[Linux-ha-dev] Clone Notifications

Lars Marowsky-Bree lmb at suse.de
Wed Aug 10 04:40:40 MDT 2005


On 2005-08-10T12:15:57, "Andrew Beekhof (GMail)" <beekhof at gmail.com> wrote:

> sorry,  "is being started" or "was started".  the tense being  
> inferred by the type of notification "pre" or "post".

Ok.

> if the start fails you wont get the post notification
> you'll next get a pre-notification for a stop
> 
> if some clones started and some didnt, the ones that did start will  
> next appear as stable <peer> entries.
> 
> does this answer your comments below?

> >I'm not convinced we need to tell them about unchanging things. For
> >example, the above makes sense whether or not :1 was already  
> >running on
> >node1 or not - we tell it that we intend that it will be started
> >afterwards, which is probably fine.
> 
> i dont think thats a good idea
> 
> how do you then differentiate between :1 starting and :1 already  
> being active?
> i can imagine different things happening in the different cases
> 
> I'd instead go with CLONES_IS_ACTIVE="1"

ACTIVE bad. STARTED good. (Then we can add promoted/demoted later.)

> >Now I'm also a huge fan of conveying more information than  
> >necessay ;-).
> >Hm. We could list those in _STOPPED, _STARTED, _PROMOTED, to  
> >indicate we
> >ain't doing anything with them, but leave them in whatever state they
> >are in
> 
> started/stopped is fine.
> not sure about promoted yet - i'll defer that until master/slave gets  
> written

That was just looking forward ;-)

> >(and in a post-notification, that would of course imply that
> >everything would be in _STARTED, for example).
> 
> i dont think thats a good idea, then the RA no longer knows what we did
> 
> again, i can imagine there might be different logic for nodes that  
> started/stopped
> versus nodes that were already started/stopped
> 
> or maybe include them in both lists, though it can easily be inferred

Well, OK, I have this narrow view of the GFS/drbd/OCFS2 resources which
don't care about this distinction. But yes, two lists probably makes
more sense, as I also elaborated below in the example...

> >(And, of course, for "_STOPPED", there'd be no _NODE_ entry. Or one  
> >just
> >listing them as "NULL". ;-)
> >
> 
> as per started

"started"? Dear sir, for sure "started" will have to have a NODE entry,
because it is certainly running on a given node.

> >Do we want to head down the path and list _FAILED too? ie, which  
> >clones
> >have failed on which nodes? We have the info, just not sure whether we
> >want to / can easily expose it.
> >
> i would say "not at this time"

OK. Just asking.


Sincerely,
    Lars Marowsky-Brée <lmb at suse.de>

-- 
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business	 -- Charles Darwin
"Ignorance more frequently begets confidence than does knowledge"




More information about the Linux-HA-Dev mailing list