org.apache.hadoop.yarn.api.records.PreemptionMessage Maven / Gradle / Ivy
/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.yarn.api.records;
import org.apache.hadoop.classification.InterfaceAudience.Private;
import org.apache.hadoop.classification.InterfaceAudience.Public;
import org.apache.hadoop.classification.InterfaceStability.Evolving;
import org.apache.hadoop.classification.InterfaceStability.Unstable;
import org.apache.hadoop.yarn.util.Records;
/**
* A {@link PreemptionMessage} is part of the RM-AM protocol, and it is used by
* the RM to specify resources that the RM wants to reclaim from this
* ApplicationMaster (AM). The AM receives a {@link
* StrictPreemptionContract} message encoding which containers the platform may
* forcibly kill, granting it an opportunity to checkpoint state or adjust its
* execution plan. The message may also include a {@link PreemptionContract}
* granting the AM more latitude in selecting which resources to return to the
* cluster.
*
*
The AM should decode both parts of the message. The {@link
* StrictPreemptionContract} specifies particular allocations that the RM
* requires back. The AM can checkpoint containers' state, adjust its execution
* plan to move the computation, or take no action and hope that conditions that
* caused the RM to ask for the container will change.
*
*
In contrast, the {@link PreemptionContract} also includes a description of
* resources with a set of containers. If the AM releases containers matching
* that profile, then the containers enumerated in {@link
* PreemptionContract#getContainers()} may not be killed.
*
*
Each preemption message reflects the RM's current understanding of the
* cluster state, so a request to return N containers may not
* reflect containers the AM is releasing, recently exited containers the RM has
* yet to learn about, or new containers allocated before the message was
* generated. Conversely, an RM may request a different profile of containers in
* subsequent requests.
*
*
The policy enforced by the RM is part of the scheduler. Generally, only
* containers that have been requested consistently should be killed, but the
* details are not specified.
*/
@Public
@Evolving
public abstract class PreemptionMessage {
@Private
@Unstable
public static PreemptionMessage newInstance(StrictPreemptionContract set,
PreemptionContract contract) {
PreemptionMessage message = Records.newRecord(PreemptionMessage.class);
message.setStrictContract(set);
message.setContract(contract);
return message;
}
/**
* @return Specific resources that may be killed by the
* ResourceManager
*/
@Public
@Evolving
public abstract StrictPreemptionContract getStrictContract();
@Private
@Unstable
public abstract void setStrictContract(StrictPreemptionContract set);
/**
* @return Contract describing resources to return to the cluster.
*/
@Public
@Evolving
public abstract PreemptionContract getContract();
@Private
@Unstable
public abstract void setContract(PreemptionContract contract);
}