#musicbrainz-devel

http://codereview.musicbrainz.org/r/391/ ping? 8 days, no review... :)

aCiD2: trying to follow your last note in CR - are we good then? :D

luks: thanks; I just realized that somehow I've not actually been seeing all tickets in CR, as I wasn't part of the ngs group.

"good" might be pushing it, but I'll give you approval to carry on :P

luks: have you seen http://search.cpan.org/~jswartz/CHI-0.28/lib/CH... ?

seem to be a lot of people I see in perl irc channels advocated its use if you ever want to do caching

mind you, they advocate DBIx::Class everywhere, so no guarantee it's useful for MusicBrainz

I haven't seen it

how would it help us?

I generally don't see a reason to change something that works

Well we use a memory cache and a memcached cache

A unified front end might simplify stuff

but I haven't used any of the caching stuff we use atm, nor chi, so I can't comment

well, the code is already written and works

I'm not saying it's a good idea to use something like this

but we already have a working solution, and any change can just bring more problems

er, s/it's a good idea/it's not a good idea/

sure

i'm not really suggesting the change now, no doubt I'll end up having a play with CHI to see what all the hype is about

just wanted to let you know too

luks: $(element).attr('name') is a core jQuery function; since when would that not work in IE?

:P

IE doesn't allow you to set name of a cloned input

I know that, because I spent days hopelessly looking for a solution a long time ago

Which IE?

any IE

ouch

so luks, you think edit work should let you change artist credit too, right?

yes

oh, I can't do this now, because we don't have an artist credit editor, heh

same with edit release group, which has been stalled for about 2 months now

ohh, they somehow managed to workaround the name issue in IE

with new jquery it works

aCiD2: hm, the work patch is good then

but please add some TODO comments there

good idea, will do

Anyone using the main web site at the moment who wants to be a guinea pig, please msg me. It involves no work on your part other than confirming whether or not the site keeps working for you. Thanks!

djce: if it's not too late -- take me ;)

current status -- works

if anyone wants to call "WTF?!" out loud every five seconds, look here mms://live.msmedia.zdf.newmedia.nacamar.net/zdf...

this is the german election fight

hey ruaok

hey

It /almost/ worked... but not quite.

oh, I'm too late, eh?

what didn't work?

For some reason, some requests didn't get the real client IP substituted in place.

So for example the ratelimit key used was "ws ip=10.1.1.13"

i.e. lots of clients sharing the same ratelimit key,

which is bad, of course.

Not quite sure where it's going wrong.

Anyway, disabled for everyone except me and pronik, who volunteered earlier :-)

hmmm. was this specific to the approach used or would all approaches have the same problem?

The approach used was to minimise downtime,

small backwards-compatible steps.

Another approach would be a bit of downtime, a bit of mb_server rewriting, and a big bang release.

simpler, but far more disruptive.

"mb_server rewriting" ?

The problem is any parts of the code which needs to know the client's IP address.

Namely: ratelimiting, and cookie checking.

Currently it uses the mod_perl API to query remote_addr,

but there's already a hack in place (has been for years) to bodge that,

because the client IP is 127.0.0.1 (apache2)

would it be better for us to do that for NGS and not worry about the old mason codebase?

It's a prequisite for the load balancer.

So yes, if you're prepared to wait.

s/yes/it's an option/

have you gauged how much work it would be?

Well I thought I'd done the work already.

So I'm confused as to what I've missed.

maybe we need a more realistic test sandbox?

It's some difference between dexter and stimpy.

So it looks like a codebase diff should turn up the problem.

it works if the load balancer sends the request to dexter, but not if stimpy.

I'd start by diffing DBDefs.pm on both boxes.

the other SVN code *should* be the same.

Ah, I think I've found it...

some nginx config I was playing with which I forgot to remove.

right, I'll give it another go.

looks good this time.

\ø/

well done!

Yes, that was just some config I was trying earlier, seeing if I could short-cut one of the steps.

Apparently, I can't :-)

lo

ok, going offline for a moment, back in a couple of minutes

k

ok I'm having some dinner now,

ping me if anyone sees any problems

and if all's well I'll continue the transition after dinner.

sounds good.

happy noms

damn libunac!

crashes again and again on me and I can't find the workaround luks gave me sometime ago

ruaok: everyone's proxying now.

seems to be happy.

sweet!

I'm going to put the java search server into rotation later today.

and probably keep if there for a bit.

the log analysis gave us one bug to take care of.

cool, only one. nice

yeah, paul seemed to think that would work fine.

the only tricky thing I see is having a recent JVM on all the boxes.

catbus might be out of date for that.

catbus==generator right?

yep.