You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Title: Default worker thread number is the amount of HW threads, not the cpuset size.
Description:
If --concurrency option is not specified, the default number of worker threads depends on the number of HW threads in the system. However, this number can be overly large, if Envoy is run in an environment where the allowed cpuset is small. The original intent seems to be that the number of worker threads should scale linearly with the available CPUs.
For example, if Envoy is started in a Docker container like this, Envoy can run only on four CPU cores:
$ docker run --cpuset-cpus="0-3" envoy-container
However, on a server with two physical cores with twenty cores each and hyperthreading enabled, 80 worker threads are started. If Envoy checked the assigned cpuset size, it would be clear that only four CPU cores are assigned to Envoy. It might make more sense then to start only four worker threads.
There's an associated PR #5975 for a possible fix for the issue.
The text was updated successfully, but these errors were encountered:
Title: Default worker thread number is the amount of HW threads, not the cpuset size.
Description:
If
--concurrency
option is not specified, the default number of worker threads depends on the number of HW threads in the system. However, this number can be overly large, if Envoy is run in an environment where the allowed cpuset is small. The original intent seems to be that the number of worker threads should scale linearly with the available CPUs.For example, if Envoy is started in a Docker container like this, Envoy can run only on four CPU cores:
However, on a server with two physical cores with twenty cores each and hyperthreading enabled, 80 worker threads are started. If Envoy checked the assigned cpuset size, it would be clear that only four CPU cores are assigned to Envoy. It might make more sense then to start only four worker threads.
There's an associated PR #5975 for a possible fix for the issue.
The text was updated successfully, but these errors were encountered: