Re: [pcp] pmcd gives up on slow starting Perl PMDA

[Ken McDonell <kenj@xxxxxxxxxxxxxxxx>]

On 25/03/14 00:25, Martins Innus wrote:
> ...
> We have proc pmdas that are timing out on heavily loaded systems, my
> next task was going to be looking into this.  Should I wait until this
> new api/method is fleshed out before doing so?
Waiting would be good ... 8^)>

In the interim, you have a sledge hammer that can be used.

The timeout in pmcd is configurable ... both at pmcd start up (via -t 
<n>, usually using /etc/pcp/pmcd/pmcd.options) and dynamically (via 
pmstore pmcd.control.timeout <n>) ... see the -t option in pmcd(1).
The down-side of making this bigger than the default 5 (seconds) is that 
you'll wait longer for a really dead PMDA to be noticed, and the 
semantics of the data returned in a pmResult may be a little fuzzy ... 
if the requested metrics span more than one PMDA they all come back with 
a single timestamp, but the data may have been collected several seconds 
apart from the different PMDAs (comparing a fast PMDA to a slow PMDA)