[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Cluster-devel] skipping unused services, cman_tool join -A

On 27/01/10 21:38, David Teigland wrote:
On Wed, Jan 27, 2010 at 01:51:38PM -0700, Steven Dake wrote:
On Wed, 2010-01-27 at 14:49 -0600, David Teigland wrote:
On Fri, Sep 25, 2009 at 02:27:52PM -0500, David Teigland wrote:
To avoid loading+running services that you don't use (e.g. to avoid bugs
crashing the system from a service you're not using)

add to cluster.conf

<service name="corosync_cman" ver="0"/>
<service name="openais_ckpt" ver="0"/>

run cman_tool join -A

(or set CMAN_JOIN_OPTS env var)

Currently cman_tool loads the list of services defined by
"openaisserviceenablestable", which is clm, evt, ckpt, msg, lck, tmr.

I suggest that by default cman_tool only load the services that we use,
namely ckpt (and cman of course).  If someone wants to use another of the
services, they can use cluster.conf<service>  to load it.

I don't think this suggestion went over very well last time I brought it
up, but recently the trend has been shifting toward narrowing our exposure
to things we've tested... so I'm tossing it out one more time.

Unfortunately this is not possible because it would break rolling
upgrades as I have stated in the past.

cman makes config adjustments upon seeing<cman upgrading="yes"/>, could
it not do that for services?

So, let me get this straight. For a cman upgrade you want to load the same services as before. But for a new full-featured RHEL6 cluster suite you think we should load fewer services, thus removing features ?

To me it seems like a pointless bit of added complication, for us and users. If the service is still to be shipped then what do we gain by simply making it more difficult to use? So far Perry's argument is that we don't *ship* things we can't support, thus preventing people from using them at all unless they compile it themselves. But this is different in that we have to ship the openais modules anyway, so disabling them is merely being wilfully obscure. It's a bit like shipping a command-line tool that was in RHEL5, but on RHEL6 insists on having the first parameter "PLEASE" before it will do anything.

I'm not really passionately in favour of either way, I just don't see any benefit to us, or (more importantly) to customers, from changing it.


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]