Haven’t blogged in a while, had some time to write now.
Since I last blogged, I made IRC Browse. It’s a service which allows you to browse the IRC logs of the #haskell and #lisp channels of Freenode. The logs come from tunes, and, for Haskell, they go back to 2001. I like IRC. I don’t go on it that frequently anymore, but I like to read the logs and I see it for the useful communication and coordination tool it is. I’ve always wanted a trivial way to view and share IRC logs as a service, so I made one. The source code is here.
It boasts these features:
It’s written in Haskell, using Snap, PostgreSQL for the database, and Sphinx for search. It’s fast.
I made it ages ago, really, but thought it worth blogging about once.
The IRC summary is generated upon request, and reveals some possibly interesting insights into channel activity and the top contributors.
Of interest the most is the activity by year, which indicates that 2009 was the apex of the IRC channel’s activity, which has since dwindled, and appears to be continuing to dwindle: despite sustained activity, conversation generally is decreasing.
There are various hypotheses put forth for this. I speculate that
This is where the name “IRC Browse” comes from. There used to be a service at ircbrowse.com, a few years back, providing a similar browsing service. I asked the author of that old site whether I could use the name ircbrowse.net, and they approved and wished me luck.
One thing that bugged me about the old IRC Browse was the speed. It was god-awfully slow. It would take ages just to display one page of logs. What I wanted was to have a log browsing service that would be instantaneous and snappy.
After some learning with PostgreSQL, I discovered some ways to make
paginating 26 million rows of a table quite fast. Simply using
LIMIT is far too slow—takes about one
second to retrieve a result. I couldn’t simply query on the IDs, because
there isn’t just one channel, or one pagination type. So I created a
separate table to store paging indexes. For every row of the “event”
table, I created a corresponding, ordered, row in the index table. After
that, it was snappy.
Another thing I discovered is that my pgsql-simple library was a little sluggish. The pages would retrieve in, say, 50ms, rather than, say, 2ms. So I switched the library to postgresql-simple and got the extremely snappy responsiveness that I wanted.
For searching I learned how to use the tool called Sphinx. It takes in a configuration and a database, and then populates a search index. From that search index, it provides very fast full text search.
I couldn’t get the Sphinx library to work with the version of Sphinx I was using at the time. I made a trivial wrapper to the command line program instead. That worked. At some point I will replace this with use of the Haskell sphinx library.
Another optimization I can do is split the indexes into #haskell and #lisp.
Profiles give a nice way to tell when someone probably goes to sleep and is probably available. It also tells whether someone has been active lately. If they haven’t been active lately, you can check their complete history by year, and if you see it dwindling, perhaps they’re not on the IRC anymore.
There are also quotes @remember’d by lambdabot, which can be fun to read.
Importing the logs happens daily, at 10:30 UTC time. One day I might update this so that it connects to the IRC directly and updates the logs in real time. But I’m not sure it’s worth it.
I also did a social graph thing, but it’s not that good and I will probably remove it. There’s a word cloud, which looks pretty enough, I’ll keep that.