use of org.apache.zookeeper.server.quorum.Leader in project zookeeper by apache.
the class MonitorCommand method commandRun.
@Override
public void commandRun() {
if (!isZKServerRunning()) {
pw.println(ZK_NOT_SERVING);
return;
}
ZKDatabase zkdb = zkServer.getZKDatabase();
ServerStats stats = zkServer.serverStats();
print("version", Version.getFullVersion());
print("avg_latency", stats.getAvgLatency());
print("max_latency", stats.getMaxLatency());
print("min_latency", stats.getMinLatency());
print("packets_received", stats.getPacketsReceived());
print("packets_sent", stats.getPacketsSent());
print("num_alive_connections", stats.getNumAliveClientConnections());
print("outstanding_requests", stats.getOutstandingRequests());
print("server_state", stats.getServerState());
print("znode_count", zkdb.getNodeCount());
print("watch_count", zkdb.getDataTree().getWatchCount());
print("ephemerals_count", zkdb.getDataTree().getEphemeralsCount());
print("approximate_data_size", zkdb.getDataTree().approximateDataSize());
OSMXBean osMbean = new OSMXBean();
if (osMbean != null && osMbean.getUnix() == true) {
print("open_file_descriptor_count", osMbean.getOpenFileDescriptorCount());
print("max_file_descriptor_count", osMbean.getMaxFileDescriptorCount());
}
if (stats.getServerState().equals("leader")) {
Leader leader = ((LeaderZooKeeperServer) zkServer).getLeader();
print("followers", leader.getLearners().size());
print("synced_followers", leader.getForwardingFollowers().size());
print("pending_syncs", leader.getNumPendingSyncs());
print("last_proposal_size", leader.getProposalStats().getLastProposalSize());
print("max_proposal_size", leader.getProposalStats().getMaxProposalSize());
print("min_proposal_size", leader.getProposalStats().getMinProposalSize());
}
}
use of org.apache.zookeeper.server.quorum.Leader in project zookeeper by apache.
the class QuorumTest method testLeaderShutdown.
@Test
public void testLeaderShutdown() throws IOException, InterruptedException, KeeperException {
ZooKeeper zk = new DisconnectableZooKeeper(qb.hostPort, ClientBase.CONNECTION_TIMEOUT, DummyWatcher.INSTANCE);
zk.create("/blah", new byte[0], ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
zk.create("/blah/blah", new byte[0], ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
Leader leader = qb.s1.leader;
if (leader == null) {
leader = qb.s2.leader;
}
if (leader == null) {
leader = qb.s3.leader;
}
if (leader == null) {
leader = qb.s4.leader;
}
if (leader == null) {
leader = qb.s5.leader;
}
assertNotNull(leader);
for (int i = 0; i < 5000; i++) {
zk.setData("/blah/blah", new byte[0], -1, (rc, path, ctx, stat) -> {
counter++;
if (rc != 0) {
errors++;
}
}, null);
}
for (LearnerHandler f : leader.getForwardingFollowers()) {
f.getSocket().shutdownInput();
}
for (int i = 0; i < 5000; i++) {
zk.setData("/blah/blah", new byte[0], -1, (rc, path, ctx, stat) -> {
counter++;
if (rc != 0) {
errors++;
}
}, null);
}
// check if all the followers are alive
assertTrue(qb.s1.isAlive());
assertTrue(qb.s2.isAlive());
assertTrue(qb.s3.isAlive());
assertTrue(qb.s4.isAlive());
assertTrue(qb.s5.isAlive());
zk.close();
}
use of org.apache.zookeeper.server.quorum.Leader in project zookeeper by apache.
the class FollowerResyncConcurrencyTest method testResyncByDiffAfterFollowerCrashes.
/**
* This test:
* Starts up 3 ZKs. The non-leader ZKs are writing to cluster
* Shut down one of the non-leader ZKs.
* Restart after sessions have expired but less than 500 txns have taken place (get a diff)
* Shut down immediately after restarting, start running separate thread with other transactions
* Restart to a diff while transactions are running in leader
*
* Before fixes for ZOOKEEPER-962, restarting off of diff could get an inconsistent view of data missing transactions that
* completed during diff syncing. Follower would also be considered "restarted" before all forwarded transactions
* were completely processed, so restarting would cause a snap file with a too-high zxid to be written, and transactions
* would be missed
*
* This test should pretty reliably catch the failure of restarting the server before all diff messages have been processed,
* however, due to the transient nature of the system it may not catch failures due to concurrent processing of transactions
* during the leader's diff forwarding.
*
* @throws IOException
* @throws InterruptedException
* @throws KeeperException
* @throws Throwable
*/
@Test
public void testResyncByDiffAfterFollowerCrashes() throws IOException, InterruptedException, KeeperException, Throwable {
final Semaphore sem = new Semaphore(0);
QuorumUtil qu = new QuorumUtil(1);
qu.startAll();
CountdownWatcher watcher1 = new CountdownWatcher();
CountdownWatcher watcher2 = new CountdownWatcher();
CountdownWatcher watcher3 = new CountdownWatcher();
int index = 1;
while (qu.getPeer(index).peer.leader == null) {
index++;
}
Leader leader = qu.getPeer(index).peer.leader;
assertNotNull(leader);
/* Reusing the index variable to select a follower to connect to */
index = (index == 1) ? 2 : 1;
LOG.info("Connecting to follower: {}", index);
final ZooKeeper zk1 = createClient(qu.getPeer(index).peer.getClientPort(), watcher1);
LOG.info("zk1 has session id 0x{}", Long.toHexString(zk1.getSessionId()));
final ZooKeeper zk2 = createClient(qu.getPeer(index).peer.getClientPort(), watcher2);
LOG.info("zk2 has session id 0x{}", Long.toHexString(zk2.getSessionId()));
final ZooKeeper zk3 = createClient(qu.getPeer(3).peer.getClientPort(), watcher3);
LOG.info("zk3 has session id 0x{}", Long.toHexString(zk3.getSessionId()));
zk1.create("/first", new byte[0], ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
zk2.create("/mybar", null, ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL_SEQUENTIAL);
final AtomicBoolean runNow = new AtomicBoolean(false);
Thread mytestfooThread = new Thread(new Runnable() {
@Override
public void run() {
int inSyncCounter = 0;
while (inSyncCounter < 400) {
if (runNow.get()) {
zk3.create("/mytestfoo", null, ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL_SEQUENTIAL, (rc, path, ctx, name) -> {
pending.decrementAndGet();
counter.incrementAndGet();
if (rc != 0) {
errors.incrementAndGet();
}
if (counter.get() > 7300) {
sem.release();
}
}, null);
pending.incrementAndGet();
try {
Thread.sleep(10);
} catch (Exception e) {
}
inSyncCounter++;
} else {
Thread.yield();
}
}
}
});
mytestfooThread.start();
for (int i = 0; i < 5000; i++) {
zk2.create("/mybar", null, ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL_SEQUENTIAL, (rc, path, ctx, name) -> {
pending.decrementAndGet();
counter.incrementAndGet();
if (rc != 0) {
errors.incrementAndGet();
}
if (counter.get() > 7300) {
sem.release();
}
}, null);
pending.incrementAndGet();
if (i == 1000) {
qu.shutdown(index);
Thread.sleep(1100);
LOG.info("Shutting down s1");
}
if (i == 1100 || i == 1150 || i == 1200) {
Thread.sleep(1000);
}
if (i == 1200) {
qu.startThenShutdown(index);
runNow.set(true);
qu.restart(index);
LOG.info("Setting up server: {}", index);
}
if (i >= 1000 && i % 2 == 0) {
zk3.create("/newbaz", null, ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL_SEQUENTIAL, (rc, path, ctx, name) -> {
pending.decrementAndGet();
counter.incrementAndGet();
if (rc != 0) {
errors.incrementAndGet();
}
if (counter.get() > 7300) {
sem.release();
}
}, null);
pending.incrementAndGet();
}
if (i == 1050 || i == 1100 || i == 1150) {
Thread.sleep(1000);
}
}
// Wait until all updates return
if (!sem.tryAcquire(ClientBase.CONNECTION_TIMEOUT, TimeUnit.MILLISECONDS)) {
LOG.warn("Did not aquire semaphore fast enough");
}
mytestfooThread.join(ClientBase.CONNECTION_TIMEOUT);
if (mytestfooThread.isAlive()) {
LOG.error("mytestfooThread is still alive");
}
assertTrue(waitForPendingRequests(60));
assertTrue(waitForSync(qu, index, 10));
// Verify that server is following and has the same epoch as the leader
verifyState(qu, index, leader);
zk1.close();
zk2.close();
zk3.close();
qu.shutdownAll();
}
use of org.apache.zookeeper.server.quorum.Leader in project zookeeper by apache.
the class QuorumTest method testNoLogBeforeLeaderEstablishment.
/**
* Tests if closeSession can be logged before a leader gets established, which
* could lead to a locked-out follower (see ZOOKEEPER-790).
*
* The test works as follows. It has a client connecting to a follower f and
* sending batches of 1,000 updates. The goal is that f has a zxid higher than
* all other servers in the initial leader election. This way we can crash and
* recover the follower so that the follower believes it is the leader once it
* recovers (LE optimization: once a server receives a message from all other
* servers, it picks a leader.
*
* It also makes the session timeout very short so that we force the false
* leader to close the session and write it to the log in the buggy code (before
* ZOOKEEPER-790). Once f drops leadership and finds the current leader, its epoch
* is higher, and it rejects the leader. Now, if we prevent the leader from closing
* the session by only starting up (see Leader.lead()) once it obtains a quorum of
* supporters, then f will find the current leader and support it because it won't
* have a highe epoch.
*
*/
@Test
public void testNoLogBeforeLeaderEstablishment() throws Exception {
final Semaphore sem = new Semaphore(0);
qu = new QuorumUtil(2, 10);
qu.startQuorum();
int index = 1;
while (qu.getPeer(index).peer.leader == null) index++;
Leader leader = qu.getPeer(index).peer.leader;
Assert.assertNotNull(leader);
/*
* Reusing the index variable to select a follower to connect to
*/
index = (index == 1) ? 2 : 1;
ZooKeeper zk = new DisconnectableZooKeeper("127.0.0.1:" + qu.getPeer(index).peer.getClientPort(), ClientBase.CONNECTION_TIMEOUT, new Watcher() {
public void process(WatchedEvent event) {
}
});
zk.create("/blah", new byte[0], ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
for (int i = 0; i < 50000; i++) {
zk.setData("/blah", new byte[0], -1, new AsyncCallback.StatCallback() {
public void processResult(int rc, String path, Object ctx, Stat stat) {
counter++;
if (rc != 0) {
errors++;
}
if (counter == 20000) {
sem.release();
}
}
}, null);
if (i == 5000) {
qu.shutdown(index);
LOG.info("Shutting down s1");
}
if (i == 12000) {
qu.start(index);
LOG.info("Setting up server: " + index);
}
if ((i % 1000) == 0) {
Thread.sleep(500);
}
}
// Wait until all updates return
sem.tryAcquire(15, TimeUnit.SECONDS);
// Verify that server is following and has the same epoch as the leader
Assert.assertTrue("Not following", qu.getPeer(index).peer.follower != null);
long epochF = (qu.getPeer(index).peer.getActiveServer().getZxid() >> 32L);
long epochL = (leader.getEpoch() >> 32L);
Assert.assertTrue("Zxid: " + qu.getPeer(index).peer.getActiveServer().getZxid() + "Current epoch: " + epochF, epochF == epochL);
zk.close();
}
use of org.apache.zookeeper.server.quorum.Leader in project zookeeper by apache.
the class QuorumOracleMajTest method dropConnectionTest.
private void dropConnectionTest(QuorumPeer s, int leader) {
Leader.Proposal p = new Leader.Proposal();
p.addQuorumVerifier(s.getQuorumVerifier());
ArrayList<LearnerHandler> fake = new ArrayList<>();
LearnerHandler f = null;
fake.add(f);
s.getQuorumVerifier().updateNeedOracle(fake);
// still have valid followers, the oracle should not take place
assertEquals(false, s.getQuorumVerifier().getNeedOracle());
fake.remove(0);
s.getQuorumVerifier().updateNeedOracle(fake);
// lose all of followers, the oracle should take place
assertEquals(true, s.getQuorumVerifier().getNeedOracle());
// when leader is 1, we expect false.
// when leader is 2, we expect true.
assertEquals(leader != 1, p.hasAllQuorums());
}
Aggregations