[Pacemaker] RFC: Any interesting in 2.0.0 betas?

[Andrew Beekhof]

On Fri, Oct 26, 2012 at 2:52 PM, Vladislav Bogdanov
<bubble at hoster-ok.com> wrote:
> 26.10.2012 04:06, Andrew Beekhof wrote:
>> On Thu, Oct 25, 2012 at 7:42 PM, Vladislav Bogdanov
>> <bubble at hoster-ok.com> wrote:
>>> 25.10.2012 07:50, Andrew Beekhof wrote:
>>>> On Thu, Oct 25, 2012 at 3:08 PM, Vladislav Bogdanov
>>>> <bubble at hoster-ok.com> wrote:
>>>>> 25.10.2012 04:47, Andrew Beekhof wrote:
>>>>>> Does anyone out there have the capacity and interest to test betas of 2.0.0 if I release them?
>>>>>
>>>>> Sure.
>>>>>
>>>>>> If so, for what distro and version?
>>>>>
>>>>> Git tag would be enough for me.
>>>>
>>>> "HEAD" ? :-)
>>>
>>> Yeah, it is usually enough too, but it is better to know that you
>>> consider some revision to be "stable enough" or at least "not broken
>>> very much" so it is tagged. :)
>>
>> The way David and I work these days, is that things stay in our
>> private trees until they are fully baked.
>> Ie. we never intentionally push brokenness into 'master'.
>>
>> So HEAD should be quite usable at any point.
>>
>> But I can also do some tags.
>>
>>>
>>> One issue we discussed earlier - node names in CIB. I saw you did the
>>> move from uname to a consistent resolving after 1.1.8 I currently
>>> evaluating (with my patch I sent earlier), But, I did not see any code
>>> for node name post-processing in case of DNS with FQDN names. So, in
>>> case on DNS-only setup (without /etc/hosts) node names will contain what
>>> reverse DNS lookup returns.
>>
>> Ah, right, that.
>>
>>>
>>> What I would expect to be "right" is to have DNS _domain_ name (what
>>> dnsdomainname returns, actually everything after the first dot) stripped
>>> from them. On the other hand, I saw setups where node names were made
>>> FQDN intentionally to differentiate between clusters in crm_mon output.
>>> Although IMHO it could be convenient for two-node clusters, in case of
>>> 8-16 nodes that output becomes a mess. Actually in my clusters I have
>>> FQDNs longer than 35 characters, with host names of 4-5 chars. And I
>>> feel much better when I do not need to parse two or three lines of text
>>> by eyes just to determine on which node do I have a problem.
>>>
>>> So, I think it would be nice to have a way to affect how cluster node
>>> name is constructed in case of DNS-resolved names without need to patch
>>> code. F.e. "leave as is", "strip after Nth dot", "strip what is in
>>> 'domain' clause in /etc/resolv.conf", "strip what is in Nth part of
>>> 'search' clause in /etc/resolv.conf", "strip what is in
>>> 'totem.cluster_name' corosync parameter", etc. Sure, that should be
>>> consistent over the whole cluster, but that is another story.
>>>
>>
>> Some stripping is fine, but I don't want to overthink it.
>> What about:
>>
>> PCMK_strip_name=N
>
> PCMK_strip_nodenames ?

Possible

>
>>
>> N=0 (default), leave untouched
>> N=1, drop everything after the first dot
>> N=2, drop everything after the second dot
>> ...
>>
>> I'd rather avoid trying to parse /etc/resolv.conf
>
> May be also set it forcibly to uname if uname contains full lexem found
> in dns name?

Run that past me again?