- If cluster is big i.e >50GB and each machines has minimum 10GB resources
- If yarn-system queue doesn't exist then we shall create a queue with zero capacity.
- All the necessary setting for queue such as queue-priority, queue-acls, disabling pre-emption for this queue are set.
- ats-hbase app queue name also set to yarn-system
- But do NOT enable system service mode by default. Let upgrade happens in embedded mode successfully.
- Post upgrade, admin can change flag to enable system_service mode if required by providing sufficient capacity to yarn-system queue. This step need to be documented.
- If cluster is small
- Continue upgrade in embedded mode without creating a yarn-system queue.