[YARN-914] (Umbrella) Support graceful decommission of nodemanager

• Type: New Feature
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.0.4-alpha
• Component/s: graceful
• Assignee: Junping Du
• Reporter: Luke Lu
• Created: 10/Jul/13 23:22
• Updated: 25/Oct/19 20:26

Description

When NMs are decommissioned for non-fault reasons (capacity change etc.), it's desirable to minimize the impact to running applications.

Currently if a NM is decommissioned, all running containers on the NM need to be rescheduled on other NMs. Further more, for finished map tasks, if their map output are not fetched by the reducers of the job, these map tasks will need to be rerun as well.

We propose to introduce a mechanism to optionally gracefully decommission a node manager.

Issue Links

• depends upon: YARN-311 Dynamic node resource configuration: core scheduler changes
• duplicates: YARN-671 Add an interface on the RM to move NMs into a maintenance state
• is blocked by: YARN-291 [Umbrella] Dynamic resource configuration
• is blocked by: HADOOP-9160 Adopt Jolokia as the JMX HTTP/JSON bridge.
• is related to: YARN-671 Add an interface on the RM to move NMs into a maintenance state
• is related to: YARN-4939 the decommissioning Node should keep alive if NM restart
• is related to: YARN-1996 Provide alternative policies for UNHEALTHY nodes.
• relates to: MAPREDUCE-5381 Support graceful decommission of tasktracker

Sub-Tasks

• Server-Side NM Graceful Decommissioning with RM HA: Patch Available
• Server-Side NM Graceful Decommissioning subsequent call behavior: Open
• Multiple format support (JSON, etc.) for exclude node file in NM graceful decommission with timeout: Open
• Better utilize gracefully decommissioning node managers: Open
• An easy method to exclude a nodemanager from the yarn cluster cleanly: Open
• Add option to graceful decommission to not wait for applications: Patch Available