launchpad-dev team mailing list archive

[Robert Collins <robertc@xxxxxxxxxxxxxxxxx>]

On Sat, Oct 8, 2011 at 3:26 AM, Gary Poster <gary.poster@xxxxxxxxxxxxx> wrote:
> Sorry, for some reason this has been sitting around unsent.  I wondered why no one had replied! :-P  So anyway...

No worries ;)

> Until the oops tools have more valuable searching available, I'm skeptical of switching the format.  I must admit that every time I've needed to do "real" analysis of OOPSes in the past several months, I've gone straight to devpad and skipped the oops tool entirely.
>
> Until we  build a much more powerful oops search tool than we have now, I'm -0.5.  I don't go the full way to -1 because I believe Robert's argument is that they are still greppable.  Maybe so, but I'm worried about the change.

One way to get comfort would be to play with it yourself - so I've
included a setup you can use to do that below.

> I'd like to see some experiments with actually trying to do OOPS analysis on the filesystem with the proposed format before actually switching.  That's no demand, of course, just stating my position.

Naturally. I think the tradeoffs are worth it (or I wouldn't have
started the thread) :).

I have a usable combination of branches now. If you get both
lp:~lifeless/python-oops-tools/amqp and lp:~lifeless/launchpad/useoops
running at once you can see how it all hangs together.

Do the normal make schema dance, and follow the oops-tools README to
get a django environment up and running for it on a new postgresql
cluster.

Then:
1) make run to start Launchpad
2) start oops-tools django server. I use bin/django runserver
192.168.x.yyy:8000 because I'm running it in a lucid LXC container.
3) start the amqp consumer for oops-tools:
bin/amqp2disk --host localhost:56720 --username guest --password guest
--vhost / --queue oopses --output $(pwd)/../amqp-oopsdir --bind-to
oopses -v
(the rabbit config is from 'make run', the output path is where to
store the oopses, the bind-to creates the exchange and queue and -v
prints received oopses).
4) visit https://launchpad.dev/++oops++
5) ???
6) Profit!

This will let you populate bson oopses yourself, which will simply
show you how binary they are (though still greppable via grep) and get
a feeling for how much adhoc python you'd want to be able to do a
sensible console map-reduce over them :)

It will also let you see how convenient getting a development OOPS up
in oops-tools is : now I can do this, I'm going to be doing it all the
time, just because of the sweet sweet query collation.

-Rob