fishy commented on pull request #2190: URL: https://github.com/apache/thrift/pull/2190#issuecomment-652472186
@dcelasun Currently our prod is still running pre-bcae3bb (we plan to deploy bcae3bb to prod next week). I've been running prod version (old) vs. bcae3bb+this PR (new) side-by-side in our staging environment for a while now. We have 2 go services, A & B, while A is also client to B (and another Java service). Here are some graphs for the last 12hr (in all the graphs, cyan is new and green is old): A's server side latency p99:  A's server side latency p50:  B's server side latency p99:  B's server side latency p50:  A's client side latency p99 for A->B requests:  A's client side latency p50 for A->B requests:  A's goroutines (this is a ticker of 10s getting `runtime.NumGoroutine()` and report as a gauge):  B's goroutines:  (in p50 charts, the unit of y axis is ms) So yes this has an overhead on goroutines (more obvious on B as it has fewer goroutines in general), and an overhead on latency (more obvious on B's server side p50, which is constantly ~0.5ms. less obvious on all p99 charts). I think those overheads are acceptable? I'll run bcae3bb vs. this PR now and post results a few hours later. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
