HDDS-2592 Add Datanode command to allow the datanode to persist its admin state

hadoop-hdds/common/src/main/java/org/apache/hadoop/hdds/protocol/DatanodeDetails.java

@@ -49,6 +49,8 @@ public class DatanodeDetails extends NodeImpl implements
   private String hostName;
   private List<Port> ports;
   private String certSerialId;
+  private HddsProtos.NodeOperationalState persistedOpState;
+  private long persistedOpStateExpiryEpochSec = 0;
 
   /**
    * Constructs DatanodeDetails instance. DatanodeDetails.Builder is used
@@ -59,15 +61,23 @@ public class DatanodeDetails extends NodeImpl implements
    * @param networkLocation DataNode's network location path
    * @param ports Ports used by the DataNode
    * @param certSerialId serial id from SCM issued certificate.
+   * @param persistedOpState Operational State stored on DN.
+   * @param persistedOpStateExpiryEpochSec Seconds after the epoch the stored
+   *                                       state should expire.
    */
+  @SuppressWarnings("checkstyle:ParameterNumber")
   private DatanodeDetails(String uuid, String ipAddress, String hostName,
-      String networkLocation, List<Port> ports, String certSerialId) {
+      String networkLocation, List<Port> ports, String certSerialId,
+      HddsProtos.NodeOperationalState persistedOpState,
+      long persistedOpStateExpiryEpochSec) {
     super(hostName, networkLocation, NetConstants.NODE_COST_DEFAULT);
     this.uuid = UUID.fromString(uuid);
     this.ipAddress = ipAddress;
     this.hostName = hostName;
     this.ports = ports;
     this.certSerialId = certSerialId;
+    this.persistedOpState = persistedOpState;
+    this.persistedOpStateExpiryEpochSec = persistedOpStateExpiryEpochSec;
   }
 
   protected DatanodeDetails(DatanodeDetails datanodeDetails) {
@@ -78,6 +88,9 @@ protected DatanodeDetails(DatanodeDetails datanodeDetails) {
     this.hostName = datanodeDetails.hostName;
     this.ports = datanodeDetails.ports;
     this.setNetworkName(datanodeDetails.getNetworkName());
+    this.persistedOpState = datanodeDetails.getPersistedOpState();
+    this.persistedOpStateExpiryEpochSec =
+        datanodeDetails.getPersistedOpStateExpiryEpochSec();
   }
 
   /**
@@ -155,6 +168,46 @@ public List<Port> getPorts() {
     return ports;
   }
 
+  /**
+   * Return the persistedOpState. If the stored value is null, return the
+   * default value of IN_SERVICE.
+   *
+   * @return The OperationalState persisted on the datanode.
+   */
+  public HddsProtos.NodeOperationalState getPersistedOpState() {
+    if (persistedOpState == null) {
+      return HddsProtos.NodeOperationalState.IN_SERVICE;
+    } else {
+      return persistedOpState;
+    }
+  }
+
+  /**
+   * Set the persistedOpState for this instance.
+   *
+   * @param state The new operational state.
+   */
+  public void setPersistedOpState(HddsProtos.NodeOperationalState state) {
+    this.persistedOpState = state;
+  }
+
+  /**
+   * Get the persistedOpStateExpiryEpochSec for the instance.
+   * @return Seconds from the epoch when the operational state should expire.
+   */
+  public long getPersistedOpStateExpiryEpochSec() {
+    return persistedOpStateExpiryEpochSec;
+  }
+
+  /**
+   * Set persistedOpStateExpiryEpochSec.
+   * @param expiry The number of second after the epoch the operational state
+   *               should expire.
+   */
+  public void setPersistedOpStateExpiryEpochSec(long expiry) {
+    this.persistedOpStateExpiryEpochSec = expiry;
+  }
+
   /**
    * Given the name returns port number, null if the asked port is not found.
    *
@@ -200,6 +253,13 @@ public static DatanodeDetails getFromProtoBuf(
     if (datanodeDetailsProto.hasNetworkLocation()) {
       builder.setNetworkLocation(datanodeDetailsProto.getNetworkLocation());
     }
+    if (datanodeDetailsProto.hasPersistedOpState()) {
+      builder.setPersistedOpState(datanodeDetailsProto.getPersistedOpState());
+    }
+    if (datanodeDetailsProto.hasPersistedOpStateExpiry()) {
+      builder.setPersistedOpStateExpiry(
+          datanodeDetailsProto.getPersistedOpStateExpiry());
+    }
     return builder.build();
   }
 
@@ -226,6 +286,10 @@ public HddsProtos.DatanodeDetailsProto getProtoBufMessage() {
     if (!Strings.isNullOrEmpty(getNetworkLocation())) {
       builder.setNetworkLocation(getNetworkLocation());
     }
+    if (persistedOpState != null) {
+      builder.setPersistedOpState(persistedOpState);
+    }
+    builder.setPersistedOpStateExpiry(persistedOpStateExpiryEpochSec);
 
     for (Port port : ports) {
       builder.addPorts(HddsProtos.Port.newBuilder()
@@ -246,6 +310,8 @@ public String toString() {
         ", networkLocation: " +
         getNetworkLocation() +
         ", certSerialId: " + certSerialId +
+        ", persistedOpState: " + persistedOpState +
+        ", persistedOpStateExpiryEpochSec: " + persistedOpStateExpiryEpochSec +
         "}";
   }
 
@@ -285,6 +351,8 @@ public static final class Builder {
     private String networkLocation;
     private List<Port> ports;
     private String certSerialId;
+    private HddsProtos.NodeOperationalState persistedOpState;
+    private long persistedOpStateExpiryEpochSec = 0;
 
     /**
      * Default private constructor. To create Builder instance use
@@ -373,6 +441,31 @@ public Builder setCertSerialId(String certId) {
       return this;
     }
 
+    /**
+     * Adds persistedOpState.
+     *
+     * @param state The operational state persisted on the datanode
+     *
+     * @return DatanodeDetails.Builder
+     */
+    public Builder setPersistedOpState(HddsProtos.NodeOperationalState state) {
+      this.persistedOpState = state;
+      return this;
+    }
+
+    /**
+     * Adds persistedOpStateExpiryEpochSec.
+     *
+     * @param expiry The seconds after the epoch the operational state should
+     *              expire.
+     *
+     * @return DatanodeDetails.Builder
+     */
+    public Builder setPersistedOpStateExpiry(long expiry) {
+      this.persistedOpStateExpiryEpochSec = expiry;
+      return this;
+    }
+
     /**
      * Builds and returns DatanodeDetails instance.
      *
@@ -384,7 +477,8 @@ public DatanodeDetails build() {
         networkLocation = NetConstants.DEFAULT_RACK;
       }
       DatanodeDetails dn = new DatanodeDetails(id, ipAddress, hostName,
-          networkLocation, ports, certSerialId);
+          networkLocation, ports, certSerialId, persistedOpState,
+          persistedOpStateExpiryEpochSec);
       if (networkName != null) {
         dn.setNetworkName(networkName);
       }

hadoop-hdds/common/src/main/proto/hdds.proto

@@ -37,6 +37,8 @@ message DatanodeDetailsProto {
     // network name, can be Ip address or host name, depends
     optional string networkName = 6;
     optional string networkLocation = 7; // Network topology location
+    optional NodeOperationalState persistedOpState = 8; // The Operational state persisted in the datanode.id file
+    optional int64 persistedOpStateExpiry = 9; // The seconds after the epoch when the OpState should expire
 }
 
 /**

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/common/helpers/DatanodeIdYaml.java

@@ -29,6 +29,7 @@
 import org.apache.commons.collections.CollectionUtils;
 import org.apache.commons.collections.MapUtils;
 import org.apache.hadoop.hdds.protocol.DatanodeDetails;
+import org.apache.hadoop.hdds.protocol.proto.HddsProtos;
 import org.yaml.snakeyaml.DumperOptions;
 import org.yaml.snakeyaml.Yaml;
 
@@ -82,6 +83,12 @@ public static DatanodeDetails readDatanodeIdFile(File path)
           .setIpAddress(datanodeDetailsYaml.getIpAddress())
           .setHostName(datanodeDetailsYaml.getHostName())
           .setCertSerialId(datanodeDetailsYaml.getCertSerialId());
+      if (datanodeDetailsYaml.getPersistedOpState() != null) {
+        builder.setPersistedOpState(HddsProtos.NodeOperationalState.valueOf(
+            datanodeDetailsYaml.getPersistedOpState()));
+      }
+      builder.setPersistedOpStateExpiry(
+          datanodeDetailsYaml.getPersistedOpStateExpiryEpochSec());
 
       if (!MapUtils.isEmpty(datanodeDetailsYaml.getPortDetails())) {
         for (Map.Entry<String, Integer> portEntry :
@@ -105,6 +112,8 @@ public static class DatanodeDetailsYaml {
     private String ipAddress;
     private String hostName;
     private String certSerialId;
+    private String persistedOpState;
+    private long persistedOpStateExpiryEpochSec = 0;
     private Map<String, Integer> portDetails;
 
     public DatanodeDetailsYaml() {
@@ -113,11 +122,15 @@ public DatanodeDetailsYaml() {
 
     private DatanodeDetailsYaml(String uuid, String ipAddress,
                                 String hostName, String certSerialId,
+                                String persistedOpState,
+                                long persistedOpStateExpiryEpochSec,
                                 Map<String, Integer> portDetails) {
       this.uuid = uuid;
       this.ipAddress = ipAddress;
       this.hostName = hostName;
       this.certSerialId = certSerialId;
+      this.persistedOpState = persistedOpState;
+      this.persistedOpStateExpiryEpochSec = persistedOpStateExpiryEpochSec;
       this.portDetails = portDetails;
     }
 
@@ -137,6 +150,14 @@ public String getCertSerialId() {
       return certSerialId;
     }
 
+    public String getPersistedOpState() {
+      return persistedOpState;
+    }
+
+    public long getPersistedOpStateExpiryEpochSec() {
+      return persistedOpStateExpiryEpochSec;
+    }
+
     public Map<String, Integer> getPortDetails() {
       return portDetails;
     }
@@ -157,6 +178,14 @@ public void setCertSerialId(String certSerialId) {
       this.certSerialId = certSerialId;
     }
 
+    public void setPersistedOpState(String persistedOpState) {
+      this.persistedOpState = persistedOpState;
+    }
+
+    public void setPersistedOpStateExpiryEpochSec(long opStateExpiryEpochSec) {
+      this.persistedOpStateExpiryEpochSec = opStateExpiryEpochSec;
+    }
+
     public void setPortDetails(Map<String, Integer> portDetails) {
       this.portDetails = portDetails;
     }
@@ -172,11 +201,17 @@ private static DatanodeDetailsYaml getDatanodeDetailsYaml(
       }
     }
 
+    String persistedOpString = null;
+    if (datanodeDetails.getPersistedOpState() != null) {
+      persistedOpString = datanodeDetails.getPersistedOpState().name();
+    }
     return new DatanodeDetailsYaml(
         datanodeDetails.getUuid().toString(),
         datanodeDetails.getIpAddress(),
         datanodeDetails.getHostName(),
         datanodeDetails.getCertSerialId(),
+        persistedOpString,
+        datanodeDetails.getPersistedOpStateExpiryEpochSec(),
         portDetails);
   }
 }

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/common/statemachine/DatanodeStateMachine.java

@@ -50,6 +50,7 @@
     .DeleteContainerCommandHandler;
 import org.apache.hadoop.ozone.container.common.statemachine.commandhandler
     .ReplicateContainerCommandHandler;
+import org.apache.hadoop.ozone.container.common.statemachine.commandhandler.SetNodeOperationalStateCommandHandler;
 import org.apache.hadoop.ozone.container.keyvalue.TarContainerPacker;
 import org.apache.hadoop.ozone.container.ozoneimpl.OzoneContainer;
 import org.apache.hadoop.ozone.container.replication.ContainerReplicator;
@@ -138,6 +139,7 @@ public DatanodeStateMachine(DatanodeDetails datanodeDetails,
             dnConf.getContainerDeleteThreads()))
         .addHandler(new ClosePipelineCommandHandler())
         .addHandler(new CreatePipelineCommandHandler(conf))
+        .addHandler(new SetNodeOperationalStateCommandHandler(conf))
         .setConnectionManager(connectionManager)
         .setContainer(container)
         .setContext(context)