Opps, I forgot to do reply all.
On Wed, Jul 14, 2010 at 2:41 PM, Darryl L. Pierce dpierce@redhat.com wrote:
On Wed, Jul 14, 2010 at 01:29:53PM +0200, Andrew Beekhof wrote:
- <property name="cpu_model" type="lstr" access="RO" desc="The process model description." />
- <property name="cpu_cores" type="uint8" access="RO" desc="The number of processor cores." />
Number of cpus? Should we assume all CPUs are the same?
That's a good question. My experience has been with systems where the CPUs are all the same model. Is it possible for a system to have multiple different physical processors? If so, should we have a "list_processors" API and a Processor agent to represent it.
Unless anyone knows different, I'd vote for assuming they're all the same and adding: cpu_count
- <!-- statistics -->
- <statistic name="last_updated" type="absTime" desc="The last time a heartbeat occurred." />
Not sure last_updated is useful. If the system is up sufficiently to give us this (or any) value, then its also up sufficiently to update it. Though perhaps I'm misunderstanding the distinction between property and statistic.
A property is a fixed value, while a statistic is one that changes.
In this case, with the last_updated timestamp, the Qpid broker will cache for an indeterminate period of time, the vaues from an agent. The agent can actually die and the value will still be in the broker. So this gives a way of seeing how long it's been since the last time the agent sent a heartbeat and infer from that if the agent is dead.
Ah. Ok, sounds good.
- <statistic name="load_average" type="double" desc="The current processing load average." />
You'll probably want to go with the standard and make the 1,5, and 15 minute load averages available.
Okay, that's cool. I'll update the wiki and schema to have "load_average_1", "load_average_5" and "load_average_15". Does that naming seem reasonable?
Sure.
<eventArguments>
- <arg name="timestamp" type="absTime" />
- <arg name="sequence" type="uint32" />
</eventArguments>
<event name="heartbeat" args="timestamp,sequence" />
Strange syntax there. Does this mean all events must contain the same arguments?
No. It just defines the "timestamp" and "sequence" argument types, and then declares heartbeat as sending out those two types during its event.
Ah, I see how it works now.
On Wed, Jul 14, 2010 at 03:20:20PM +0200, Andrew Beekhof wrote:
On Wed, Jul 14, 2010 at 2:41 PM, Darryl L. Pierce dpierce@redhat.com wrote:
On Wed, Jul 14, 2010 at 01:29:53PM +0200, Andrew Beekhof wrote:
- <property name="cpu_model" type="lstr" access="RO" desc="The process model description." />
- <property name="cpu_cores" type="uint8" access="RO" desc="The number of processor cores." />
Number of cpus? Should we assume all CPUs are the same?
That's a good question. My experience has been with systems where the CPUs are all the same model. Is it possible for a system to have multiple different physical processors? If so, should we have a "list_processors" API and a Processor agent to represent it.
Unless anyone knows different, I'd vote for assuming they're all the same and adding: cpu_count
Sure. And if, in future, we find out that there can be different CPUs on the same system, we can expand the APIs to include an index argument, and default it to 0.
matahari@lists.fedorahosted.org