Producer network thread is leaking via DirectByteBuffer's

Hi all, I also posted this here, but I didn’t get a response unfortunately.

I have a problem. The direct memory is linearly increasing by about 500 MB/6h. It does not seem to go down. After inspecting the heapdump with OQL (this query: SELECT x, x.capacity FROM java.nio.DirectByteBuffer x WHERE ((x.capacity > 1024 * 1024) and (x.cleaner != null))) I find that there are roughly 400 DirectByteBuffers with a capacity of ~4MB. It also does will not go down.

I am using (mostly) default properties for the producer settings, with a linger time of 100ms and acks=all.

After restarting, this goes to 0 again, and will gradually increase unitll we restart our service.

Again, see the stackoverflow for a detailed overview of my problem with some images.

Thanks in advance,

Bram