envoyproxy/envoy

[http] Measure per-connection and per-request memory usage

Open

#11 421 ouverte le 3 juin 2020

Voir sur GitHub
 (3 commentaires) (0 réactions) (0 assignés)C++ (5 373 forks)batch import
area/httparea/perfhelp wanted

Métriques du dépôt

Stars
 (27 997 stars)
Métriques de merge PR
 (Merge moyen 8j) (378 PRs mergées en 30 j)

Description

It would be good to have some benchmarks that measure memory usage in the following scenarios:

  • H1 and H2 request after proxying request headers to upstream
  • H1 and H2 request after proxying response headers to client
  • Idle H1 client connection using raw sockets and TLS sockets
  • Idle H2 client connection with empty/full compression context

Implementing strict tests like test/integration/stats_integration_test.cc for codec memory usage may not be feasible due to the codecs having a higher rate of change. Use of a mechanism like the one proposed in https://github.com/envoyproxy/envoy/issues/11266 to track these performance metrics and flamegraphs of the heapz output from CI would allow us to detect and track down improvements/regressions.

@oschaaf

Guide contributeur