public void restart(int siteId) throws Exception {
EnvironmentConfig ec = makeBasicConfig();
int p = config.getMyPort(siteId);
ReplicationManagerSiteConfig dbsite = new ReplicationManagerSiteConfig("localhost", p);
dbsite.setLocalSite(true);
dbsite.setLegacy(true);
ec.addReplicationManagerSite(dbsite);
p = config.getOtherPort(siteId);
dbsite = new ReplicationManagerSiteConfig("localhost", p);
dbsite.setLegacy(true);
ec.addReplicationManagerSite(dbsite);
MyEventHandler monitor = new MyEventHandler();
ec.setEventHandler(monitor);
File clientDir = new File(config.getBaseDir(), "dir" + siteId);
assertTrue(clientDir.exists());
Environment client = new Environment(clientDir, ec);
client.setReplicationConfig(ReplicationConfig.STRICT_2SITE, false);
envs[siteId] = client;
monitors[siteId] = monitor;
// we want to make sure we don't retry from here after the
// initial failure, because we want to make the old master
// connect to us.
client.setReplicationTimeout(ReplicationTimeoutType.CONNECTION_RETRY, Integer.MAX_VALUE);
client.replicationManagerStart(3, ReplicationManagerStartPolicy.REP_CLIENT);
}
public void joinExistingClient(int site, boolean useHB) throws Exception {
EnvironmentConfig ec = makeBasicConfig();
int p = config.getMyPort(site);
ReplicationManagerSiteConfig dbsite = new ReplicationManagerSiteConfig("localhost", p);
dbsite.setLocalSite(true);
dbsite.setLegacy(true);
ec.addReplicationManagerSite(dbsite);
p = config.getOtherPort(site);
dbsite = new ReplicationManagerSiteConfig("localhost", p);
dbsite.setLegacy(true);
ec.addReplicationManagerSite(dbsite);
MyEventHandler monitor = new MyEventHandler();
monitors[site] = monitor;
ec.setEventHandler(monitor);
File clientDir = new File(config.getBaseDir(), "dir" + site);
assertTrue(clientDir.exists());
Environment client = new Environment(clientDir, ec);
client.setReplicationConfig(ReplicationConfig.STRICT_2SITE, false);
if (useHB) {
client.setReplicationTimeout(ReplicationTimeoutType.HEARTBEAT_SEND, 3000000);
client.setReplicationTimeout(ReplicationTimeoutType.HEARTBEAT_MONITOR, 6000000);
}
envs[site] = client;
client.setReplicationTimeout(ReplicationTimeoutType.CONNECTION_RETRY, 1000000); // be impatient
client.replicationManagerStart(3, ReplicationManagerStartPolicy.REP_CLIENT);
monitor.await();
assertTrue(client.getReplicationStats(StatsConfig.DEFAULT).getStartupComplete());
}
public void run() {
String homedirName = baseDirName + threadNumber;
TestUtils.removeDir(homedirName);
try {
homedir = new File(homedirName);
homedir.mkdir();
} catch (Exception e) {
TestUtils.DEBUGOUT(
2, "Warning: initialization had a problem creating a clean directory.\n" + e);
}
try {
homedir = new File(homedirName);
} catch (NullPointerException npe) {
// can't really happen :)
}
TestUtils.DEBUGOUT(1, "Creating worker: " + threadNumber);
envConfig = new EnvironmentConfig();
envConfig.setErrorStream(TestUtils.getErrorStream());
envConfig.setErrorPrefix("RepmgrElectionTest test(" + threadNumber + ")");
envConfig.setAllowCreate(true);
envConfig.setRunRecovery(true);
envConfig.setThreaded(true);
envConfig.setInitializeLocking(true);
envConfig.setInitializeLogging(true);
envConfig.setInitializeCache(true);
envConfig.setTransactional(true);
envConfig.setTxnNoSync(true);
envConfig.setInitializeReplication(true);
envConfig.setVerboseReplication(false);
ReplicationManagerSiteConfig localConfig =
new ReplicationManagerSiteConfig(address, basePort + threadNumber);
localConfig.setLocalSite(true);
envConfig.addReplicationManagerSite(localConfig);
envConfig.setReplicationPriority(priorities[threadNumber]);
envConfig.setEventHandler(this);
envConfig.setReplicationManagerAckPolicy(ReplicationManagerAckPolicy.ALL);
if (masterThreadIndex >= 0) {
// If we already have the master, then set it as the bootstrap helper,
// otherwise, set local site as new master.
ReplicationManagerSiteConfig remoteConfig =
new ReplicationManagerSiteConfig(address, basePort + masterThreadIndex);
remoteConfig.setBootstrapHelper(true);
envConfig.addReplicationManagerSite(remoteConfig);
}
try {
dbenv = new Environment(homedir, envConfig);
} catch (FileNotFoundException e) {
fail("Unexpected FNFE in standard environment creation." + e);
} catch (DatabaseException dbe) {
fail("Unexpected database exception came from environment create." + dbe);
}
try {
// If we do not have master, then set local site as new master.
if (masterThreadIndex == -1)
dbenv.replicationManagerStart(NUM_WORKER_THREADS, ReplicationManagerStartPolicy.REP_MASTER);
else
dbenv.replicationManagerStart(NUM_WORKER_THREADS, ReplicationManagerStartPolicy.REP_CLIENT);
} catch (DatabaseException dbe) {
fail("Unexpected database exception came from replicationManagerStart." + dbe);
}
TestUtils.DEBUGOUT(1, "Started replication site: " + threadNumber);
lastSiteStarted = true;
try {
java.lang.Thread.sleep(1000 * (1 + threadNumber));
} catch (InterruptedException ie) {
}
if (masterThreadIndex != -1) {
// Wait for "Start-up done" for each client, then add next client.
ReplicationStats rs = null;
int i = 0;
do {
try {
java.lang.Thread.sleep(2000);
} catch (InterruptedException e) {
}
try {
rs = dbenv.getReplicationStats(StatsConfig.DEFAULT);
} catch (DatabaseException dbe) {
dbe.printStackTrace();
fail("Unexpected database exception came from getReplicationStats." + dbe);
}
} while (!rs.getStartupComplete() && i++ < maxLoopWait);
assertTrue(rs.getStartupComplete());
}
}
@Test
public void testDraining() throws Exception {
EnvironmentConfig masterConfig = makeBasicConfig();
masterConfig.setReplicationLimit(100000000);
ReplicationManagerSiteConfig site = new ReplicationManagerSiteConfig("localhost", masterPort);
site.setLocalSite(true);
site.setLegacy(true);
masterConfig.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", clientPort);
site.setLegacy(true);
masterConfig.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", client2Port);
site.setLegacy(true);
masterConfig.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", client3Port);
site.setLegacy(true);
masterConfig.addReplicationManagerSite(site);
Environment master = new Environment(mkdir("master"), masterConfig);
setTimeouts(master);
// Prevent connection retries, so that all connections
// originate from clients
master.setReplicationTimeout(ReplicationTimeoutType.CONNECTION_RETRY, Integer.MAX_VALUE);
master.replicationManagerStart(2, ReplicationManagerStartPolicy.REP_MASTER);
DatabaseConfig dc = new DatabaseConfig();
dc.setTransactional(true);
dc.setAllowCreate(true);
dc.setType(DatabaseType.BTREE);
dc.setPageSize(4096);
Database db = master.openDatabase(null, "test.db", null, dc);
DatabaseEntry key = new DatabaseEntry();
DatabaseEntry value = new DatabaseEntry();
value.setData(data);
for (int i = 0; ((BtreeStats) db.getStats(null, null)).getPageCount() < 500; i++) {
String k = "The record number is: " + i;
key.setData(k.getBytes());
db.put(null, key, value);
}
// tell fiddler to stop reading once it sees a PAGE message
Socket s = new Socket("localhost", mgrPort);
OutputStreamWriter w = new OutputStreamWriter(s.getOutputStream());
String path1 = "{" + masterPort + "," + clientPort + "}"; // looks like {6000,6001}
w.write("{init," + path1 + ",page_clog}\r\n");
w.flush();
BufferedReader br = new BufferedReader(new InputStreamReader(s.getInputStream()));
br.readLine();
assertEquals("ok", br.readLine());
// create client
//
EnvironmentConfig ec = makeBasicConfig();
site = new ReplicationManagerSiteConfig("localhost", clientPort);
site.setLocalSite(true);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", masterPort);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", client2Port);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", client3Port);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
Environment client = new Environment(mkdir("client"), ec);
setTimeouts(client);
client.replicationManagerStart(1, ReplicationManagerStartPolicy.REP_CLIENT);
// wait til it gets stuck
Thread.sleep(5000); // FIXME
// Do the same for another client, because the master has 2
// msg processing threads. (It's no longer possible to
// configure just 1.)
String path2 = "{" + masterPort + "," + client2Port + "}";
w.write("{init," + path2 + ",page_clog}\r\n");
w.flush();
br = new BufferedReader(new InputStreamReader(s.getInputStream()));
br.readLine();
assertEquals("ok", br.readLine());
ec = makeBasicConfig();
site = new ReplicationManagerSiteConfig("localhost", client2Port);
site.setLocalSite(true);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", masterPort);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", clientPort);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", client3Port);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
Environment client2 = new Environment(mkdir("client2"), ec);
setTimeouts(client2);
client2.replicationManagerStart(1, ReplicationManagerStartPolicy.REP_CLIENT);
// wait til it gets stuck
Thread.sleep(5000);
// With the connection stuck, the master cannot write out log
// records for new "live" transactions. Knowing we didn't
// write the record, we should not bother waiting for an ack
// that cannot possibly arrive; so we should simply return
// quickly. The duration should be very quick, but anything
// less than the ack timeout indicates correct behavior (in
// case this test runs on a slow, overloaded system).
//
long startTime = System.currentTimeMillis();
key.setData("one extra record".getBytes());
db.put(null, key, value);
long duration = System.currentTimeMillis() - startTime;
assertTrue("txn duration: " + duration, duration < 29000);
System.out.println("txn duration: " + duration);
db.close();
// Tell fiddler to close the connections. That should trigger
// us to abandon the timeout. Then create another client and
// see that it can complete its internal init quickly. Since
// we have limited threads at the master, this demonstrates
// that they were abandoned.
//
path1 = "{" + clientPort + "," + masterPort + "}"; // looks like {6001,6000}
w.write("{" + path1 + ",shutdown}\r\n");
w.flush();
assertEquals("ok", br.readLine());
path2 = "{" + client2Port + "," + masterPort + "}"; // looks like {6001,6000}
w.write("{" + path2 + ",shutdown}\r\n");
w.flush();
assertEquals("ok", br.readLine());
ec = makeBasicConfig();
site = new ReplicationManagerSiteConfig("localhost", client3Port);
site.setLocalSite(true);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", masterPort);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", clientPort);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
site = new ReplicationManagerSiteConfig("localhost", client2Port);
site.setLegacy(true);
ec.addReplicationManagerSite(site);
EventHandler clientMonitor = new EventHandler();
ec.setEventHandler(clientMonitor);
Environment client3 = new Environment(mkdir("client3"), ec);
setTimeouts(client3);
startTime = System.currentTimeMillis();
client3.replicationManagerStart(2, ReplicationManagerStartPolicy.REP_CLIENT);
clientMonitor.await();
duration = System.currentTimeMillis() - startTime;
assertTrue("sync duration: " + duration, duration < 20000); // 20 seconds should be plenty
client3.close();
master.close();
w.write("shutdown\r\n");
w.flush();
assertEquals("ok", br.readLine());
s.close();
}