgewall.org/