How to find a memory leak

time to read 2 min | 289 words

Originally posted at 10/17/2010

I got the following message in the rhino tools mailing list:

I am looking into Rhino-esb and NServiceBus for a smart client application. I was doing some stress testing lately and I noticed some very strange behavior with rhino-esb. I tried to send large numbers of requests at the same time (6000-10000) and the memory of my back-end when using rhino-esb was continuously rising.

Since RSB is in production for the last two or three years, that seemed suspicious. Luckily, there was a reproduction that I could run. I tried it out, and indeed, memory seems to be taken, in proportion to the number of messages sent. That had me worried, really worried.

I run the application under memory profiling (using JetBrains dotTrace), and tried it. Which gave me this:


I went Ouch! and Huh?! at the same time. The next step was to find who was holding those. Luckily, that was as easy as asking the profiler.


And a short hop to the code explained what was actually going on.

There is a LRU buffer there to prevent duplicate messages from being sent, and the default limit for the buffer is 10,000. And since the buffer is swept once every 3 minutes. It would look like a memory leak.

But what really pleased me wasn’t so much the answer, but how easy it was to figure it out.