inherit IO to get kubectl logs output

[foxish]

@ash211 @mccheah I was thinking about the logging that we have, where we redirect output to files.

• There are no imposed file size limits
• Fetching it for inspection isn't currently very easy.

Using the container's stdout/stderr instead of redirecting to file the way we do it currently may be advantageous, because it has the logic to rotate logs, and keep 5 of the most recent ones (each 10M in size). This also allows for the use of kubectl logs directly, which would be analogous to yarn logs.

[foxish]

Just throwing this out there. The change appears to work for me. Was there some reason why this was not considered a viable alternative?

[foxish]

The biggest advantage is log streaming using kubectl logs -f <pod-name> which is convenient.

[mccheah]

Yup this is a good idea.

[ash211]

Looks nice to me -- @mccheah would you like to merge?

kubectl logs -f <pod-name> is super useful!