Skip to content

Should EG skip publishing for non-retriable errors at the provider layer, as well as errors from the Gatway API translator or the xDS translator? #6448

Description

@zhaohuabing

the goal of the provider layer is to an accurate snapshot of the resources, the only reason it may be unable to do so is because of any faulty I/O, in that case we skip a publish . In the future we can improve this to skip public per GC, which is a bigger design change

Originally posted by @arkodg in #6299 (comment)

Also mentioned in this comment from @guydc #6299 (comment)

Notes from the community meeting:

  • Provider layer should publish an accurate state-of-the-world.
  • If transient errors make it impossible to build an accurate snapshot, the reconciliation should be retried by returning early with an error
  • If validation errors are encountered, state should be published, so that status can be updated and user notified about issues with resources.
  • If users want to implement alternative strategies for handling invalid configuration (e.g. pausing XDS updates), that should be implemented in other layers.

Metadata

Metadata

Assignees

No one assigned

    Labels

    kind/decisionA record of a decision made by the community.provider/kubernetesIssues related to the Kubernetes providerstale

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions