Pod tolerations cause recreate loop with recreate updateStrategy

[ikreymer]

I noticed an issue when using CompositeController mode with controlling pods directly: it seems that defining the tolerations field can cause a pod to go into a recreate loop if the tolerations aren't exactly matched.
Since kubernetes adds tolerations like node.kubernetes.io/not-ready and node.kubernetes.io/unreachable automatically, a CompositeController that manages pods with Recreate will enter into a recreate loop if these tolerations aren't defined, but other tolerations are.

It seems like Recreate should ignore fields like tolerations, which are designed to be added to pods while they're running?
Or am I missing something?

It appears that I can work around this by adding the node.kubernetes.io/not-ready and node.kubernetes.io/unreachable tolerations explicitly to the child pods (or not using tolerations at all).

[grzesuav]

hmm I guess it has something to do with how metacontroller try to smart-detect merge keys for list #562 (comment)

[grzesuav]

@ikreymer can you attach metacontroller logs with debug level 7 ? Or simple case to recreate ? I will see if it is possible to mitigate it