[
https://issues.apache.org/jira/browse/YARN-8202?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16467731#comment-16467731
]
Robert Kanter commented on YARN-8202:
-------------------------------------
Thanks for the patch. A few comments:
# Why is {{TestResourceRequest}} necessary? Can't we just directly use
{{Resource}}?
# I don't think we need to make the variables ALL CAPS in
{{RMContainerAllocator}}. While these are final, they're lists and we usually
only do ALL CAPS for "real" constants (e.g. Strings, ints, etc).
# In {{UnitsConversionUtil#checkUnitArgument}}, the check for a null unit also
unnecessarily checks that the unit is known, which is then repeated in the
second if statement. As it is now (and the original code suffered the same
issue), it's going to throw a message saying the unit is null for an unknown
unit. We should remove the unknown unit check from the first if statement.
# In {{ResourceTypesTestHelper}}, we should make the {{Pattern}} a static class
variable, so we don't have to "compile" it everytime the method is called.
# In {{SchedulerUtils#checkResourceRequestAgainstAvailableResource}}, when
doing the debug logging, we should call {{LOG.isDebugEnabled()}} first so we
avoid doing String concatenation if debug logging is not turned on.
> DefaultAMSProcessor should properly check units of requested custom resource
> types against minimum/maximum allocation
> ---------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-8202
> URL: https://issues.apache.org/jira/browse/YARN-8202
> Project: Hadoop YARN
> Issue Type: Bug
> Reporter: Szilard Nemeth
> Assignee: Szilard Nemeth
> Priority: Blocker
> Attachments: YARN-8202-001.patch, YARN-8202-002.patch,
> YARN-8202-003.patch, YARN-8202-004.patch, YARN-8202-005.patch,
> YARN-8202-006.patch, YARN-8202-007.patch, YARN-8202-008.patch
>
>
>
> When I execute a pi job with arguments:
> {code:java}
> -Dmapreduce.map.resource.memory-mb=200
> -Dmapreduce.map.resource.resource1=500M 1 1000{code}
> and I have one node with 5GB of resource1, I get the following exception on
> every second and the job hangs:
> {code:java}
> 2018-04-24 08:42:03,694 INFO org.apache.hadoop.ipc.Server: IPC Server handler
> 20 on 8030, call Call#386 Retry#0
> org.apache.hadoop.yarn.api.ApplicationMasterProtocolPB.allocate from
> 172.31.119.172:58138
> org.apache.hadoop.yarn.exceptions.InvalidResourceRequestException: Invalid
> resource request, requested resource type=[resource1] < 0 or greater than
> maximum allowed allocation. Requested resource=<memory:200, vCores:1,
> resource1: 500M>, maximum allowed allocation=<memory:6144, vCores:8,
> resource1: 5G>, please note that maximum allowed allocation is calculated by
> scheduler based on maximum resource of registered NodeManagers, which might
> be less than configured maximum allocation=<memory:8192, vCores:8192,
> resource1: 9223372036854775807G>
> at
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerUtils.validateResourceRequest(SchedulerUtils.java:286)
> at
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerUtils.normalizeAndValidateRequest(SchedulerUtils.java:242)
> at
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerUtils.normalizeAndvalidateRequest(SchedulerUtils.java:258)
> at
> org.apache.hadoop.yarn.server.resourcemanager.RMServerUtils.normalizeAndValidateRequests(RMServerUtils.java:249)
> at
> org.apache.hadoop.yarn.server.resourcemanager.DefaultAMSProcessor.allocate(DefaultAMSProcessor.java:230)
> at
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.constraint.processor.DisabledPlacementProcessor.allocate(DisabledPlacementProcessor.java:75)
> at
> org.apache.hadoop.yarn.server.resourcemanager.AMSProcessingChain.allocate(AMSProcessingChain.java:92)
> at
> org.apache.hadoop.yarn.server.resourcemanager.ApplicationMasterService.allocate(ApplicationMasterService.java:433)
> at
> org.apache.hadoop.yarn.api.impl.pb.service.ApplicationMasterProtocolPBServiceImpl.allocate(ApplicationMasterProtocolPBServiceImpl.java:60)
> at
> org.apache.hadoop.yarn.proto.ApplicationMasterProtocol$ApplicationMasterProtocolService$2.callBlockingMethod(ApplicationMasterProtocol.java:99)
> at
> org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:523)
> at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:991)
> at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:872)
> at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:818)
> at java.security.AccessController.doPrivileged(Native Method)
> at javax.security.auth.Subject.doAs(Subject.java:422)
> at
> org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1682)
> at org.apache.hadoop.ipc.Server$Handler.run(Server.java:2678)
> {code}
> *This is because
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerUtils#validateResourceRequest
> does not take resource units into account.*
>
> However, if I start a job with arguments:
> {code:java}
> -Dmapreduce.map.resource.memory-mb=200 -Dmapreduce.map.resource.resource1=1G
> 1 1000{code}
> and I still have 5GB of resource1 on one node then the job runs successfully.
>
> I also tried a third job run, when I request 1GB of resource1 and I have no
> nodes with any amount of resource1, then I restart the node with 5GBs of
> resource1, the job ultimately completes, but just after the node with enough
> resources registered in RM, which is the desired behaviour.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
