Closed myaser closed 1 year ago
Changes Missing Coverage | Covered Lines | Changed/Added Lines | % | ||
---|---|---|---|---|---|
internal/servicecheck/servicecheck.go | 17 | 27 | 62.96% | ||
<!-- | Total: | 25 | 35 | 71.43% | --> |
Files with Coverage Reduction | New Missed Lines | % | ||
---|---|---|---|---|
internal/kubediscovery/nodewatcher.go | 7 | 81.63% | ||
<!-- | Total: | 7 | --> |
Totals | |
---|---|
Change from base Build 3323701561: | -2.2% |
Covered Lines: | 433 |
Relevant Lines: | 552 |
Hi @myaser Thanks for your PR and catching these issues on that scale!
I (or @zbindenren ?) will probably implement these feature flags also on the helm chart later.
Hi @myaser Thanks for your PR and catching these issues on that scale!
I (or @zbindenren ?) will probably implement these feature flags also on the helm chart later.
@djboris9 No worries, I will implement this in helm. Making a PR.
@zbindenren & @djboris9 thank you very much :)
Running kubenurse at scale (~ 1000 node) can have a negative performance impact on the Kubernetes control plane for example, creating a service for the kubenurse daemonset requires CNI Providers (flannel in my case) to track ALL pods in the iptable rules of all nodes, which will be huge and may lead to increased memory consumption on flannel
Also, neighborhood checks can overload the Kubernetes API server for discovery information.
In addition, I have tools to monitor some stuff that kubenurse covers and I wanted to use only what I need from kubenurse
This PR creates control to disable some of the checks when needed, but does not change the behavior as all checks are enabled by default