[YARN-7146] Many RM unit tests failing with FairScheduler

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.0.0-beta1
• Fix Version/s: 2.9.0, 3.0.0-beta1, 3.1.0
• Component/s: test
• Target Version/s: 2.9.0, 3.0.0-beta1
• Hadoop Flags: Reviewed
• Assignee: Robert Kanter
• Reporter: Robert Kanter
• Created: 01/Sep/17 01:16
• Updated: 01/Nov/17 21:14
• Resolved: 01/Nov/17 21:14

Description

Many of the RM unit tests are failing when using the FairScheduler.

Here is a list of affected test classes:

TestYarnClient
TestApplicationCleanup
TestApplicationMasterLauncher
TestDecommissioningNodesWatcher
TestKillApplicationWithRMHA
TestNodeBlacklistingOnAMFailures
TestRM
TestRMAdminService
TestRMRestart
TestResourceTrackerService
TestWorkPreservingRMRestart
TestAMRMRPCNodeUpdates
TestAMRMRPCResponseId
TestAMRestart
TestApplicationLifetimeMonitor
TestNodesListManager
TestRMContainerImpl
TestAbstractYarnScheduler
TestSchedulerUtils
TestFairOrderingPolicy
TestAMRMTokens
TestDelegationTokenRenewer

Most of the test methods in these classes are failing, though some do succeed.

There's two main categories of issues:

1. The test submits an application to the MockRM and waits for it to enter a specific state, which it never does, and the test times out. We need to call update() on the scheduler.
2. The test throws a ClassCastException on FSQueueMetrics to CSQueueMetrics. This is because QueueMetrics metrics are static, and a previous test using FairScheduler initialized it, and the current test is using CapacityScheduler. We need to reset the metrics.