[YARN-9084] Service Upgrade: With default readiness check, the status of upgrade is reported to be successful prematurely

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.1.2, 3.3.0, 3.2.1
• Target Version/s: 3.1.2, 3.3.0, 3.2.1
• Release Note: Improve transient container status accuracy for upgrade.
• Assignee: Chandni Singh
• Reporter: Chandni Singh
• Created: 05/Dec/18 20:41
• Updated: 18/Dec/18 23:21
• Resolved: 18/Dec/18 23:08

Description

With YARN-9071 we do clear the IP address and Hostname from AM, NM and yarn registry before upgrade. However it is observed that after the container is launched again as part of reinit, the ContainerStatus received from NM has an IP and host even though the container fails as soon as it is launched.

On Yarn Service side this results in the component instance transitioning to READY state when it checks just the presence of IP address.

Issue Links

• relates to: YARN-9071 NM and service AM don't have updated status for reinitialized containers