Examples with IN_MAINTENANCE org.apache.hadoop.hdds.protocol.proto.HddsProtos.NodeOperationalState.IN_MAINTENANCE used on opensource projects

Example 1 with IN_MAINTENANCE

use of org.apache.hadoop.hdds.protocol.proto.HddsProtos.NodeOperationalState.IN_MAINTENANCE in project ozone by apache.

the class TestReplicationManager method testOverReplicatedClosedContainerWithDecomAndMaint.

/**
 * When a CLOSED container is over replicated, ReplicationManager
 * deletes the excess replicas. While choosing the replica for deletion
 * ReplicationManager should not attempt to remove a DECOMMISSION or
 * MAINTENANCE replica.
 */
@Test
public void testOverReplicatedClosedContainerWithDecomAndMaint() throws IOException {
    final ContainerInfo container = createContainer(LifeCycleState.CLOSED);
    addReplica(container, NodeStatus.inServiceHealthy(), CLOSED);
    addReplica(container, new NodeStatus(DECOMMISSIONED, HEALTHY), CLOSED);
    addReplica(container, new NodeStatus(IN_MAINTENANCE, HEALTHY), CLOSED);
    addReplica(container, NodeStatus.inServiceHealthy(), CLOSED);
    addReplica(container, NodeStatus.inServiceHealthy(), CLOSED);
    addReplica(container, NodeStatus.inServiceHealthy(), CLOSED);
    addReplica(container, NodeStatus.inServiceHealthy(), CLOSED);
    final int currentDeleteCommandCount = datanodeCommandHandler.getInvocationCount(SCMCommandProto.Type.deleteContainerCommand);
    replicationManager.processAll();
    eventQueue.processAll(1000);
    Assert.assertEquals(currentDeleteCommandCount + 2, datanodeCommandHandler.getInvocationCount(SCMCommandProto.Type.deleteContainerCommand));
    Assert.assertEquals(currentDeleteCommandCount + 2, replicationManager.getMetrics().getNumDeletionCmdsSent());
    Assert.assertEquals(1, replicationManager.getInflightDeletion().size());
    Assert.assertEquals(1, replicationManager.getMetrics().getInflightDeletion());
    // Get the DECOM and Maint replica and ensure none of them are scheduled
    // for removal
    Set<ContainerReplica> decom = containerStateManager.getContainerReplicas(container.containerID()).stream().filter(r -> r.getDatanodeDetails().getPersistedOpState() != IN_SERVICE).collect(Collectors.toSet());
    for (ContainerReplica r : decom) {
        Assert.assertFalse(datanodeCommandHandler.received(SCMCommandProto.Type.deleteContainerCommand, r.getDatanodeDetails()));
    }
    assertOverReplicatedCount(1);
}

Example 2 with IN_MAINTENANCE

use of org.apache.hadoop.hdds.protocol.proto.HddsProtos.NodeOperationalState.IN_MAINTENANCE in project ozone by apache.

the class TestDecommissionAndMaintenance method testMaintenanceEndsAutomaticallyAtTimeout.

@Test
public // and new replicas created.
void testMaintenanceEndsAutomaticallyAtTimeout() throws Exception {
    // Generate some data on the empty cluster to create some containers
    generateData(20, "key", ReplicationFactor.THREE, ReplicationType.RATIS);
    ContainerInfo container = waitForAndReturnContainer();
    DatanodeDetails dn = getOneDNHostingReplica(getContainerReplicas(container));
    scmClient.startMaintenanceNodes(Arrays.asList(getDNHostAndPort(dn)), 0);
    waitForDnToReachPersistedOpState(dn, IN_MAINTENANCE);
    long newEndTime = System.currentTimeMillis() / 1000 + 5;
    // Update the maintenance end time via NM manually. As the current
    // decommission interface only allows us to specify hours from now as the
    // end time, that is not really suitable for a test like this.
    nm.setNodeOperationalState(dn, IN_MAINTENANCE, newEndTime);
    waitForDnToReachOpState(dn, IN_SERVICE);
    waitForDnToReachPersistedOpState(dn, IN_SERVICE);
    // Put the node back into maintenance and then stop it and wait for it to
    // go dead
    scmClient.startMaintenanceNodes(Arrays.asList(getDNHostAndPort(dn)), 0);
    waitForDnToReachPersistedOpState(dn, IN_MAINTENANCE);
    cluster.shutdownHddsDatanode(dn);
    waitForDnToReachHealthState(dn, DEAD);
    newEndTime = System.currentTimeMillis() / 1000 + 5;
    nm.setNodeOperationalState(dn, IN_MAINTENANCE, newEndTime);
    waitForDnToReachOpState(dn, IN_SERVICE);
    // Ensure there are 3 replicas not including the dead node, indicating a new
    // replica was created
    GenericTestUtils.waitFor(() -> getContainerReplicas(container).stream().filter(r -> !r.getDatanodeDetails().equals(dn)).count() == 3, 200, 30000);
}