[YARN-10139] ValidateAndGetSchedulerConfiguration API fails when cluster max allocation > default 8GB

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.3.0
• Fix Version/s: 3.3.0, 3.2.2, 3.1.4
• Hadoop Flags: Reviewed
• Assignee: Prabhu Joseph
• Reporter: Prabhu Joseph
• Created: 14/Feb/20 15:49
• Updated: 22/May/20 13:28
• Resolved: 19/Feb/20 05:50

Description

ValidateAndGetSchedulerConfiguration fails when the cluster max allocation (yarn.scheduler.maximum-allocation-mb) is set to resource (eg: 16GB) > default 8GB in yarn-site.xml.

As part of validation API, there are two configuration used - CapacitySchedulerConfiguration and Configuration (yarn-site.xml). When CapacityScheduler is initialized with CapacitySchedulerConfiguration, as part of queues initialization, it checks the queue maximum allocation which is not present and so checks cluster max allocation which is not present (it is present only in YarnConfiguration) and defaults to 8GB. This will fail as queue max allocation 8GB is decreased from previous 16GB.

2020-02-14 07:38:46,087 WARN org.apache.hadoop.yarn.server.resourcemanager.webapp.RMWebServices: CapacityScheduler configuration validation failed:java.io.IOException: Failed to re-init queues : Trying to reinitialize root.default.c1.c3 the maximum allocation size can not be decreased! Current setting: <memory:164860, vCores:88>, trying to set it to: <memory:8192, vCores:4>

CapacityScheduler initialize code reads a yarn config from CapacitySchedulerConfiguration causing the issue.

Issue Links

• relates to: YARN-10022 Create RM Rest API to validate a CapacityScheduler Configuration