launchpad-dev team mailing list archive

[Michael Hudson <michael.hudson@xxxxxxxxxxxxx>]

On Wed, 10 Nov 2010 18:11:23 -0800, Max Kanat-Alexander <mkanat@xxxxxxxxxxxx> wrote:
> On 11/10/2010 06:04 PM, Martin Pool wrote:
> >  * make a 1.18 branch and tarball; in future this will get only safe
> > bugfixes and Launchpad should deploy this until a 1.19 is stable
> 
> 	This should be happening later today or tomorrow, and I'll email the
> list in a separate thread about it when it's done.
> 
> 	However, I do need to know exactly what revision of loggerhead
> Launchpad is currently running.

You can see that in the utilities/sourcedeps.conf file in the launchpad
tree.  Currently it says:

loggerhead lp:~launchpad-pqm/loggerhead/devel;revno=176

> >  * turn off per-line annotations in the default view
> 
> 	Yep, and I think that this will solve quite a few of the existing
> performance problems. Basically all that will happen is that the
> "Revision" column of the default view will disappear unless you choose
> to see it, probably by clicking on a new text link at the top of the page.

+1.  I have a few branches on Launchpad that sort of start towards this,
if you want to look.

> 	After this, I may also remove the "Latest Rev" and "Last Changed" view
> from the default "Files" view, which would mean that we wouldn't have to
> go into bzr's history at all for two most-used pages in Loggerhead.

If you have a revision tree, this information is pretty much there
(except for the revision numbering, I guess).

> >  * add a 'just plain text' view
> 
> 	Yes, although as mentioned when we discussed this before, there are
> security concerns. I've dealt with such concerns thoroughly in another
> app with much higher security requirements, though, so I know how to
> deal with it.

I'm glad someone else is willing to tackle this nest of vipers :-)

> >  * work with http caches - suggested refresh interval, perhaps etag
> 
> 	Yeah, etag at least. Expires is a bit more questionable for dynamic
> pages, but the etag will do most of what we need in terms of making the
> actual performance difference, I think.

A problem with etag is that squid doesn't pay any attention to it.  But
it will help browsers, I'm sure :-)

> >  * performance/memory usage serving a single page
> 
> 	Well, more than serving a single page, running as a single thread. The
> problem right now is that loggerhead is heavily dependent upon its
> in-memory cache that it shares between its threads in order to have
> decent performance. Removing the revision data from some of the default
> views (and possibly improving interactions with bzr-history-db) will
> make a big difference here, though.

Is Launchpad using the history-db inspired stuff yet?  It would be great
to get that deployed -- one of the things I didn't get done before
leaving the launchpad-code team :(

Cheers,
mwh