pcp
[Top] [All Lists]

Re: pmParseMetricSpec(3) problems

To: nscott@xxxxxxxxxx
Subject: Re: pmParseMetricSpec(3) problems
From: Mark Goodwin <markgw@xxxxxxx>
Date: Thu, 01 May 2008 09:18:51 +1000
Cc: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>, pcp@xxxxxxxxxxx
In-reply-to: <54508.192.168.3.1.1209591565.squirrel@mail.aconex.com>
Organization: SGI Engineering
References: <40997.192.168.3.1.1209514118.squirrel@mail.aconex.com> <1209590407.2870.19.camel@localhost.localdomain> <54508.192.168.3.1.1209591565.squirrel@mail.aconex.com>
Reply-to: markgw@xxxxxxx
Sender: pcp-bounce@xxxxxxxxxxx
User-agent: Thunderbird 1.5.0.14 (Windows/20071210)


nscott@xxxxxxxxxx wrote:
I think these are both solvable.

>From a spec point of view (see PCPIntro(1))
a) eating multiple colons in the archive name is a no brainer

*nod*. Simple Matter Of Programming there.

b) if neither host: nor archive/ is present the metric spec is still
valid, so disk.all.total and disk.dev.total[sda1] and
disk.dev.total[mydisk,yourdisk theirdisk] are all valid and refer to the
local pmcd's view of the metrics world.

Thats pretty much how it is today - except I think that if there is no archive/host specified at all, then the passed in value "isarch" is used in the result structure passed back out of pmParseMetricSpec.

But, my issue was local contexts (not pmcd) - I don't see any way for a
pmMetricSpec to specify this third kind of context today, and not sure
what the best approach to take to implement that is.

perhaps a reserved word, e.g. LOCALCONTEXT:

Do we also want to try and entertain a chain of proxies, separated by ':'?
I think you mentioned you have no need for more than one, and our
need has gone away entirely now.

Cheers



On Wed, 2008-04-30 at 10:08 +1000, nscott@xxxxxxxxxx wrote:
...
Second, I've just recently come across the fact that theres no way to
specify
use of the local context through this interface - it handles only
hostnames
and archive filenames in its parsing, and the API "int isarch" parameter
makes
resolving this quite tricky.

I'm not sure how best to fix this one.  It may be that we need a new
interface
here, which allows all context types to be passed in?  Or could we
change the
"int isarch" field to be the context "int type" and also make
localhost:/metric[]
to mean local context?  (the latter bit seems quite dodgey to me - but I
don't
see a better way).

cheers.

--
Nathan




--

 Mark Goodwin                                  markgw@xxxxxxx
 Engineering Manager for XFS and PCP    Phone: +61-3-99631937
 SGI Australian Software Group           Cell: +61-4-18969583
-------------------------------------------------------------


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