This would eliminate problem when a new executor without h2o tries to join the computation
Already testing the solution in case of conversion from Spark -> H2O
The opposite conversion should work automatically as we are converting from the node on which we already have h2o
Tested on YARN where I requested 1000 executors, 168 were discovered. All run fine. Then I enforced location only for 2 executors out of 1000 to really see it works.