Add ability to use custom kube-dns zone. #593
Conversation
The default `cluster.local` zone is not something set in stone, so if `kube-dns` is configured to use a different zone, Spark needs to know about it in order to allow the executors reach the driver.
|
I have no idea why the integration tests have crapped out. I don't think it's related to my change, but if it is, I'd be grateful to get some pointers to what's happening. |
|
Those integration test failures aren’t related to your additions. That failure is sporadically seen across pull requests. |
|
@f355 - have you considered just leaving the dns zone entirely out of the equation? If the search path is setup correctly, we shouldn't need it at all. |
|
@foxish you mean I think if we go with your suggestion by default, and allow to specify a domain for the weird cases where it's needed, we should be good. But before I jump to implementing it, I'd like to hear from the maintainers. |
|
Just fixed in upstream in apache#20187 and confirmed with SIG network that it was an acceptable way to proceed. Thanks for sending this PR btw, we wouldn't have discovered this latent issue if it hadn't been brought up here. Cheers! |
|
Awesome, thank you! |
3 participants
What changes were proposed in this pull request?
The default
cluster.localzone is not something set in stone, so ifkube-dnsis configured to use a different zone, Spark needs to know about it in order to allow the executors reach the driver.How was this patch tested?
The patch includes changes to two unit tests, both are passing. It is hard to integration-test this patch because that would require a custom kubernetes setup, however, current integration tests pass while using the default value for the added option.