Examples with ExponentialBackoffTimeoutStrategy org.apache.ignite.raft.jraft.util.ExponentialBackoffTimeoutStrategy used on opensource projects

Example 1 with ExponentialBackoffTimeoutStrategy

use of org.apache.ignite.raft.jraft.util.ExponentialBackoffTimeoutStrategy in project ignite-3 by apache.

the class ItNodeTest method testElectionTimeoutAutoAdjustWhenBlockedMessages.

private void testElectionTimeoutAutoAdjustWhenBlockedMessages(BiPredicate<Object, String> blockingPredicate) throws Exception {
    List<PeerId> peers = TestUtils.generatePeers(4);
    int maxElectionRoundsWithoutAdjusting = 3;
    cluster = new TestCluster("unittest", dataPath, peers, new LinkedHashSet<>(), ELECTION_TIMEOUT_MILLIS, opts -> opts.setElectionTimeoutStrategy(new ExponentialBackoffTimeoutStrategy(11_000, maxElectionRoundsWithoutAdjusting)), testInfo);
    for (PeerId peer : peers) {
        assertTrue(cluster.start(peer.getEndpoint()));
    }
    cluster.waitLeader();
    Node leader = cluster.getLeader();
    int initElectionTimeout = leader.getOptions().getElectionTimeoutMs();
    LOG.warn("Current leader {}, electTimeout={}", leader.getNodeId().getPeerId(), leader.getOptions().getElectionTimeoutMs());
    List<Node> followers = cluster.getFollowers();
    for (Node follower : followers) {
        NodeImpl follower0 = (NodeImpl) follower;
        assertEquals(initElectionTimeout, follower0.getOptions().getElectionTimeoutMs());
    }
    blockMessagesOnFollowers(followers, blockingPredicate);
    LOG.warn("Stop leader {}, curTerm={}", leader.getNodeId().getPeerId(), ((NodeImpl) leader).getCurrentTerm());
    assertTrue(cluster.stop(leader.getNodeId().getPeerId().getEndpoint()));
    assertNull(cluster.getLeader());
    assertTrue(waitForCondition(() -> followers.stream().allMatch(f -> f.getOptions().getElectionTimeoutMs() > initElectionTimeout), (long) maxElectionRoundsWithoutAdjusting * // need to multiply to 2 because stepDown happens after voteTimer timeout
    (initElectionTimeout + followers.get(0).getOptions().getRaftOptions().getMaxElectionDelayMs()) * 2));
    stopBlockingMessagesOnFollowers(followers);
    // elect new leader
    cluster.waitLeader();
    leader = cluster.getLeader();
    LOG.info("Elected new leader is {}, curTerm={}", leader.getLeaderId(), ((NodeImpl) leader).getCurrentTerm());
    assertTrue(waitForCondition(() -> followers.stream().allMatch(f -> f.getOptions().getElectionTimeoutMs() == initElectionTimeout), 3_000));
}

Example 2 with ExponentialBackoffTimeoutStrategy

use of org.apache.ignite.raft.jraft.util.ExponentialBackoffTimeoutStrategy in project ignite-3 by apache.

the class TestCluster method start.

public boolean start(Endpoint listenAddr, boolean emptyPeers, int snapshotIntervalSecs, boolean enableMetrics, SnapshotThrottle snapshotThrottle, RaftOptions raftOptions, int priority) throws IOException {
    this.lock.lock();
    try {
        if (this.serverMap.get(listenAddr) != null) {
            return true;
        }
        // Start node in non shared pools mode. Pools will be managed by node itself.
        NodeOptions nodeOptions = new NodeOptions();
        nodeOptions.setServerName(listenAddr.toString());
        nodeOptions.setElectionTimeoutMs(this.electionTimeoutMs);
        nodeOptions.setEnableMetrics(enableMetrics);
        nodeOptions.setSnapshotThrottle(snapshotThrottle);
        nodeOptions.setSnapshotIntervalSecs(snapshotIntervalSecs);
        nodeOptions.setServiceFactory(this.raftServiceFactory);
        if (raftOptions != null) {
            nodeOptions.setRaftOptions(raftOptions);
        }
        String serverDataPath = this.dataPath + File.separator + listenAddr.toString().replace(':', '_');
        new File(serverDataPath).mkdirs();
        nodeOptions.setLogUri(serverDataPath + File.separator + "logs");
        nodeOptions.setRaftMetaUri(serverDataPath + File.separator + "meta");
        nodeOptions.setSnapshotUri(serverDataPath + File.separator + "snapshot");
        nodeOptions.setElectionPriority(priority);
        // Align rpc options with election timeout.
        nodeOptions.setRpcConnectTimeoutMs(this.electionTimeoutMs / 3);
        nodeOptions.setRpcDefaultTimeout(this.electionTimeoutMs / 2);
        // Reduce default threads count per test node.
        nodeOptions.setRaftRpcThreadPoolSize(Utils.cpus());
        nodeOptions.setTimerPoolSize(Utils.cpus() * 2);
        nodeOptions.setRpcProcessorThreadPoolSize(Utils.cpus() * 3);
        nodeOptions.setElectionTimeoutStrategy(new ExponentialBackoffTimeoutStrategy());
        MockStateMachine fsm = new MockStateMachine(listenAddr);
        nodeOptions.setFsm(fsm);
        if (!emptyPeers)
            nodeOptions.setInitialConf(new Configuration(this.peers, this.learners));
        List<NetworkAddress> addressList = (emptyPeers ? Stream.<PeerId>empty() : peers.stream()).map(PeerId::getEndpoint).map(JRaftUtils::addressFromEndpoint).collect(toList());
        NodeManager nodeManager = new NodeManager();
        ClusterService clusterService = ClusterServiceTestUtils.clusterService(testInfo, listenAddr.getPort(), new StaticNodeFinder(addressList), new TestScaleCubeClusterServiceFactory());
        var rpcClient = new IgniteRpcClient(clusterService);
        nodeOptions.setRpcClient(rpcClient);
        ExecutorService requestExecutor = JRaftUtils.createRequestExecutor(nodeOptions);
        var rpcServer = new TestIgniteRpcServer(clusterService, nodeManager, nodeOptions, requestExecutor);
        clusterService.start();
        if (optsClo != null)
            optsClo.accept(nodeOptions);
        RaftGroupService server = new RaftGroupService(this.name, new PeerId(listenAddr, 0, priority), nodeOptions, rpcServer, nodeManager) {

            @Override
            public synchronized void shutdown() {
                // This stop order is consistent with JRaftServerImpl
                rpcServer.shutdown();
                ExecutorServiceHelper.shutdownAndAwaitTermination(requestExecutor);
                super.shutdown();
                // Network service must be stopped after a node because raft initiates timeoutnowrequest on stop for faster
                // leader election.
                clusterService.stop();
            }
        };
        this.serverMap.put(listenAddr, server);
        Node node = server.start();
        this.fsms.put(new PeerId(listenAddr, 0), fsm);
        this.nodes.add((NodeImpl) node);
        return true;
    } finally {
        this.lock.unlock();
    }
}