[YARN-2230] Fix description of yarn.scheduler.maximum-allocation-vcores in yarn-default.xml (or code)

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 2.4.0
• Fix Version/s: 2.7.0
• Component/s: client, documentation, scheduler
• Target Version/s: 2.7.0
• Assignee: Vijay Bhat
• Reporter: Adam Kawa
• Created: 28/Jun/14 23:01
• Updated: 11/Feb/16 23:20
• Resolved: 07/Jan/15 18:58

Description

When a user requests more vcores than the allocation limit (e.g. mapreduce.map.cpu.vcores is larger than yarn.scheduler.maximum-allocation-vcores), then InvalidResourceRequestException is thrown - https://svn.apache.org/repos/asf/hadoop/common/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/SchedulerUtils.java

    if (resReq.getCapability().getVirtualCores() < 0 ||
        resReq.getCapability().getVirtualCores() >
        maximumResource.getVirtualCores()) {
      throw new InvalidResourceRequestException("Invalid resource request"
          + ", requested virtual cores < 0"
          + ", or requested virtual cores > max configured"
          + ", requestedVirtualCores="
          + resReq.getCapability().getVirtualCores()
          + ", maxVirtualCores=" + maximumResource.getVirtualCores());
    }

According to documentation - yarn-default.xml http://hadoop.apache.org/docs/current/hadoop-yarn/hadoop-yarn-common/yarn-default.xml, the request should be capped to the allocation limit.

  <property>
    <description>The maximum allocation for every container request at the RM,
    in terms of virtual CPU cores. Requests higher than this won't take effect,
    and will get capped to this value.</description>
    <name>yarn.scheduler.maximum-allocation-vcores</name>
    <value>32</value>
  </property>

This means that:

• Either documentation or code should be corrected (unless this exception is handled elsewhere accordingly, but it looks that it is not).

This behavior is confusing, because when such a job (with mapreduce.map.cpu.vcores is larger than yarn.scheduler.maximum-allocation-vcores) is submitted, it does not make any progress. The warnings/exceptions are thrown at the scheduler (RM) side e.g.

2014-06-29 00:34:51,469 WARN org.apache.hadoop.yarn.server.resourcemanager.ApplicationMasterService: Invalid resource ask by application appattempt_1403993411503_0002_000001
org.apache.hadoop.yarn.exceptions.InvalidResourceRequestException: Invalid resource request, requested virtual cores < 0, or requested virtual cores > max configured, requestedVirtualCores=32, maxVirtualCores=3
	at org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerUtils.validateResourceRequest(SchedulerUtils.java:237)
	at org.apache.hadoop.yarn.server.resourcemanager.RMServerUtils.validateResourceRequests(RMServerUtils.java:80)
	at org.apache.hadoop.yarn.server.resourcemanager.ApplicationMasterService.allocate(ApplicationMasterService.java:420)
        .....
	at org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:585)
	at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:1026)
	at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1986)
	at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1982)
	at java.security.AccessController.doPrivileged(Native Method)
	at javax.security.auth.Subject.doAs(Subject.java:416)
	at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1548)
	at org.apache.hadoop.ipc.Server$Handler.run(Server.java:1980)

• IMHO, such an exception should be forwarded to client. Otherwise, it is non obvious to discover why a job does not make any progress.

The same looks to be related to memory.