ActiveMQ

From TipperWiki

Q1/Q2 2008 we tried to put ActiveMQ into production. It failed. Here's a discussion I had today about that experience. --Jhannah 12:56, 1 March 2009 (UTC)

06:42 -!- Irssi: Starting query in Freenode with rjbs
06:42 <deafferret> In 2008 we tried to put ActiveMQ in our critical path. Everything went swimingly until multiple 
                   unexplained crashes. We abandoned it and went back to database tables
06:45 <deafferret> I/O to ActiveMQ via Net::Stomp. works great and easy until ActiveMQ itself pukes after a few 
                   million messages
06:45 <deafferret> but then we'd load test it and it would be fine for 10M records. Then we'd roll it back into 
                   production and it would barf again after a while
06:46 <rjbs> morning!
06:46 <rjbs> That is a discouraging story. :/
06:47 <rjbs> We currently have our own db-based job queue (like basically eveybody does) but it stinks and I'd 
             rather replace it with something well-known than rewrite it with a new internal thing... 
06:47 <deafferret> ya. sounds like me, Q1 2008  :)
06:47 <deafferret> it's insanely fast
06:47 <rjbs> still using your own db thing, then?
06:48 <deafferret> Maybe our guy didn't try hard enough? (I wasn't the lead for the puking, he was)
06:48 <rjbs> yeah, I played with it just enough to see that I could create and complete jobs with stomp
06:48 <deafferret> I basically set the whole thing up for him and handed it over
06:48 <rjbs> I -think- that Net::Stomp was written to talk to ActiveMQ, so I should try talking to its author.
06:48 <deafferret> he's a java guy, so he was really comfortable on that side
06:48 <deafferret> he messed around with different persistance plugins and everything, but in the end it wasn't 
                   worth the trouble
06:49 <deafferret> maybe it's more stable now, a year later?
06:49 <rjbs> Heh.  Yeah, that's another of my (minor) problems.  All the docs are for using the Java libraries
06:49 <deafferret> well, Net::Stomp is dead easy and works, and ActiveMQ is dead easy to configure
06:49 <rjbs> looks like no releases since april
06:49 <deafferret> so theoretically the fact that the black box is java shouldn't matter
06:50 <deafferret> oh. that's not goo
06:50 <deafferret> d
06:50 <rjbs> right.  i don't mind that it's written in java, caml, haskell or anything else
06:50 <rjbs> Oh, no, I misread their site.
06:50 <deafferret> ya, I was pretty bummed when we had to give up on it
06:50 <rjbs> November.
06:50 <rjbs> that's much better
06:50 <deafferret> oh ok
06:51 <rjbs> Nov, May, April.  Those are all v5, which came out as 5.0.0 in Dec 07.  
06:51 <rjbs> Anyway, thanks for the datapoint, even if it is somewhat discouraging!;)
06:52 <deafferret> so what we did was create an abstraction namespace called OmniMQ. for now that namespace is using 
                   sub classes that use database tables, but the ActiveMQ stuff is still out there just in case we 
                   go back to it (or something else?) someday
06:52 <deafferret> but we're pretty happy with the performance of our cheesy-inefficient MySQL setup, shrug
06:52 <rjbs> cool
06:52 <rjbs> yeah, replacing our crappy one with a much better crappy internal one woud surely be sufficient
06:53 <deafferret> it's just AWESOME when ActiveMQ pushess thousands of messages with no delays, all in RAM  :)
06:53 <rjbs> I just really like not writing code for things other than our core competencies.
06:53 <deafferret> ya
06:53 <deafferret> the benchmarking of ActiveMQ freaked me out. the create-new-queues-on-the-fly feature freaked me 
                   out. the multiple subscriptions was way cool
06:53 <deafferret> oh well   shrug
06:55 <rjbs> I'm also looking at TheSchwartz, but it's less documented.
06:55 <deafferret> :)