launchpad-dev team mailing list archive

[Gary Poster <gary.poster@xxxxxxxxxxxxx>]

On Jun 3, 2010, at 5:19 PM, Robert Collins wrote:

> A couple of thoughts:
> 
> - consider grabbing the RPM agent format and using that as the output
> protocol - this would permit:
>   - using RPM if we want to
>   - telling other folk currently using RPM that they should
> collaborate on $whatever tool we write instead

We already have a log format from the existing tool.  I'd be +1 on investigating that possibility for a later iteration.

> Also I'd like to track:
> - time spent in accept() queue

Yes, already tracked.

> - time spent in apache SSL negotiation

I don't understand why that would vary significantly per URI/pageid, except with repeated SSL handshakes for new connections that download page components such as images, CSS, and JS.  We won't be able to track that easily on the server side, and we do have some visibility right now on the client side via the webpagetest.org stuff.  ...Am I missing something?

> - cache hit rate with squids

Yes.  This is a somewhat lower priority IMO, since Squid is only used for anonymous users.  Right now I'd just want it to validate my assumption that a majority of anonymous requests are cache hits in Squid.  That would be a good thing to validate, but we should be able to use simpler metrics.

If that assumption is not validated, or if we change our Squid scheme into something that is used for authenticated users, then I'll change my mind.  Even now, I agree that it is a nice to have.

> all back against the same URI's/pageids
> 
> I think URI's matter sometimes - bugs.launchpad.net/ubuntu, for instance.

Yeah.  We have URIs now.  Maybe it will be appropriate to look at the report generated from the URI perspective and then generated from the pageid perspective (once we have it).

Gary