com.splout.db.hazelcast.DistributedRegistry Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of splout-server Show documentation
Show all versions of splout-server Show documentation
Splout SQL is a read only, horizontally scalable and
partitioned SQL database that plays well with Hadoop.
The newest version!
package com.splout.db.hazelcast;
/*
* #%L
* Splout SQL Server
* %%
* Copyright (C) 2012 Datasalt Systems S.L.
* %%
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU Affero General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see
* Main difficulty with the registry is the case of network partitions, or split-brain. In this case, all
* registry replicas of one member could be lost. It is an improbable case, but possible. For this case,
* we have a thread that periodically check that the node is registered so assuring eventual consistency.
* Another design principle of this registry is the idea of trying to minimize network
* communication between nodes, as that could affect the system scalability.
*/
public class DistributedRegistry {
private final static Log log = LogFactory.getLog(DistributedRegistry.class);
private final String registryName;
private Object nodeInfo;
private final HazelcastInstance hzInstance;
private final int minutesToCheckRegister;
private final int oldestMembersLeading;
private final AtomicBoolean amIRegistered = new AtomicBoolean(false);
private final AtomicBoolean disableChecking = new AtomicBoolean(false);
private final ScheduledExecutorService scheduler = Executors.newScheduledThreadPool(1);
private final AtomicReference> checker = new AtomicReference>();
private final Random random = new Random();
public class MyListener implements MembershipListener {
@Override
public void memberAdded(MembershipEvent membershipEvent) {
// Nothing to do. I should be registered. Just in case...
scheduleCheck();
}
@Override
public void memberRemoved(MembershipEvent membershipEvent) {
synchronized(DistributedRegistry.this) {
/*
* When a node leaves, somebody in the cluster must remove its info from the distributed map. We
* restrict this removal to some of the oldest members, in order to reduce coordination traffic
* while keeping replication
*/
if(HazelcastUtils.isOneOfOldestMembers(hzInstance, hzInstance.getCluster().getLocalMember(),
oldestMembersLeading)) {
String member = getHZAddress(membershipEvent.getMember());
log.info("Member " + member + " leaves. Unregistering it from registry [" + registryName + "]");
ConcurrentMap members = hzInstance.getMap(registryName);
members.remove(member);
}
}
// Just in case...
scheduleCheck();
}
/**
* In the case of a network partition where some data is lost (unprovable, as replication should
* mitigate that), we could have some members that believe they are registered but they are not. We
* set a periodical check to test that. We schedule a test each time a new member arrives or leaves
* to the cluster. But in order to avoid to much checking, only one check can be performed in a
* period of time. Also, to reduce the herd behavior, we schedule the check randomly in this period.
*/
private void scheduleCheck() {
if(disableChecking.get()) {
return;
}
ScheduledFuture checkerFuture = checker.get();
if(checkerFuture == null || checkerFuture.isDone()) {
int seconds = random.nextInt(Math.max(1, minutesToCheckRegister * 60));
checker.set(scheduler.schedule(new Runnable() {
@Override
public void run() {
synchronized(DistributedRegistry.this) {
if(amIRegistered.get()) {
String member = localMember();
log.info("Checking if registered [" + member + "] ...");
ConcurrentMap members = hzInstance.getMap(registryName);
if(members.get(member) == null) {
log.warn("Detected wrongly unregistered ["
+ member
+ "]. Could be due a network partition problem or due to a software bug. Reregistering.");
register();
}
}
}
}
}, seconds, TimeUnit.SECONDS));
}
}
@Override
public void memberAttributeChanged(MemberAttributeEvent memberAttributeEvent) {
// TODO Since HZ 3.2
}
}
/**
* When a split-brain happens, and two clusters will merge, the member of the smallest cluster are
* restarted. When we detect that, we reregister the clusters in order to assure that their info is
* present in the registry distributed map.
* This alone does not assure complete coherence in the case of a network partition merge, as members
* of the bigger cluster could have an incomplete registry of themselves as well.
*/
public class RestartListener implements LifecycleListener {
@Override
public void stateChanged(LifecycleEvent event) {
if(event.getState() == LifecycleState.MERGED) {
synchronized(DistributedRegistry.this) {
if(amIRegistered.get()) {
log.info("Hazelcast RESTARTED event received. Reregistering myself to asure I'm properly registered");
register();
}
}
}
}
}
public DistributedRegistry(String registryName, Object nodeInfo, HazelcastInstance hzInstance,
int minutesToCheckRegister, int oldestMembersLeading) {
this.registryName = registryName;
this.nodeInfo = nodeInfo;
this.hzInstance = hzInstance;
hzInstance.getCluster().addMembershipListener(new MyListener());
hzInstance.getLifecycleService().addLifecycleListener(new RestartListener());
this.minutesToCheckRegister = minutesToCheckRegister;
this.oldestMembersLeading = oldestMembersLeading;
}
public synchronized void register() {
String myself = localMember();
log.info("Registering myself [" + myself + "] on registry [" + registryName + "]");
ConcurrentMap members = hzInstance.getMap(registryName);
members.put(myself, nodeInfo);
amIRegistered.set(true);
}
public synchronized void unregister() {
String myself = localMember();
log.info("Unregistering myself [" + myself + " -> " + nodeInfo + "] on registry [" + registryName
+ "]");
ConcurrentMap members = hzInstance.getMap(registryName);
members.remove(myself);
amIRegistered.set(false);
}
public synchronized void changeInfo(Object nodeInfo) {
// Changing memory information. Needed for future reregistration
this.nodeInfo = nodeInfo;
String myself = localMember();
log.info("Changing my info [" + myself + "] on registry [" + registryName + "]");
ConcurrentMap members = hzInstance.getMap(registryName);
members.put(myself, nodeInfo);
amIRegistered.set(true);
}
/**
* Enables or disable preventive registration checking.
*/
protected void disableChecking(boolean disable) {
disableChecking.set(disable);
}
private String localMember() {
return getHZAddress(hzInstance.getCluster().getLocalMember());
}
public void dumpRegistry() {
ConcurrentMap members = hzInstance.getMap(registryName);
System.out.println("Registry [" + registryName + "] {");
for(Entry entry : members.entrySet()) {
System.out.println("\t" + entry.getKey() + " -> " + entry.getValue());
}
System.out.println("}");
}
}