Massive performance degredation from 0.8.0.109 -> 0.9.0.49

[richslaughter]

Massive performance degredation from 0.8.0.109 -> 0.9.0.49

[aclemmensen]

I've been experiencing something similar, although with a much lower throughput level than described here. I've been using kafka-net inside an ASP.NET Web API application. I was experiencing frequent producer timeouts when publishing messages individually so I implemented a batching layer after experimenting in a console application.

Initial results were good, but after fully implementing it I could see timings for the production application's endpoints getting progressively worse:

Expected worst-case timings are in the 150-200 msec range. Restarting/recycling the application returns things to normal but timings start climbing again right away.

After experimenting with tweaks and profiling for hours, including some things that are mentioned in other issues here, I've given up and deployed version 0.8.0.114 which has been working perfectly so far.

[hsubram1]

Can you let me know how was the performance measurement done ? Was it via some tool like Sematext SPM?

[hsubram1]

Which tool did you use it to to run these Performance tests?

[aclemmensen]

This was observed in a production system and timings are collected through Graphite. Our message sizes are fairly large (can be multiple MB, but in most cases will probably be 50-300 KB). I don't know if this would influence performance.