YARN-7458. TestContainerManagerSecurity is still flakey

(Contributed by Robert Kanter via Daniel Templeton)

Change-Id: Ibb1975ad086c3a33f8af0b4f8b9a13c3cdca3f7d
This commit is contained in:
Daniel Templeton 2017-11-08 17:31:14 -08:00
parent 0de10680b7
commit 49b4c0b334
1 changed files with 24 additions and 14 deletions

View File

@ -28,7 +28,9 @@
import java.util.Collection;
import java.util.LinkedList;
import java.util.List;
import java.util.concurrent.TimeoutException;
import com.google.common.base.Supplier;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.CommonConfigurationKeysPublic;
import org.apache.hadoop.io.DataInputBuffer;
@ -36,6 +38,7 @@
import org.apache.hadoop.net.NetUtils;
import org.apache.hadoop.security.UserGroupInformation;
import org.apache.hadoop.security.token.SecretManager.InvalidToken;
import org.apache.hadoop.test.GenericTestUtils;
import org.apache.hadoop.yarn.api.ContainerManagementProtocol;
import org.apache.hadoop.yarn.api.protocolrecords.GetContainerStatusesRequest;
import org.apache.hadoop.yarn.api.protocolrecords.GetContainerStatusesResponse;
@ -49,6 +52,7 @@
import org.apache.hadoop.yarn.api.records.ContainerId;
import org.apache.hadoop.yarn.api.records.ContainerLaunchContext;
import org.apache.hadoop.yarn.api.records.ContainerState;
import org.apache.hadoop.yarn.api.records.ContainerStatus;
import org.apache.hadoop.yarn.api.records.NodeId;
import org.apache.hadoop.yarn.api.records.Priority;
import org.apache.hadoop.yarn.api.records.Resource;
@ -404,27 +408,33 @@ private void testNMTokens(Configuration testConf) throws Exception {
newContainerToken, attempt1NMToken, false).isEmpty());
}
private void waitForContainerToFinishOnNM(ContainerId containerId) {
private void waitForContainerToFinishOnNM(ContainerId containerId)
throws TimeoutException, InterruptedException {
Context nmContext = yarnCluster.getNodeManager(0).getNMContext();
int interval = 4 * 60; // Max time for container token to expire.
Assert.assertNotNull(nmContext.getContainers().containsKey(containerId));
// Get the container first, as it may be removed from the Context
// by asynchronous calls.
// This was leading to a flakey test as otherwise the container could
// be removed and end up null.
// If the container is null, then it has already completed and been removed
// from the Context by asynchronous calls.
Container waitContainer = nmContext.getContainers().get(containerId);
while ((interval-- > 0)
&& !waitContainer.cloneAndGetContainerStatus()
.getState().equals(ContainerState.COMPLETE)) {
if (waitContainer != null) {
try {
LOG.info("Waiting for " + containerId + " to complete.");
Thread.sleep(1000);
} catch (InterruptedException e) {
LOG.info("Waiting for " + containerId + " to get to state " +
ContainerState.COMPLETE);
GenericTestUtils.waitFor(new Supplier<Boolean>() {
@Override
public Boolean get() {
return ContainerState.COMPLETE.equals(
waitContainer.cloneAndGetContainerStatus().getState());
}
}, 10, interval);
} catch (TimeoutException te) {
fail("Was waiting for " + containerId + " to get to state " +
ContainerState.COMPLETE + " but was in state " +
waitContainer.cloneAndGetContainerStatus().getState() +
" after the timeout");
}
}
// Normally, Containers will be removed from NM context after they are
// explicitly acked by RM. Now, manually remove it for testing.
yarnCluster.getNodeManager(0).getNodeStatusUpdater()