[Linux-HA] Re: new crm commands (a proposal)
dejanmm at fastmail.fm
Fri Jan 9 02:38:22 MST 2009
On Thu, Jan 08, 2009 at 10:50:27AM -0800, Joe Bill wrote:
> On Wed, Jan 7, 2009 at 14:49, Dejan Muhamedagic <dejanmm[at]fastmail.fm> wrote:
> > Most of you are probably back from holidays, so now may be good
> > time to give you some food for thought.
> Happy to oblige.
> >1. Detach monitor operations from primitives
> >old (one statement):
> >new (three statements):
> Why the "params" keyword in the new syntax ?
Because there are also meta attributes. In the very beginning, I
tried to make it work without the "params" keyword, but for some
reason I can't recall anymore it didn't work.
> Agree for the number of statements, but keep the old, unambiguous syntax like this:
> primitive drbd0 drbd drbd_resource=drbd0
> monitor drbd0 role=Master interval=59s timeout=30s
> monitor drbd0 role=Slave interval=60s timeout=30s
> Strongly suggest to support *both* syntaxes.
Of course, both will be supported.
> >2. Shorter specification of boolean variables
> Seems more cosmetic to me but have no reason to oppose.
> Again strongly support *both* syntaxes.
> CAUTION: Do not mix keyword lists (a la xml) and language constructs.
> "not globally-unique" : NOK
> "!globally-unique" : OK, preferred over above
Well, these are the same, some people of course do prefer '!'
over 'not', but I still believe that most of the users are not
programmers, so I assumed that they'd appreciate 'not' rather
than C syntax.
> "not-globally-unique" : OK
This is an interesting proposition, but probably hard and error
prone to do right.
> > 3. New "prefer" statement/clause for location preference
> >location ms-drbd0-master-on-xen-1 ms-drbd0 rule role=master 100: #uname eq >xen-1
> >prefer ms-drbd0:Master xen-1 # score 100 is implied
> Ok, but support also:
> prefer ms-drbd0 on=xen-1 role=master # score 100 is implied
> ... and combinations like:
> prefer ms-drbd0:master on=xen-1 ... or ...
> prefer ms-drbd0 xen-1 role=master
That should be covered by the location statement. The new prefer
statement would be only for simple location preferences, i.e.
node plus score [plus role].
> >new1 (two statements):
> >group g1 global-ip web-server
> >prefer g1 node1
> >new2 (one statement):
> >group g1 global-ip web-server prefer node1
> Agree with the two-statement.
> Disagree with the one-statement.
> Ambiguous. Represents an unspecified language construct.
> Creates a name exception to handle the nature of this language construct.
> How do you parse, what rules used to parse:
> group g1 global-ip prefer prefer node1
> group g1 global-ip prefer node1
> group g1 global-ip web-server prefer prefer
The prefer keyword would of course be reserved. Too bad for
people naming nodes "prefer" ;-)
> >4. Shorter specification of colocation constraints
> >colocation apache-group-on-ms-drbd0 inf: apache-group ms-drbd0:Master
> >colocation apache-not-with-slave -inf: apache-group ms-drbd0:Slave
> Ugh. Definitely needed.
> I apologize in advance, I really looked everywhere but may have
> overlooked this, exactly which HA facility accepts this
> "colocation" command with this syntax ?
This post is all about the new crm shell and configuration
utility which is available as of pacemaker 1.0.
Sorry for not being more precise:
When I wrote "old" in the original post, I definitely didn't mean
"obsolete". The "old" syntax will be kept.
> >colocate apache-group ms-drbd0:Master # inf is implied
> >separate apache-group ms-drbd0:Slave # -inf is implied
> Unfortunately, ambiguous.
The first one should mean: keep the two together, the second:
separate the resources. How exactly ambiguous?
> Also, "colocate" and "separate" not the best words to describe
> the actions I understand they are supposed to undertake.
I looked through the Roget's for ages and could find anything
better for collocate. Any suggestions? I'm all ears :)
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> See also: http://linux-ha.org/ReportingProblems
More information about the Linux-HA