[
https://issues.apache.org/jira/browse/YARN-7561?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16269324#comment-16269324
]
Robert Kanter commented on YARN-7561:
-------------------------------------
That's just the way it works: you have to *always* specify the "less specific"
request(s) when specifying a request. For example, if you want a rack, you
have to specify ANY, and if you want a node, you have to specify a rack and
ANY. So you effectively always have to specify at least the ANY request no
matter what.
As to _why_ it works that way? I'm not sure - it looks to have been that way
for quite a long time, possibly since the beginning. It does seem like a
rather confusing requirement.
> Why hasContainerForNode() return false directly when there is no request of
> ANY locality without considering NODE_LOCAL and RACK_LOCAL?
> ---------------------------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-7561
> URL: https://issues.apache.org/jira/browse/YARN-7561
> Project: Hadoop YARN
> Issue Type: Task
> Components: fairscheduler
> Affects Versions: 2.7.3
> Reporter: wuchang
>
> I am studying the FairScheduler source cod of yarn 2.7.3.
> By the code of class FSAppAttempt:
> {code}
> public boolean hasContainerForNode(Priority prio, FSSchedulerNode node) {
> ResourceRequest anyRequest = getResourceRequest(prio,
> ResourceRequest.ANY);
> ResourceRequest rackRequest = getResourceRequest(prio,
> node.getRackName());
> ResourceRequest nodeRequest = getResourceRequest(prio,
> node.getNodeName());
>
> return
> // There must be outstanding requests at the given priority:
> anyRequest != null && anyRequest.getNumContainers() > 0 &&
> // If locality relaxation is turned off at *-level, there must be
> a
> // non-zero request for the node's rack:
> (anyRequest.getRelaxLocality() ||
> (rackRequest != null && rackRequest.getNumContainers() > 0))
> &&
> // If locality relaxation is turned off at rack-level, there must
> be a
> // non-zero request at the node:
> (rackRequest == null || rackRequest.getRelaxLocality() ||
> (nodeRequest != null && nodeRequest.getNumContainers() > 0))
> &&
> // The requested container must be able to fit on the node:
> Resources.lessThanOrEqual(RESOURCE_CALCULATOR, null,
> anyRequest.getCapability(),
> node.getRMNode().getTotalCapability());
> }
> {code}
> I really cannot understand why when there is no anyRequest ,
> *hasContainerForNode()* return false directly without considering whether
> there is NODE_LOCAL or RACK_LOCAL requests.
> And , *AppSchedulingInfo.allocateNodeLocal()* and
> *AppSchedulingInfo.allocateRackLocal()* will also decrease the number of
> containers for *ResourceRequest.ANY*, this is another place where I feel
> confused.
> Really thanks for some prompt.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
