pcp
[Top] [All Lists]

Re: [pcp] One pmwebapi against multiple pcp running hosts

To: pcp@xxxxxxxxxxx
Subject: Re: [pcp] One pmwebapi against multiple pcp running hosts
From: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Sun, 07 Apr 2013 16:15:41 +1000
Cc: junwang123@xxxxxxxxx, devpriya00@xxxxxxxxx
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <CANHDG4bfT6b3zYy2bOHhhK0Rp2q5et-GMzEPN4HaZ_TCW9TZYQ@xxxxxxxxxxxxxx>
References: <CADPXOeknVWyG8ZWgEcnVnxtHoYr3tq+g5EVof4AdAhxGZH4eiA@xxxxxxxxxxxxxx> <20130402205337.GD3659@xxxxxxxxxx> <CADPXOe=NbKrgxCtvePnAMM_MCUHibdWseAni5p+_e4k4Hwp1ow@xxxxxxxxxxxxxx> <20130402222722.GE3659@xxxxxxxxxx> <CADPXOekTUhrF-82ZNALf68rFfSftqOGVM0ahVEqxSVgs+MMRyw@xxxxxxxxxxxxxx> <CANHDG4bfT6b3zYy2bOHhhK0Rp2q5et-GMzEPN4HaZ_TCW9TZYQ@xxxxxxxxxxxxxx>
User-agent: Mozilla/5.0 (X11; Linux i686; rv:17.0) Gecko/20130308 Thunderbird/17.0.4
On 05/04/13 10:31, Jun Wang wrote:
Hi Frank/Nathan,

I was trying to use the summary PMDA and have two questions.

1. I wasn't sure whether the followings are right.
2. We noticed that we lost the Unit for summary.avg.memory in pmwebapi.
mem.util.used is using u64 and kbyte, but the summary is a double (no
kbyte unit). I tried to add "summary.avg.memory = avg_host $memoryhosts
Kbyte", but got a syntax error.
Can someone help?

Maybe ... 8^)>

I suspect what's happening here is that the metrics are being converted to cannonical units (byte and sec for the dimensions of Space and Time) and cannonical type of double ... this is being done by pmie(1) which is the real workhorse behind the summary PMDA. As a result you cannot get non-cannoical units or types for the summary metrics, so you're stuck with type double and units of byte.

So this should have produced meta data for the metric like this:

summary.qa.mem
    Data Type: double  InDom: PM_INDOM_NULL 0xffffffff
    Semantics: instant  Units: byte

BUT you've uncovered a very obscure bug in pmie (or the summary agent, I'm not sure which yet), where the addition of the avg_host predicate to the summary metric declaration causes the Units to be botched and set to none instead of byte, as in ...

summary.qa.avg_mem
    Data Type: double  InDom: PM_INDOM_NULL 0xffffffff
    Semantics: instant  Units: none

The new QA test 525 reproduces this fault.

I don't expect the fix will be too hard to find, says he bravely!

Cheers, Ken.

<Prev in Thread] Current Thread [Next in Thread>