[YARN-2573] Integrate ReservationSystem with the RM failover mechanism

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.8.0, 3.0.0-alpha1
• Component/s: capacityscheduler, fairscheduler, resourcemanager
• Assignee: Subramaniam Krishnan
• Reporter: Subramaniam Krishnan
• Created: 19/Sep/14 19:30
• Updated: 30/Aug/16 01:30
• Resolved: 28/Oct/15 20:33

Description

YARN-1051 introduces the ReservationSystem and the current implementation is completely in-memory based. YARN-149 brings in the notion of RM HA with a highly available state store. This JIRA proposes persisting the Plan into the RMStateStore and recovering it post RM failover

Issue Links

• relates to: YARN-5326 Support for recurring reservations in the YARN ReservationSystem

Sub-Tasks

• Add RMStateStore apis to store and load accepted reservations for failover: Resolved
• Add support for recovery of reserved apps (running under dynamic queues) to Fair Scheduler: Resolved
• Add support for recovery of reserved apps running under dynamic queues: Resolved
• Add reservation system recovery to RM recovery process: Resolved
• Reduce storage footprint for ReservationAllocation: Resolved
• Make ReservationSystem persist state using RMStateStore reservation APIs: Resolved
• Remove update reservation state api from state store as its not used by ReservationSystem: Resolved