1, gracefully decommission a node A
2, restart node A
3, node A could not register to RM
(Umbrella) Support graceful decommission of nodemanager
This message was automatically generated.
./bin/yarn node -list -states DECOMMISSIONING couldn't get the decommissioning node, update the patch to fix
Thanks, sandflee. The patch looks good to me. Could you please add tests to cover the scenario the patch addresses?
Thanks Daniel Templeton, I have add a test to the patch, and test failure is not related to the patch
hi, Daniel Templeton,Junping Du could you help to review this? thanks
v4 patch LGTM. Jenkins test seems to have problem. Kick off again manually. +1 pending on Jenkins result.
The build failure is due to rm.NMwaitForState() is actually not define yet. sandflee, I think you may want to use rm.waitForState(). Isn't it?
Put a patch with this minor fix and adjust some import. Will commit if Mr. Jenkins give it a +1.
Thanks Junping Du, the test failure seems not related, could run pass locally, file YARN-5317,YARN-5318 to track
Agree. I verified locally that two tests can pass so it just failed intermittently and not related to patch here. +1. Committing v5 patch in.
SUCCESS: Integrated in Hadoop-trunk-Commit #10067 (See https://builds.apache.org/job/Hadoop-trunk-Commit/10067/)
YARN-4939. The decommissioning Node should keep alive during NM restart. (junping_du: rev 30ee57ceb1e80c30ea3adfe7736d4d4c7d5c8386)
I have commit the patch to trunk, branch-2 and branch-2.8. Thanks sandflee for patch contribution!
Thanks Junping and Daniel for reviewing and committing.