[pcp] [resend] Derived Metrics - RFC

Max Matveev makc at gmx.co.uk
Sun Nov 8 02:41:51 CST 2009


On Fri, 06 Nov 2009 06:54:59 +1100, Ken McDonell wrote:

 kenj> On Fri, 2009-11-06 at 00:15 +1100, Max Matveev wrote:
 >> On Thu, 05 Nov 2009 12:28:58 +1100, Ken McDonell wrote:

 kenj> As an aside does any know of a living user of these asynchronous
 kenj> interface extensions to the original PMAPI?
 >> Yes, I do - when collecting metrics from multiple hosts I cannot
 >> afford for TCP to pick its nose when one of the hosts go down.

 kenj> That's a fair justification.  But would it be reasonable to assume that
 kenj> 99.99% of these cases are related to pmFetch() as this is the only PMAPI
 kenj> routine most monitoring apps call in steady-state after initialization?
There is occasional indom interrogation when new instances appear (it
happens much more often now due to silly implementation). The others
are used too when extra sources come online and after the
reconnect to the pmcd on a host which went AWOL.

Reconnect itself is by far the most convoluted part with multistep
procedure just to establish the connection between the client and the
pmcd.

 kenj> If so, a compromise may be to dump all the async stuff in places like
 kenj> PMNS manipulation, pmInDom interrogation, help text retrieval, pmStore,
 kenj> pmDesc?
I think the only viable dumpee in this list is text unless Nathan has
plans to switch kmchart to this interface - text is only useful for
humans and humans are not going to sit inside the automated tools
which monitor multiple hosts.

max



More information about the pcp mailing list