All Downloads are FREE. Search and download functionalities are using the official Maven repository.

org.apache.kudu.client.ConnectToCluster Maven / Gradle / Ivy

Go to download

org.apache.kudu:kudu-client with netty package relocations reverted and netty classes stripped away so that camel-quarkus-kudu can use quarkus-netty as a replacement

There is a newer version: 3.15.0
Show newest version
// 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.kudu.client;

import java.net.InetAddress;
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.atomic.AtomicReference;

import com.google.common.base.Functions;
import com.google.common.base.Joiner;
import com.google.common.base.Preconditions;
import com.google.common.collect.Lists;
import com.stumbleupon.async.Callback;
import com.stumbleupon.async.Deferred;
import io.netty.util.Timer;
import org.apache.yetus.audience.InterfaceAudience;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import org.apache.kudu.Common.HostPortPB;
import org.apache.kudu.consensus.Metadata.RaftPeerPB.Role;
import org.apache.kudu.master.Master.ConnectToMasterResponsePB;
import org.apache.kudu.rpc.RpcHeader.ErrorStatusPB.RpcErrorCodePB;
import org.apache.kudu.util.NetUtil;
import org.apache.kudu.util.Pair;

/**
 * Class responsible for fanning out RPCs to all of the configured masters,
 * finding a leader, and responding when the leader has been located.
 */
@InterfaceAudience.Private
final class ConnectToCluster {

  private static final Logger LOG = LoggerFactory.getLogger(ConnectToCluster.class);

  private final List masterAddrs;
  private final Deferred responseD;

  // Used to avoid calling 'responseD' twice.
  private final AtomicBoolean responseDCalled = new AtomicBoolean(false);

  /** Number of responses we've received so far */
  private final AtomicInteger countResponsesReceived = new AtomicInteger(0);

  // Exceptions received so far: kept for debugging purposes.
  private final List exceptionsReceived =
      Collections.synchronizedList(new ArrayList<>());

  /**
   * If we've received a response from a master which indicates the full
   * list of masters in the cluster, it is stored here. Otherwise, null.
   */
  private AtomicReference> knownMasters = new AtomicReference<>();

  private int numMasters;

  /**
   * Creates an object that holds the state needed to retrieve master table's location.
   * @param masterAddrs Addresses of all master replicas that we want to retrieve the
   *                    registration from.
   */
  ConnectToCluster(List masterAddrs) {
    this.masterAddrs = masterAddrs;
    this.responseD = new Deferred<>();
    this.numMasters = masterAddrs.size();
  }

  @InterfaceAudience.LimitedPrivate("Test")
  public Deferred getDeferred() {
    return responseD;
  }

  @InterfaceAudience.LimitedPrivate("Test")
  List getExceptionsReceived() {
    return exceptionsReceived;
  }

  private static Deferred connectToMaster(
      final KuduTable masterTable,
      final RpcProxy masterProxy,
      KuduRpc parentRpc,
      Timer timer,
      long defaultTimeoutMs) {
    // TODO: Handle the situation when multiple in-flight RPCs all want to query the masters,
    // basically reuse in some way the master permits.
    long timeoutMillis = parentRpc == null ? defaultTimeoutMs :
                                             parentRpc.timeoutTracker.getMillisBeforeTimeout();
    final ConnectToMasterRequest rpc =
        new ConnectToMasterRequest(masterTable, timer, timeoutMillis);
    rpc.setParentRpc(parentRpc);
    Deferred d = rpc.getDeferred();
    rpc.attempt++;
    masterProxy.sendRpc(rpc);

    // If we are connecting to an older version of Kudu, we'll get an invalid request
    // error. In that case, we resend using the older version of the RPC.
    d.addErrback(new Callback, Exception>() {
      @Override
      public Deferred call(Exception result)
          throws Exception {
        if (result instanceof RpcRemoteException) {
          RpcRemoteException rre = (RpcRemoteException)result;
          if (rre.getErrPB().getCode() == RpcErrorCodePB.ERROR_INVALID_REQUEST &&
              rre.getErrPB().getUnsupportedFeatureFlagsCount() > 0) {
            AsyncKuduClient.LOG.debug("Falling back to GetMasterRegistration() RPC to connect " +
                "to server running Kudu < 1.3.");
            final Deferred newAttempt =
                Preconditions.checkNotNull(rpc.getDeferred());
            rpc.setUseOldMethod();
            masterProxy.sendRpc(rpc);
            return newAttempt;
          }
        }
        return Deferred.fromError(result);
      }
    });

    return d;
  }

  /**
   * Locate the leader master and retrieve the cluster information
   * (see {@link ConnectToClusterResponse}.
   *
   * @param masterTable the "placeholder" table used by AsyncKuduClient
   * @param masterAddresses the addresses of masters to fetch from
   * @param parentRpc RPC that prompted a master lookup, can be null
   * @param defaultTimeoutMs timeout to use for RPCs if the parentRpc has no timeout
   * @param credentialsPolicy credentials policy to use for connection negotiation
   * @return a Deferred object for the cluster connection status
   */
  public static Deferred run(
      KuduTable masterTable,
      List masterAddresses,
      KuduRpc parentRpc,
      long defaultTimeoutMs,
      Connection.CredentialsPolicy credentialsPolicy) {
    ConnectToCluster connector = new ConnectToCluster(masterAddresses);
    connector.connectToMasters(masterTable, parentRpc,
                               defaultTimeoutMs, credentialsPolicy);
    return connector.responseD;
  }

  @InterfaceAudience.LimitedPrivate("Test")
  List> connectToMasters(
      KuduTable masterTable,
      KuduRpc parentRpc,
      long defaultTimeoutMs,
      Connection.CredentialsPolicy credentialsPolicy) {
    // Try to connect to each master. The ConnectToCluster instance
    // waits until it gets a good response before firing the returned
    // deferred.
    List> deferreds = new ArrayList<>();
    List> masterAddrsWithNames = new ArrayList<>();
    for (HostAndPort hostAndPort : masterAddrs) {
      InetAddress[] inetAddrs = NetUtil.getAllInetAddresses(hostAndPort.getHost());
      if (inetAddrs != null) {
        if (inetAddrs.length > 1) {
          LOG.info("Specified master server address {} resolved to multiple IPs {}. " +
                   "Connecting to each one of them.", hostAndPort.getHost(), inetAddrs);
        }
        for (InetAddress addr : inetAddrs) {
          masterAddrsWithNames.add(
              new Pair<>(addr, new HostAndPort(addr.getHostAddress(), hostAndPort.getPort())));
        }
      } else {
        masterAddrsWithNames.add(new Pair<>(null, hostAndPort));
      }
    }

    this.numMasters = masterAddrsWithNames.size();
    for (Pair masterPair : masterAddrsWithNames) {
      InetAddress addr = masterPair.getFirst();
      HostAndPort hostAndPort = masterPair.getSecond();
      Deferred d;
      if (addr != null) {
        AsyncKuduClient client = masterTable.getAsyncClient();
        RpcProxy proxy = client.newMasterRpcProxy(hostAndPort, addr, credentialsPolicy);
        d = connectToMaster(masterTable, proxy, parentRpc, client.getTimer(), defaultTimeoutMs);
      } else {
        String message = "Couldn't resolve this master's address " + hostAndPort.toString();
        LOG.warn(message);
        Status statusIOE = Status.IOError(message);
        d = Deferred.fromError(new NonRecoverableException(statusIOE));
      }
      d.addCallbacks(callbackForNode(hostAndPort), errbackForNode(hostAndPort));
      deferreds.add(d);
    }
    return deferreds;
  }

  /**
   * Creates a callback for a ConnectToMaster RPC that was sent to 'hostAndPort'.
   * @see ConnectToMasterCB
   * @param hostAndPort Host and port for the RPC we're attaching this to.
   *                    Host and port must be valid.
   * @return The callback object that can be added to the RPC request.
   */
  @InterfaceAudience.LimitedPrivate("Test")
  Callback callbackForNode(HostAndPort hostAndPort) {
    return new ConnectToMasterCB(hostAndPort);
  }

  /**
   * Creates an errback for a ConnectToMaster that was sent to 'hostAndPort'.
   * @see ConnectToMasterErrCB
   * @param hostAndPort Host and port for the RPC we're attaching this to.
   *                Used for debugging purposes.
   * @return The errback object that can be added to the RPC request.
   */
  @InterfaceAudience.LimitedPrivate("Test")
  Callback errbackForNode(HostAndPort hostAndPort) {
    return new ConnectToMasterErrCB(hostAndPort);
  }

  /**
   * Checks if we've already received a response or an exception from every master that
   * we've sent a ConnectToMaster to. If so -- and no leader has been found
   * (that is, 'responseD' was never called) -- pass a {@link NoLeaderFoundException}
   * to responseD.
   */
  private void incrementCountAndCheckExhausted() {
    if (countResponsesReceived.incrementAndGet() == numMasters &&
        responseDCalled.compareAndSet(false, true)) {
      // We want `allUnrecoverable` to only be true if all the masters came back with
      // NonRecoverableException so that we know for sure we can't retry anymore. Just one master
      // that replies with RecoverableException or with an ok response but is a FOLLOWER is
      // enough to keep us retrying.
      boolean allUnrecoverable = true;
      if (exceptionsReceived.size() == countResponsesReceived.get()) {
        for (Exception ex : exceptionsReceived) {
          if (!(ex instanceof NonRecoverableException)) {
            allUnrecoverable = false;
            break;
          }
        }
      } else {
        allUnrecoverable = false;
      }

      String allHosts = NetUtil.hostsAndPortsToString(masterAddrs);
      if (allUnrecoverable) {
        // This will stop retries.
        String msg = String.format("Couldn't find a valid master in (%s). " +
            "Exceptions received: [%s]", allHosts,
            Joiner.on(", ").join(Lists.transform(
                exceptionsReceived, Functions.toStringFunction())));
        Status s = Status.ServiceUnavailable(msg);
        responseD.callback(new NonRecoverableException(s));
      } else {
        // We couldn't find a leader master. A common case here is that the user only
        // specified a subset of the masters, so check for that. We could try to do
        // something fancier like compare the actual host/ports to see if they don't
        // match, but it's possible that the hostnames used by clients are not the
        // same as the hostnames that the servers use for each other in some network
        // setups.

        List knownMastersLocal = knownMasters.get();
        if (knownMastersLocal != null &&
            knownMastersLocal.size() > numMasters) {
          String msg = String.format(
              "Could not connect to a leader master. " +
              "Client configured with %s master(s) (%s) but cluster indicates it expects " +
              "%s master(s) (%s)",
              numMasters, allHosts,
              knownMastersLocal.size(),
              ProtobufHelper.hostPortPbListToString(knownMastersLocal));
          LOG.warn(msg);
          Exception e = new NonRecoverableException(Status.ConfigurationError(msg));
          if (!LOG.isDebugEnabled()) {
            // Stack trace is just internal guts of netty, etc, no need for the detail
            // level.
            e.setStackTrace(new StackTraceElement[]{});
          }
          responseD.callback(e);
          return;
        }

        String message = String.format("Master config (%s) has no leader.",
            allHosts);
        Exception ex;
        if (exceptionsReceived.isEmpty()) {
          LOG.warn("None of the provided masters {} is a leader; will retry", allHosts);
          ex = new NoLeaderFoundException(Status.ServiceUnavailable(message));
        } else {
          LOG.warn("Unable to find the leader master {}; will retry", allHosts);
          String joinedMsg = message + " Exceptions received: " +
              Joiner.on(",").join(Lists.transform(
                  exceptionsReceived, Functions.toStringFunction()));
          Status s = Status.ServiceUnavailable(joinedMsg);
          ex = new NoLeaderFoundException(s,
              exceptionsReceived.get(exceptionsReceived.size() - 1));
        }
        responseD.callback(ex);
      }
    }
  }

  /**
   * Callback for each ConnectToCluster RPC sent in connectToMaster() above.
   * If a request (paired to a specific master) returns a reply that indicates it's a leader,
   * the callback in 'responseD' is invoked with an initialized GetTableLocationResponsePB
   * object containing the leader's RPC address.
   * If the master is not a leader, increment 'countResponsesReceived': if the count equals to
   * the number of masters, pass {@link NoLeaderFoundException} into
   * 'responseD' if no one else had called 'responseD' before; otherwise, do nothing.
   */
  final class ConnectToMasterCB implements Callback {
    private final HostAndPort hostAndPort;

    public ConnectToMasterCB(HostAndPort hostAndPort) {
      this.hostAndPort = hostAndPort;
    }

    @Override
    public Void call(ConnectToMasterResponsePB r) throws Exception {
      recordKnownMasters(r);
      if (!r.getRole().equals(Role.LEADER)) {
        incrementCountAndCheckExhausted();
        return null;
      }
      // We found a leader!
      if (!responseDCalled.compareAndSet(false,  true)) {
        // Someone else already found a leader. This is somewhat unexpected
        // because this means two nodes think they're the leader, but it's
        // not impossible. We'll just ignore it.
        LOG.debug("Callback already invoked, discarding response({}) from {}", r, hostAndPort);
        return null;
      }

      responseD.callback(new ConnectToClusterResponse(hostAndPort, r));
      return null;
    }

    @Override
    public String toString() {
      return "ConnectToMasterCB for " + hostAndPort.toString();
    }

    private void recordKnownMasters(ConnectToMasterResponsePB r) {
      // Old versions don't set this field.
      if (r.getMasterAddrsCount() == 0) {
        return;
      }

      knownMasters.compareAndSet(null, r.getMasterAddrsList());
    }
  }

  /**
   * Errback for each ConnectToMaster RPC sent in connectToMaster() above.
   * Stores each exception in 'exceptionsReceived'. Increments 'countResponseReceived': if
   * the count is equal to the number of masters and no one else had called 'responseD' before,
   * pass a {@link NoLeaderFoundException} into 'responseD'; otherwise, do
   * nothing.
   */
  final class ConnectToMasterErrCB implements Callback {
    private final HostAndPort hostAndPort;

    public ConnectToMasterErrCB(HostAndPort hostAndPort) {
      this.hostAndPort = hostAndPort;
    }

    @Override
    public Void call(Exception e) throws Exception {
      LOG.info("Unable to connect to master {}: {}", hostAndPort, e.getMessage());
      exceptionsReceived.add(e);
      incrementCountAndCheckExhausted();
      return null;
    }

    @Override
    public String toString() {
      return "ConnectToMasterErrCB for " + hostAndPort.toString();
    }
  }
}




© 2015 - 2024 Weber Informatics LLC | Privacy Policy