Julian Taylor

actually info=pdslotreport is not even needed, the controller information contains the physical disk data already so this works on all our currently in use hardware generations. I still have to...

yes, the problem occurs before it even starts talking to the server regarding the manifests (and after it has setup its resource caches from the servers api-resources)

it is not an endpoint that is slow, it is spending all its team reading the certificate authority over and over again at 100%-150% cpu usage

here is the output with network log: ``` $ kubectl apply --dry-run=client -f manifests -v 6 |& ts Dez 05 13:31:33 I1205 13:31:33.126524 397199 loader.go:395] Config loaded from file: config...

if you give me some pointers how to create a cpu profile I can pinpoint the exact cause for you, I'm not very familiar with golang tooling.

please read my full report, reading the cluster pem does not take 11 seconds, reading the cluster pem hundreds of times (it seems twice per input manifest) and the other...

using ` --validate=false` removes the performance problem

the cluster version is 1.30.6 but it is not a cluster problem, there is no network traffic at all during this part. have you tried running my reproducer yourself? I...

why does the openapi schema need to be decoded for every single manifest? I can load the schema with python and validate all the manifests it in fractions of a...

proof of concept patch just caching the parsed schemas: ```patch --- a/staging/src/k8s.io/client-go/openapi3/root.go +++ b/staging/src/k8s.io/client-go/openapi3/root.go @@ -59,6 +59,7 @@ type Root interface { type root struct { // OpenAPI client to...