[Pacemaker] Stonith setup hostname params

Andrew Beekhof andrew at beekhof.net
Fri Apr 1 09:05:29 UTC 2011


On Fri, Apr 1, 2011 at 10:45 AM, Dejan Muhamedagic <dejanmm at fastmail.fm> wrote:
> On Fri, Apr 01, 2011 at 09:41:26AM +0200, Andrew Beekhof wrote:
>> On Tue, Mar 29, 2011 at 3:28 PM, Dejan Muhamedagic <dejanmm at fastmail.fm> wrote:
>> > On Mon, Mar 28, 2011 at 09:08:32PM +0400, Pavel Levshin wrote:
>> >> 28.03.2011 18:35, Dejan Muhamedagic:
>> >>>> Currently, crm shell cannot handle quoted parameters with spaces from the command line. Try to enter it interactively:
>> >>> The crm shell gets what bash (or dash) passes. The quotes are
>> >>> gobbled by bash in that process. Basically, what you type is not
>> >>> what crm sees.
>> >>>
>> >>
>> >> I understand the mechanics. It is machine-centric and confusing to some
>> >> users.
>> >>
>> >> It is not absolutely impossible to accept quoted parameters from command
>> >> line. Quotes are stripped, but these parameters are preserved in single
>> >> arguments.
>> >>
>> >>> Right. I really don't understand why's everybody trying to
>> >>> configure the cluster directly from bash instead of doing it in
>> >>> crm configure.
>> >>>
>> >>
>> >> CRM shell offers this opportunity, that's why many users try to use it.
>> >
>> > The fact that the shell has this feature doesn't mean that it
>> > should be misused. It was meant mainly for one-off management
>> > commands (such as "resource stop" or "node standby") and only
>> > very seldom for one-off configuration commands. There's a
>> > plethora of things which can go wrong when used for
>> > configuration directly from the command line. It also makes the
>> > cluster work harder because every invocation has an implicit
>> > commit and the changes immediately take effect.
>>
>> There are, as CTS shows, ways around that.
>
> You mean using CIB_file? Yes, that's one possibility, though I'm
> not sure how many people know about it. Another would be to do
> the changes in a shadow CIB.
>
>> > Because of that
>> > the user also must pay attention to the order of configuration
>> > elements which otherwise isn't necessary.
>>
>> Can you define constraints for resource's you've not defined yet?
>> What about clone or group resource's that are also not yet defined?
>>
>> (If yes to either, that sounds more like a bug than a feature)
>
> What I meant is this: If you stuff all the configuration into a
> file and do "configure load" or just feed it to "crm configure",
> the order of elements in the file does not matter as long as all
> referenced elements are presents at the commit time.
>
>> > It's a bit like taking
>> > apples one by one from your garden into the house, instead of
>> > bringing over a box of apples. Just worse than that. I could go
>> > on, btw.
>>
>> I'm sorry, but this is all a pretty weak argument.
>> Not that I consider proper quote escaping the crm shell's problem, but
>> the ability to script the creation of a config is quite important.
>
> The proper way to script a configuration would be by way of
> here-document or similar.
>
>> What about adding some additional sanity checking for the params section?
>> It should be pretty easy to spot (and warn about) entries not of the
>> form "name=value" which would be indicative of a quoting error.
>
> Unfortunately, CIB allows a parameter to be defined but not have
> a value, that's why a token by itself within parameters is
> considered to be this case.

True, but almost no-one uses that intentionally - thats more of a node
attribute feature.
If you see $name by itself and there is no $name parameter advertised
in the metadata - its highly likely that there's a quoting issue.

You could require the syntax for that to be name= instead of just name though.
But anyway, thats why I suggested producing a warning rather than
always trying to fix it.




More information about the Pacemaker mailing list