Closed mccheah closed 7 years ago
I created this PR so that I didn't have to overwrite the original branch from #486. The rebase was pretty tricky so I want to keep the old history around just in case.
Tests are broken because ExecutorPodFactory doesn't consider the mounted empty dirs. I'll work on fixing that.
@mccheah ready to review? Any particular parts to focus on?
still haven't fixed the tests yet
@ash211 @kimoonkim ready for review again. I changed the architecture a bit to make testing easier.
(killed the integration test run on the intermediate commit to save time)
Are there plans to merge this to spark master?
Rebased version of #486.
Closes #439.
This is extremely important for performance, especially in shuffle-heavy computations where the executors perform a large amount of disk I/O. We only provision these volumes in static allocation mode without using the shuffle service because using a shuffle service requires mounting hostPath volumes, instead.