Election issues


We are seeing an issue where a cluster running on Ubuntu 14.04.1 / vmware are ending up with a never ending master election loop. All of the servers can see and tcp connect to each other. Any ideas?

-Svein Arne

Local network or? Are they timing out?

I had this same issue in my vagrant setup…

Setting the environment variable MONO_THREADS_PER_CPU to some high number (100) was the fix.

(thanks to Matt Walkenhorst in this thread for the suggestion)

  • Rich

Ah yes should have thought to ask node size good call

This is probably worth us setting by default in the ./run-node.sh script?

Any thoughts?

Yeah but 100 is wayyyy too high maybe something like 8?

100 is high, but 8 is too low for a single or dual cpu setup. We still saw issues with 2 vCPUs at 5 threads per CPU, but were OK with 4, so I’d say 20 total threads would be about right.

an alternative would be to move away from threadpool here @james what
do you think?

It will be interesting to see what other effects this has - presumably there’s a reason the default is low. I’ll take a look and open an issue on GitHub to track this.

We will try this out. Thanks! Forgot to add email updates so never saw the answers until now.


I’ve just put in a potential fix for this in this pull request: https://github.com/EventStore/EventStore/pull/300. This will be part of the RC which will be out shortly, and it would be interesting to know if the new default fixes the issue for everyone. Failing
that it would be interesting to know what value does fix the issue for everyone (you can adjust it with --mono-min-threadpool-size at the command line/envvars/config file).