Bump base memory requirements for python and go #3473
Conversation
- When auto-instrumenting applications, I have noticed that default memory
limits are too tight for some languages. This leads to the
following:
- Intermitent OOMKilled events in init container when auto-instrumenting python applications.
Eventually the pods are able to start.
- OOMKilled events for sidecar containers in go applications. The
pods are not able to start.
- 64Mi seems to be enough to fix these issues. While some tweaking by users may still be necessary,
the operator should work out-of-the-box for all supported languages.
|
Can you reproduce the OOMKills in a kind cluster? I'm not against bumping the requirements, but I don't understand why these containers would need it. The only thing they actually do is copy some files over to a shared volume. |
|
I am able to consistently reproduce the go OOMs with both k3d and kind:
Interestingly, the python OOMs only happen with k3d, not kind.
I am using a standard setup following the README. For testing, I use the rolldice applications with their manual instrumentation code stripped out. (I will move this to an issue and link it in the changelog) |
There was a problem hiding this comment.
This looks good to me, and I'm fine merging it as-is. But I'd like to understand why our e2e tests don't catch it, and ideally modify them to do so. @jcreixell could you open a follow-up issue to track that?
Description:
Link to tracking Issue(s):
#3479