Only bootstrap and elect node in current voting configuration (#37712)

Adapts bootstrapping and leader election to only trigger on nodes that are actually part of the voting
configuration.

server/src/main/java/org/elasticsearch/cluster/coordination/ClusterBootstrapService.java

@@ -97,7 +97,7 @@ public class ClusterBootstrapService {
 
     void onFoundPeersUpdated() {
         final Set<DiscoveryNode> nodes = getDiscoveredNodes();
-        if (transportService.getLocalNode().isMasterNode() && bootstrapRequirements.isEmpty() == false
+        if (bootstrappingPermitted.get() && transportService.getLocalNode().isMasterNode() && bootstrapRequirements.isEmpty() == false
             && isBootstrappedSupplier.getAsBoolean() == false && nodes.stream().noneMatch(Coordinator::isZen1Node)) {
 
             final Tuple<Set<DiscoveryNode>,List<String>> requirementMatchingResult;
@@ -114,6 +114,13 @@ public class ClusterBootstrapService {
             logger.trace("nodesMatchingRequirements={}, unsatisfiedRequirements={}, bootstrapRequirements={}",
                 nodesMatchingRequirements, unsatisfiedRequirements, bootstrapRequirements);
 
+            if (nodesMatchingRequirements.contains(transportService.getLocalNode()) == false) {
+                logger.info("skipping cluster bootstrapping as local node does not match bootstrap requirements: {}",
+                    bootstrapRequirements);
+                bootstrappingPermitted.set(false);
+                return;
+            }
+
             if (nodesMatchingRequirements.size() * 2 > bootstrapRequirements.size()) {
                 startBootstrap(nodesMatchingRequirements, unsatisfiedRequirements);
             }

server/src/main/java/org/elasticsearch/cluster/coordination/Coordinator.java

@@ -348,6 +348,12 @@ public class Coordinator extends AbstractLifecycleComponent implements Discovery
             // The preVoteCollector is only active while we are candidate, but it does not call this method with synchronisation, so we have
             // to check our mode again here.
             if (mode == Mode.CANDIDATE) {
+                if (electionQuorumContainsLocalNode(getLastAcceptedState()) == false) {
+                    logger.trace("skip election as local node is not part of election quorum: {}",
+                        getLastAcceptedState().coordinationMetaData());
+                    return;
+                }
+
                 final StartJoinRequest startJoinRequest
                     = new StartJoinRequest(getLocalNode(), Math.max(getCurrentTerm(), maxTermSeen) + 1);
                 logger.debug("starting election with {}", startJoinRequest);
@@ -360,6 +366,13 @@ public class Coordinator extends AbstractLifecycleComponent implements Discovery
         }
     }
 
+    private static boolean electionQuorumContainsLocalNode(ClusterState lastAcceptedState) {
+        final String localNodeId = lastAcceptedState.nodes().getLocalNodeId();
+        assert localNodeId != null;
+        return lastAcceptedState.getLastCommittedConfiguration().getNodeIds().contains(localNodeId)
+            || lastAcceptedState.getLastAcceptedConfiguration().getNodeIds().contains(localNodeId);
+    }
+
     private Optional<Join> ensureTermAtLeast(DiscoveryNode sourceNode, long targetTerm) {
         assert Thread.holdsLock(mutex) : "Coordinator mutex not held";
         if (getCurrentTerm() < targetTerm) {
@@ -709,10 +722,24 @@ public class Coordinator extends AbstractLifecycleComponent implements Discovery
                 return false;
             }
 
+            if (getLocalNode().isMasterNode() == false) {
+                logger.debug("skip setting initial configuration as local node is not a master-eligible node");
+                throw new CoordinationStateRejectedException(
+                    "this node is not master-eligible, but cluster bootstrapping can only happen on a master-eligible node");
+            }
+
+            if (votingConfiguration.getNodeIds().contains(getLocalNode().getId()) == false) {
+                logger.debug("skip setting initial configuration as local node is not part of initial configuration");
+                throw new CoordinationStateRejectedException("local node is not part of initial configuration");
+            }
+
             final List<DiscoveryNode> knownNodes = new ArrayList<>();
             knownNodes.add(getLocalNode());
             peerFinder.getFoundPeers().forEach(knownNodes::add);
+
             if (votingConfiguration.hasQuorum(knownNodes.stream().map(DiscoveryNode::getId).collect(Collectors.toList())) == false) {
+                logger.debug("skip setting initial configuration as not enough nodes discovered to form a quorum in the " +
+                    "initial configuration [knownNodes={}, {}]", knownNodes, votingConfiguration);
                 throw new CoordinationStateRejectedException("not enough nodes discovered to form a quorum in the initial configuration " +
                     "[knownNodes=" + knownNodes + ", " + votingConfiguration + "]");
             }
@@ -729,6 +756,8 @@ public class Coordinator extends AbstractLifecycleComponent implements Discovery
             metaDataBuilder.coordinationMetaData(coordinationMetaData);
 
             coordinationState.get().setInitialState(ClusterState.builder(currentState).metaData(metaDataBuilder).build());
+            assert electionQuorumContainsLocalNode(getLastAcceptedState()) :
+                "initial state does not have local node in its election quorum: " + getLastAcceptedState().coordinationMetaData();
             preVoteCollector.update(getPreVoteResponse(), null); // pick up the change to last-accepted version
             startElectionScheduler();
             return true;
@@ -1022,12 +1051,20 @@ public class Coordinator extends AbstractLifecycleComponent implements Discovery
             public void run() {
                 synchronized (mutex) {
                     if (mode == Mode.CANDIDATE) {
+                        final ClusterState lastAcceptedState = coordinationState.get().getLastAcceptedState();
+
+                        if (electionQuorumContainsLocalNode(lastAcceptedState) == false) {
+                            logger.trace("skip prevoting as local node is not part of election quorum: {}",
+                                lastAcceptedState.coordinationMetaData());
+                            return;
+                        }
+
                         if (prevotingRound != null) {
                             prevotingRound.close();
                         }
-                        final ClusterState lastAcceptedState = coordinationState.get().getLastAcceptedState();
                         final List<DiscoveryNode> discoveredNodes
                             = getDiscoveredNodes().stream().filter(n -> isZen1Node(n) == false).collect(Collectors.toList());
+
                         prevotingRound = preVoteCollector.start(lastAcceptedState, discoveredNodes);
                     }
                 }

server/src/test/java/org/elasticsearch/cluster/coordination/ClusterBootstrapServiceTests.java

@@ -328,6 +328,18 @@ public class ClusterBootstrapServiceTests extends ESTestCase {
         deterministicTaskQueue.runAllTasks();
     }
 
+    public void testDoesNotBootstrapsIfLocalNodeNotInInitialMasterNodes() {
+        ClusterBootstrapService clusterBootstrapService = new ClusterBootstrapService(Settings.builder().putList(
+            INITIAL_MASTER_NODES_SETTING.getKey(), otherNode1.getName(), otherNode2.getName()).build(),
+            transportService, () ->
+            Stream.of(localNode, otherNode1, otherNode2).collect(Collectors.toList()), () -> false, vc -> {
+            throw new AssertionError("should not be called");
+        });
+        transportService.start();
+        clusterBootstrapService.onFoundPeersUpdated();
+        deterministicTaskQueue.runAllTasks();
+    }
+
     public void testDoesNotBootstrapsIfNotConfigured() {
         ClusterBootstrapService clusterBootstrapService = new ClusterBootstrapService(
             Settings.builder().putList(INITIAL_MASTER_NODES_SETTING.getKey()).build(), transportService,

server/src/test/java/org/elasticsearch/cluster/coordination/CoordinatorTests.java

@@ -19,7 +19,6 @@
 package org.elasticsearch.cluster.coordination;
 
 import com.carrotsearch.randomizedtesting.RandomizedContext;
-
 import org.apache.logging.log4j.CloseableThreadContext;
 import org.apache.logging.log4j.LogManager;
 import org.apache.logging.log4j.Logger;
@@ -53,6 +52,7 @@ import org.elasticsearch.common.settings.Settings;
 import org.elasticsearch.common.settings.Settings.Builder;
 import org.elasticsearch.common.transport.TransportAddress;
 import org.elasticsearch.common.unit.TimeValue;
+import org.elasticsearch.common.util.set.Sets;
 import org.elasticsearch.discovery.zen.PublishClusterStateStats;
 import org.elasticsearch.discovery.zen.UnicastHostsProvider.HostsResolver;
 import org.elasticsearch.env.NodeEnvironment;
@@ -93,10 +93,10 @@ import static org.elasticsearch.cluster.coordination.ClusterBootstrapService.BOO
 import static org.elasticsearch.cluster.coordination.CoordinationStateTests.clusterState;
 import static org.elasticsearch.cluster.coordination.CoordinationStateTests.setValue;
 import static org.elasticsearch.cluster.coordination.CoordinationStateTests.value;
-import static org.elasticsearch.cluster.coordination.Coordinator.PUBLISH_TIMEOUT_SETTING;
 import static org.elasticsearch.cluster.coordination.Coordinator.Mode.CANDIDATE;
 import static org.elasticsearch.cluster.coordination.Coordinator.Mode.FOLLOWER;
 import static org.elasticsearch.cluster.coordination.Coordinator.Mode.LEADER;
+import static org.elasticsearch.cluster.coordination.Coordinator.PUBLISH_TIMEOUT_SETTING;
 import static org.elasticsearch.cluster.coordination.CoordinatorTests.Cluster.DEFAULT_DELAY_VARIABILITY;
 import static org.elasticsearch.cluster.coordination.ElectionSchedulerFactory.ELECTION_BACK_OFF_TIME_SETTING;
 import static org.elasticsearch.cluster.coordination.ElectionSchedulerFactory.ELECTION_DURATION_SETTING;
@@ -117,7 +117,6 @@ import static org.elasticsearch.node.Node.NODE_NAME_SETTING;
 import static org.elasticsearch.transport.TransportService.NOOP_TRANSPORT_INTERCEPTOR;
 import static org.hamcrest.Matchers.containsString;
 import static org.hamcrest.Matchers.empty;
-import static org.hamcrest.Matchers.endsWith;
 import static org.hamcrest.Matchers.equalTo;
 import static org.hamcrest.Matchers.greaterThan;
 import static org.hamcrest.Matchers.greaterThanOrEqualTo;
@@ -745,7 +744,7 @@ public class CoordinatorTests extends ESTestCase {
             assertThat(nodeId + " should have found all peers", foundPeers, hasSize(cluster.size()));
         }
 
-        final ClusterNode bootstrapNode = cluster.getAnyNode();
+        final ClusterNode bootstrapNode = cluster.getAnyBootstrappableNode();
         bootstrapNode.applyInitialConfiguration();
         assertTrue(bootstrapNode.getId() + " has been bootstrapped", bootstrapNode.coordinator.isInitialConfigurationSet());
 
@@ -775,13 +774,13 @@ public class CoordinatorTests extends ESTestCase {
     public void testCannotSetInitialConfigurationWithoutQuorum() {
         final Cluster cluster = new Cluster(randomIntBetween(1, 5));
         final Coordinator coordinator = cluster.getAnyNode().coordinator;
-        final VotingConfiguration unknownNodeConfiguration = new VotingConfiguration(Collections.singleton("unknown-node"));
+        final VotingConfiguration unknownNodeConfiguration = new VotingConfiguration(
+            Sets.newHashSet(coordinator.getLocalNode().getId(), "unknown-node"));
         final String exceptionMessage = expectThrows(CoordinationStateRejectedException.class,
             () -> coordinator.setInitialConfiguration(unknownNodeConfiguration)).getMessage();
         assertThat(exceptionMessage,
             startsWith("not enough nodes discovered to form a quorum in the initial configuration [knownNodes=["));
-        assertThat(exceptionMessage,
+        assertThat(exceptionMessage, containsString("unknown-node"));
-            endsWith("], VotingConfiguration{unknown-node}]"));
         assertThat(exceptionMessage, containsString(coordinator.getLocalNode().toString()));
 
         // This is VERY BAD: setting a _different_ initial configuration. Yet it works if the first attempt will never be a quorum.
@@ -789,6 +788,16 @@ public class CoordinatorTests extends ESTestCase {
         cluster.stabilise();
     }
 
+    public void testCannotSetInitialConfigurationWithoutLocalNode() {
+        final Cluster cluster = new Cluster(randomIntBetween(1, 5));
+        final Coordinator coordinator = cluster.getAnyNode().coordinator;
+        final VotingConfiguration unknownNodeConfiguration = new VotingConfiguration(Sets.newHashSet("unknown-node"));
+        final String exceptionMessage = expectThrows(CoordinationStateRejectedException.class,
+            () -> coordinator.setInitialConfiguration(unknownNodeConfiguration)).getMessage();
+        assertThat(exceptionMessage,
+            equalTo("local node is not part of initial configuration"));
+    }
+
     public void testDiffBasedPublishing() {
         final Cluster cluster = new Cluster(randomIntBetween(1, 5));
         cluster.runRandomly();
@@ -1331,7 +1340,7 @@ public class CoordinatorTests extends ESTestCase {
                 assertThat("setting initial configuration may fail with disconnected nodes", disconnectedNodes, empty());
                 assertThat("setting initial configuration may fail with blackholed nodes", blackholedNodes, empty());
                 runFor(defaultMillis(DISCOVERY_FIND_PEERS_INTERVAL_SETTING) * 2, "discovery prior to setting initial configuration");
-                final ClusterNode bootstrapNode = getAnyMasterEligibleNode();
+                final ClusterNode bootstrapNode = getAnyBootstrappableNode();
                 bootstrapNode.applyInitialConfiguration();
             } else {
                 logger.info("setting initial configuration not required");
@@ -1402,8 +1411,10 @@ public class CoordinatorTests extends ESTestCase {
             return clusterNodes.stream().anyMatch(cn -> cn.getLocalNode().equals(node));
         }
 
-        ClusterNode getAnyMasterEligibleNode() {
+        ClusterNode getAnyBootstrappableNode() {
-            return randomFrom(clusterNodes.stream().filter(n -> n.getLocalNode().isMasterNode()).collect(Collectors.toList()));
+            return randomFrom(clusterNodes.stream().filter(n -> n.getLocalNode().isMasterNode())
+                .filter(n -> initialConfiguration.getNodeIds().contains(n.getLocalNode().getId()))
+                .collect(Collectors.toList()));
         }
 
         ClusterNode getAnyNode() {
@@ -1737,8 +1748,14 @@ public class CoordinatorTests extends ESTestCase {
                     Stream.generate(() -> BOOTSTRAP_PLACEHOLDER_PREFIX + UUIDs.randomBase64UUID(random()))
                         .limit((Math.max(initialConfiguration.getNodeIds().size(), 2) - 1) / 2)
                         .forEach(nodeIdsWithPlaceholders::add);
-                    final VotingConfiguration configurationWithPlaceholders = new VotingConfiguration(new HashSet<>(
+                    final Set<String> nodeIds = new HashSet<>(
-                        randomSubsetOf(initialConfiguration.getNodeIds().size(), nodeIdsWithPlaceholders)));
+                        randomSubsetOf(initialConfiguration.getNodeIds().size(), nodeIdsWithPlaceholders));
+                    // initial configuration should not have a place holder for local node
+                    if (initialConfiguration.getNodeIds().contains(localNode.getId()) && nodeIds.contains(localNode.getId()) == false) {
+                        nodeIds.remove(nodeIds.iterator().next());
+                        nodeIds.add(localNode.getId());
+                    }
+                    final VotingConfiguration configurationWithPlaceholders = new VotingConfiguration(nodeIds);
                     try {
                         coordinator.setInitialConfiguration(configurationWithPlaceholders);
                         logger.info("successfully set initial configuration to {}", configurationWithPlaceholders);