Matthew Ebeweber

09/27/2022, 2:20 AM
Saw some errors with Cerbos today triggered by a ~moderate amount of volume. We're using grpc + a load balancer and started seeing a lot of connection refused errors similar to what is described here: Curious if you've seen this before?

Charith (Cerbos)

09/27/2022, 8:10 AM
No we haven't seen that before. I wonder whether there's a mismatch between how long the LB keeps a connection open vs the default max connection age set by Cerbos (10 minutes). Try reducing the LB keep-alive idle timeout to less than 10 minutes or alternatively, increasing the Cerbos timeout (`server.advanced.grpc.maxConnectionAge`:

Matthew Ebeweber

09/27/2022, 2:09 PM
Good callout! I’ll look into this and report back!