Is your feature request related to a problem? Please describe.
kube-burner uses container_memory_rss (RSS) to collect metrics about memory usage.
I would suggest starting collecting container_memory_working_set_bytes (WSS). Many applications have stable and low RSS footprints, but the WSS is way bigger. Some examples are Postgres and Prometheus.
In most cases, when WSS reaches container resource limits, it will get OOMKilled. That's why I think that tracking the increase in WSS is more important than RSS.
Describe the solution you'd like
Add additional metrics to gather container_memory_working_set_bytes.
Describe alternatives you've considered
Replaces existing container_memory_rss with container_memory_working_set_bytes, but that will cause previously gathered results to be not comparable with the new ones.
Is your feature request related to a problem? Please describe.
kube-burner
usescontainer_memory_rss
(RSS
) to collect metrics about memory usage.I would suggest starting collecting
container_memory_working_set_bytes
(WSS
). Many applications have stable and lowRSS
footprints, but theWSS
is way bigger. Some examples are Postgres and Prometheus.In most cases, when
WSS
reaches container resource limits, it will get OOMKilled. That's why I think that tracking the increase inWSS
is more important thanRSS
.Describe the solution you'd like
Add additional metrics to gather
container_memory_working_set_bytes
.Describe alternatives you've considered
Replaces existing
container_memory_rss
withcontainer_memory_working_set_bytes
, but that will cause previously gathered results to be not comparable with the new ones.Additional context
Article about cgroups and OOMKilled decisions. A lot of details, but it points out what Kubernetes compares and when OOMKilled will happen.