Hello there, and happy New Year, guys!
We have a 1-controller 1-router ziti network that we use to expose remote HTTP services in a way very similar to how zrok exposes them (we implemented our own “zrok” frontend, local proxy, and controller. We experienced errors under a small load, both in our expected production configuration (k8s deployed) and when deploying it in a VM with docker/compose.
To ensure the problem was not related to our proxies or our deployment, I executed the same test (25 concurrent users consuming a REST API with a small SPA download) using zrok free plan with very similar result. Probably it’s just a matter of tuning our deployment. Could you please give us some hints to identify and fix the problem?
Hi @ffaraone, this seems like a tough one for us to be able to help with to be totally honest. There seem to be a lot of questions that would need to be understood and answered and it seems like it's just out of scope to try to help. Can you reproduce the problem with simple curl statements? Or possibly produce a python/bash/powershell/go example that seems to replicate the problem so we can see what's going on?
I don't know what locust is nor how to use it. "broken pipe" errors surely seem to indicate to me that something isn't happy along the path and is terminating the connection immediately. Those sorts of errors are often https/tls related with bad certs. 
I don't know if i have anything really to help here.
Hi @TheLumberjack, yeah, I understand, with so a poor description it's hard to help me. I repeated the same load test with the free Zrok account, and I got no errors. So your comment about bad certificate errors made me think about cryptographic operations and the lack of CPU. In fact, looks like under pressure, my setup reaches almost 100% of CPU usage, and starting from there, some request fails. So you helped me! Thanks!
1 Like