杂记
ZooKeeper的用途:distributed coordination;maintaining configuration information, naming, providing distributed synchronization, and providing group services.
Zookeeper的节点都是存放在内存中的,所以读写速度很快。更新日志被记录到了磁盘中,以便用于恢复数据。在更新内在中节点数之前,会先序列化到磁盘中。
为避免单点失效,zookeeper的数据是在多个server上留有备份的。不管客户端连接到的是哪个server,它看到的数据都是一致的。如果client和一个server的TCP连接失效,它会尝试连接另一个server。众多server中有一个是leader。
所有的server 都必须知道彼此的存在。
zookeeper在读写比例为10:1时性能最佳。
每个znode上data的读写都是原子操作。
读是局部性的,即client只需要从与它相连的server上读取数据即可;而client有写请求的话,与之相连的server会通知leader,然后leader会把写操作分发给所有server。所以定要比读慢很多。
在建立zookeeper连接时,给定的地址字符串可以是这样的:"192.168.1.1:3000,192.168.1.2:3000,192.168.1.3:3000/app/a",以后的所有操作就都是在/app/a下进行的。实际上只连接到一台ZooKeeper机器就可了,没必要指定每台zk机器的IP和端口,即用“192.168.1.2:3000/app/a”也是可以的。
当client与一个server断连接时(可能是因为server失效了),它就收不到任何watches;当它与另一个server建立好连接后,它就会收到"session expired"通知。
ACL不是递归的,它只针对当前节点,对子节点没有任何影响。
默认情况下日志文件和数据文件是放在同一个目录下的,为缩短延迟提高响应性,你可以把日志文件单独放在另一个目录下。
为避免swaping,运行java时最好把可用物理内在调得大一些,比如对于4G的内在,可以把它调到3G。java有以下两个运行参数:
-Xms<size>
设置虚拟机可用内存堆的初始大小,缺省单位为字节,该大小为1024的整数倍并且要大于1MB,可用k(K)或m(M)为单位来设置较大的内存数。初始堆大小为2MB。
例如:-Xms6400K,-Xms256M
-Xmx<size>
设置虚拟机内存堆的最大可用大小,缺省单位为字节。该值必须为1024整数倍,并且要大于2MB。可用k(K)或m(M)为单位来设置较大的内存数。缺省堆最大值为64MB。
例如:-Xmx81920K,-Xmx80M
CreateMode
PERSISTENT:创建后只要不删就永久存在
EPHEMERAL:会话结束年结点自动被删除,EPHEMERAL结点不允许有子节点
SEQUENTIAL:节点名末尾会自动追加一个10位数的单调递增的序号,同一个节点的所有子节点序号是单调递增的
PERSISTENT_SEQUENTIAL:结合PERSISTENT和SEQUENTIAL
EPHEMERAL_SEQUENTIAL:结合EPHEMERAL和SEQUENTIAL
package basic; import java.io.IOException; import java.util.List; import org.apache.zookeeper.CreateMode; import org.apache.zookeeper.KeeperException; import org.apache.zookeeper.ZooKeeper; import org.apache.zookeeper.ZooDefs.Ids; public class Demo { private static final int TIMEOUT = 3000; public static void main(String[] args) throws IOException { ZooKeeper zkp = new ZooKeeper("localhost:2181", TIMEOUT, null); try { // 创建一个EPHEMERAL类型的节点,会话关闭后它会自动被删除 zkp.create("/node1", "data1".getBytes(), Ids.OPEN_ACL_UNSAFE,CreateMode.EPHEMERAL); if (zkp.exists("/node1", false) != null) { System.out.println("node1 exists now."); } try { // 当节点名已存在时再去创建它会抛出KeeperException(即使本次的ACL、CreateMode和上次的不一样) zkp.create("/node1", "data1".getBytes(), Ids.OPEN_ACL_UNSAFE,CreateMode.PERSISTENT); } catch (KeeperException e) { System.out.println("KeeperException caught:" + e.getMessage()); } // 关闭会话 zkp.close(); zkp = new ZooKeeper("localhost:2181", TIMEOUT, null); //重新建立会话后node1已经不存在了 if (zkp.exists("/node1", false) == null) { System.out.println("node1 dosn't exists now."); } //创建SEQUENTIAL节点 zkp.create("/node-", "same data".getBytes(), Ids.OPEN_ACL_UNSAFE,CreateMode.PERSISTENT_SEQUENTIAL); zkp.create("/node-", "same data".getBytes(), Ids.OPEN_ACL_UNSAFE,CreateMode.PERSISTENT_SEQUENTIAL); zkp.create("/node-", "same data".getBytes(), Ids.OPEN_ACL_UNSAFE,CreateMode.PERSISTENT_SEQUENTIAL); List<String> children = zkp.getChildren("/", null); System.out.println("Children of root node:"); for (String child : children) { System.out.println(child); } zkp.close(); } catch (Exception e) { System.out.println(e.getMessage()); } } }
第一次运行输出:
node1 exists now.
KeeperException caught:KeeperErrorCode = NodeExists for /node1
node1 dosn't exists now.
Children of root node:
node-0000000003
zookeeper
node-0000000002
node-0000000001
第二次运行输出:
node1 exists now.
KeeperException caught:KeeperErrorCode = NodeExists for /node1
node1 dosn't exists now.
Children of root node:
node-0000000003
zookeeper
node-0000000002
node-0000000001
node-0000000007
node-0000000005
node-0000000006
注意两次会话中创建的PERSISTENT_SEQUENTIAL节点序号并不是连续的,比如上例中缺少了node-0000000004.
Watcher & Version
watcher分为两大类:data watches和child watches。getData()和exists()上可以设置data watches,getChildren()上可以设置child watches。
setData()会触发data watches;
create()会触发data watches和child watches;
delete()会触发data watches和child watches.
如果对一个不存在的节点调用了exists(),并设置了watcher,而在连接断开的情况下create/delete了该znode,则watcher会丢失。
在server端用一个map来存放watcher,所以相同的watcher在map中只会出现一次,只要watcher被回调一次,它就会被删除----map解释了watcher的一次性。比如如果在getData()和exists()上设置的是同一个data watcher,调用setData()会触发data watcher,但是getData()和exists()只有一个会收到通知。
1 import java.io.IOException; 2 3 import org.apache.zookeeper.CreateMode; 4 import org.apache.zookeeper.KeeperException; 5 import org.apache.zookeeper.WatchedEvent; 6 import org.apache.zookeeper.Watcher; 7 import org.apache.zookeeper.ZooDefs.Ids; 8 import org.apache.zookeeper.ZooKeeper; 9 import org.apache.zookeeper.data.Stat; 10 11 public class SelfWatcher implements Watcher{ 12 13 ZooKeeper zk=null; 14 15 @Override 16 public void process(WatchedEvent event) { 17 System.out.println(event.toString()); 18 } 19 20 SelfWatcher(String address){ 21 try{ 22 zk=new ZooKeeper(address,3000,this); //在创建ZooKeeper时第三个参数负责设置该类的默认构造函数 23 zk.create("/root", new byte[0], Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL); 24 }catch(IOException e){ 25 e.printStackTrace(); 26 zk=null; 27 }catch (KeeperException e) { 28 e.printStackTrace(); 29 } catch (InterruptedException e) { 30 e.printStackTrace(); 31 } 32 } 33 34 void setWatcher(){ 35 try { 36 Stat s=zk.exists("/root", true); 37 if(s!=null){ 38 zk.getData("/root", false, s); 39 } 40 } catch (KeeperException e) { 41 e.printStackTrace(); 42 } catch (InterruptedException e) { 43 e.printStackTrace(); 44 } 45 } 46 47 void trigeWatcher(){ 48 try { 49 Stat s=zk.exists("/root", false); //此处不设置watcher 50 zk.setData("/root", "a".getBytes(), s.getVersion()); //修改数据时需要提供version,version设为-1表示强制修改 51 }catch(Exception e){ 52 e.printStackTrace(); 53 } 54 } 55 56 void disconnect(){ 57 if(zk!=null) 58 try { 59 zk.close(); 60 } catch (InterruptedException e) { 61 e.printStackTrace(); 62 } 63 } 64 65 public static void main(String[] args){ 66 SelfWatcher inst=new SelfWatcher("127.0.0.1:2181"); 67 inst.setWatcher(); 68 inst.trigeWatcher(); 69 inst.disconnect(); 70 } 71 72 }
可以在创建Zookeeper时指定默认的watcher回调函数,这样在getData()、exists()和getChildren()收到通知时都会调用这个函数--只要它们在参数中设置了true。所以如果把代码22行的this改为null,则不会有任何watcher被注册。
上面的代码输出:
WatchedEvent state:SyncConnected type:None path:null
WatchedEvent state:SyncConnected type:NodeDataChanged path:/root
之所会输出第1 行是因为本身在建立ZooKeeper连接时就会触发watcher。输出每二行是因为在代码的第36行设置了true。
WatchEvent有三种类型:NodeDataChanged、NodeDeleted和NodeChildrenChanged。
调用setData()时会触发NodeDataChanged;
调用create()时会触发NodeDataChanged和NodeChildrenChanged;
调用delete()时上述三种event都会触发。
如果把代码的第36--39行改为:
Stat s=zk.exists("/root", false); if(s!=null){ zk.getData("/root", true, s); }
或
Stat s=zk.exists("/root", true); if(s!=null){ zk.getData("/root", true, s); }
跟上面的输出是一样的。这也证明了watcher是一次性的。
设置watcher的另外一种方式是不使用默认的watcher,而是在getData()、exists()和getChildren()中指定各自的watcher。示例代码如下:
1 public class SelfWatcher{ 2 3 ZooKeeper zk=null; 4 5 private Watcher getWatcher(final String msg){ 6 return new Watcher(){ 7 @Override 8 public void process(WatchedEvent event) { 9 System.out.println(msg+"\t"+event.toString()); 10 } 11 }; 12 } 13 14 SelfWatcher(String address){ 15 try{ 16 zk=new ZooKeeper(address,3000,null); //在创建ZooKeeper时第三个参数负责设置该类的默认构造函数 17 zk.create("/root", new byte[0], Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL); 18 }catch(IOException e){ 19 e.printStackTrace(); 20 zk=null; 21 }catch (KeeperException e) { 22 e.printStackTrace(); 23 } catch (InterruptedException e) { 24 e.printStackTrace(); 25 } 26 } 27 28 void setWatcher(){ 29 try { 30 Stat s=zk.exists("/root", getWatcher("EXISTS")); 31 if(s!=null){ 32 zk.getData("/root", getWatcher("GETDATA"), s); 33 } 34 } catch (KeeperException e) { 35 e.printStackTrace(); 36 } catch (InterruptedException e) { 37 e.printStackTrace(); 38 } 39 } 40 41 void trigeWatcher(){ 42 try { 43 Stat s=zk.exists("/root", false); //此处不设置watcher 44 zk.setData("/root", "a".getBytes(), s.getVersion()); 45 }catch(Exception e){ 46 e.printStackTrace(); 47 } 48 } 49 50 void disconnect(){ 51 if(zk!=null) 52 try { 53 zk.close(); 54 } catch (InterruptedException e) { 55 e.printStackTrace(); 56 } 57 } 58 59 public static void main(String[] args){ 60 SelfWatcher inst=new SelfWatcher("127.0.0.1:2181"); 61 inst.setWatcher(); 62 inst.trigeWatcher(); 63 inst.disconnect(); 64 } 65 66 }
输出:
GETDATA WatchedEvent state:SyncConnected type:NodeDataChanged path:/root
EXISTS WatchedEvent state:SyncConnected type:NodeDataChanged path:/root
上例中由于exists和getData分别设置了两个不同的Watcher实例,所以虽然watcher都是由同了一个NodeDataChanged触发的,但exists()和getData()都会收到通知。由于16行创建Zookeeper时没有设置watcher(参数为null),所以建立连接时没有收到通知。
关于Version:为了方便进行cache validations 和coordinated updates,每个znode都有一个stat结构体,其中包含:version的更改记录、ACL的更改记录、时间戳。znode的数据每更改一次,version就会加1。客户端每次检索data的时候都会把data的version一并读出出来。修改数据时需要提供version。
zk.delete("/root", -1); //触发data watches和children watches。version设为-1时表示要强制删除 zk.getChildren("/root", getWatcher("LISTCHILDREN")); //getChildren()上可以设置children watches
输出:
LISTCHILDREN WatchedEvent state:SyncConnected type:NodeDeleted path:/root
zk.delete("/root", -1); //触发data watches和children watches Stat s=zk.exists("/root", getWatcher("EXISTS")); //exists()上可以设置data watches if(s!=null){ zk.getChildren("/root", getWatcher("LISTCHILDREN")); }
输出:
EXISTS WatchedEvent state:SyncConnected type:NodeDeleted path:/root
LISTCHILDREN WatchedEvent state:SyncConnected type:NodeDeleted path:/root
zk.delete("/root", -1); //触发data watches和children watches Stat s=zk.exists("/root", getWatcher("EXISTS")); if(s!=null){ zk.getData("/root", getWatcher("GETDATA"), s); zk.getChildren("/root", getWatcher("LISTCHILDREN")); }
输出:
GETDATA WatchedEvent state:SyncConnected type:NodeDeleted path:/root
LISTCHILDREN WatchedEvent state:SyncConnected type:NodeDeleted path:/root
EXISTS WatchedEvent state:SyncConnected type:NodeDeleted path:/root
tat s=zk.exists("/root", false); zk.setData("/root", "a".getBytes(), s.getVersion()); zk.delete("/root", -1); Stat s=zk.exists("/root", getWatcher("EXISTS")); if(s!=null){ zk.getData("/root", getWatcher("GETDATA"), s); zk.getChildren("/root", getWatcher("LISTCHILDREN")); }
输出:
GETDATA WatchedEvent state:SyncConnected type:NodeDataChanged path:/root
EXISTS WatchedEvent state:SyncConnected type:NodeDataChanged path:/root
LISTCHILDREN WatchedEvent state:SyncConnected type:NodeDeleted path:/root
按说data watches触发了两次,但是exists()和getData()只会收到一次通知。
Barriers and Queues
1)所有的线程都到达barrier后才能进行后续的计算
或者
2)所有的线程都完成自己的计算后才能离开barrier
Double Barrier是指同时具有上述两点。
Queue就不说了,一个产生--消费模型,先生产的先被消费。
Double Barrier的实现:
enter barrier:
1.建一个根节点"/root"
2.想进入barrier的线程在"/root"下建立一个子节点"/root/c_i"
3.循环监听"/root"孩子节点数目的变化,当其达到size时就说明有size个线程都已经barrier点了
leave barrier:
1.想离开barrier的线程删除其在"/root"下建立的子节点
2.循环监听"/root"孩子节点数目的变化,当size减到0时它就可以离开barrier了
Queue的实现:
1.建立一个根节点"/root"
2.生产线程在"/root"下建立一个SEQUENTIAL子节点
3.消费线程检查"/root"有没有子节点,如果没有就循环监听"/root"子节点的变化,直到它有子节点。删除序号最小的子节点。
原代码:
package sync; import java.io.IOException; import java.net.InetAddress; import java.net.UnknownHostException; import java.nio.ByteBuffer; import java.util.List; import java.util.Random; import org.apache.zookeeper.CreateMode; import org.apache.zookeeper.KeeperException; import org.apache.zookeeper.WatchedEvent; import org.apache.zookeeper.Watcher; import org.apache.zookeeper.ZooKeeper; import org.apache.zookeeper.ZooDefs.Ids; import org.apache.zookeeper.data.Stat; public class SyncPrimitive implements Watcher { static ZooKeeper zk = null; static Integer mutex; String root; //同步原语 SyncPrimitive(String address) { if (zk == null) { try { System.out.println("Starting ZK:"); //建立Zookeeper连接,并且指定watcher zk = new ZooKeeper(address, 3000, this); //初始化锁对象 mutex = new Integer(-1); System.out.println("Finished starting ZK:" + zk); } catch (IOException e) { System.out.println(e.toString()); zk = null; } } } @Override synchronized public void process(WatchedEvent event) { synchronized (mutex) { //有事件发生时,调用notify,使其他wait()点得以继续 mutex.notify(); } } static public class Barrier extends SyncPrimitive { int size; String name; Barrier(String address, String root, int size) { super(address); this.root = root; this.size = size; if (zk != null) { try { //一个barrier建立一个根目录 Stat s = zk.exists(root, false); //不注册watcher if (s == null) { zk.create(root, new byte[0], Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT); } } catch (KeeperException e) { System.out .println("keeper exception when instantiating queue:" + e.toString()); } catch (InterruptedException e) { System.out.println("Interrupted exception."); } } try { //获取自己的主机名 name = new String(InetAddress.getLocalHost() .getCanonicalHostName().toString()); } catch (UnknownHostException e) { System.out.println(e.toString()); } } boolean enter() throws KeeperException, InterruptedException { //在根目录下创建一个子节点.create和delete都会触发children wathes,这样getChildren就会收到通知,process()就会被调用 zk.create(root + "/" + name, new byte[0], Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL_SEQUENTIAL); //一直等,直到根目录下的子节点数目达到size时,函数退出 while (true) { synchronized (mutex) { List<String> list = zk.getChildren(root, true); if (list.size() < size) { mutex.wait(); //释放mutex上的锁 } else { return true; } } } } boolean leave() throws KeeperException, InterruptedException { //删除自己创建的节点 zk.delete(root + "/" + name, 0); //一直等,直到根目录下有子节点时,函数退出 while (true) { synchronized (mutex) { List<String> list = zk.getChildren(root, true); if (list.size() > 0) { mutex.wait(); } else { return true; } } } } } static public class Queue extends SyncPrimitive { Queue(String address, String name) { super(address); this.root = name; if (zk != null) { try { //一个queue建立一个根目录 Stat s = zk.exists(root, false); if (s == null) { zk.create(root, new byte[0], Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT); } } catch (KeeperException e) { System.out .println("keeper exception when instantiating queue:" + e.toString()); } catch (InterruptedException e) { System.out.println("Interrupted exception."); } } } //参数i是要创建节点的data boolean produce(int i) throws KeeperException, InterruptedException { ByteBuffer b = ByteBuffer.allocate(4); byte[] value; b.putInt(i); value = b.array(); //根目录下创建一个子节点,因为是SEQUENTIAL的,所以先创建的节点具有较小的序号 zk.create(root + "/element", value, Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT_SEQUENTIAL); return true; } int consume() throws KeeperException, InterruptedException { int retvalue = -1; Stat stat = null; while (true) { synchronized (mutex) { List<String> list = zk.getChildren(root, true); //并不能保证list[0]就是序号最小的 //如果根目录下没有子节点就一直等 if (list.size() == 0) { System.out.println("Going to wait"); mutex.wait(); } //找到序号最小的节点将其删除 else { Integer min = new Integer(list.get(0).substring(7)); for (String s : list) { Integer tmp = new Integer(s.substring(7)); if (tmp < min) min = tmp; } System.out.println("Temporary value:" + root + "/element" + min); byte[] b = zk.getData(root + "/element" + min, false, stat); zk.delete(root + "/element" + min, 0); ByteBuffer buffer = ByteBuffer.wrap(b); retvalue = buffer.getInt(); return retvalue; } } } } } public static void main(String[] args) { if (args[0].equals("qTest")) queueTest(args); else barrierTest(args); } private static void barrierTest(String[] args) { Barrier b = new Barrier(args[1], "/b1", new Integer(args[2])); try { boolean flag = b.enter(); System.out.println("Enter barrier:" + args[2]); if (!flag) System.out.println("Error when entering the barrier"); } catch (KeeperException e) { } catch (InterruptedException e) { } Random rand = new Random(); int r = rand.nextInt(100); for (int i = 0; i < r; i++) { try { Thread.sleep(100); } catch (InterruptedException e) { } } try { b.leave(); } catch (KeeperException e) { } catch (InterruptedException e) { } System.out.println("Left barrier"); } private static void queueTest(String[] args) { Queue q = new Queue(args[1], "/app1"); System.out.println("Input:" + args[1]); int i; Integer max = new Integer(args[2]); if (args[3].equals("p")) { System.out.println("Producer"); for (i = 0; i < max; i++) try { q.produce(10 + 1); } catch (KeeperException e) { } catch (InterruptedException e) { } } else { System.out.println("Consumer"); for (i = 0; i < max; i++) try { int r = q.consume(); System.out.println("Item:" + r); } catch (KeeperException e) { i--; } catch (InterruptedException e) { } } } }
Locks
获得锁:
1.创建根节点"/root"
2.在根节点下新建子节点"/root/c-xxxxxx",SEQUENTIAL模式
3.对根节点调用getChildren(),如果第2步创建的节点是所有子节点中序号最小的,则获得锁;否则进入第4步
4.在序号最小的子节点上调用exists(),当序号最小的子节点被删除后返回第3步
释放锁:
删除自己创建的子节点即可
原代码:
package sync; import java.io.IOException; import java.net.InetAddress; import java.util.List; import org.apache.zookeeper.CreateMode; import org.apache.zookeeper.KeeperException; import org.apache.zookeeper.WatchedEvent; import org.apache.zookeeper.Watcher; import org.apache.zookeeper.ZooDefs.Ids; import org.apache.zookeeper.ZooKeeper; import org.apache.zookeeper.data.Stat; public class Locks implements Watcher{ static ZooKeeper zk=null; static Integer mutex=null; String name=null; String path=null; @Override synchronized public void process(WatchedEvent event) { synchronized(mutex){ mutex.notify(); } } Locks(String address){ try{ zk=new ZooKeeper(address,2000,this); zk.create("/lock", new byte[0], Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL); mutex=new Integer(-1); name = new String(InetAddress.getLocalHost().getCanonicalHostName().toString()); }catch(IOException e){ zk=null; } catch (KeeperException e) { e.printStackTrace(); } catch (InterruptedException e) { e.printStackTrace(); } } private int minSeq(List<String> list){ int min=Integer.parseInt(list.get(0).substring(14)); for(int i=1;i<list.size();i++){ if(min<Integer.parseInt(list.get(i).substring(14))) min=Integer.parseInt(list.get(i).substring(14)); } return min; } boolean getLock() throws KeeperException, InterruptedException{ //create方法返回新建的节点的完整路径 path=zk.create("/lock/"+name+"-", new byte[0], Ids.OPEN_ACL_UNSAFE,CreateMode.EPHEMERAL_SEQUENTIAL); int min; while(true){ synchronized(mutex){ List<String> list=zk.getChildren("/lock", false); min=minSeq(list); //如果刚建的节点是根节点的所有子节点中序号最小的,则获得了锁,可以返回true if(min==Integer.parseInt(path.substring(14))){ return true; }else{ mutex.wait(); //等待事件(新建节点或删除节点)发生 while(true){ Stat s=zk.exists("/lock/"+name+"-"+min, true); //查看序号最小的子节点还在不在 if(s!=null) //如果还在,则继续等待事件发生 mutex.wait(); else //如果不在,则跳外层循环中,查看新的最小序号的子节点是谁 break; } } } } } boolean releaseLock() throws KeeperException, InterruptedException{ if(path!=null){ zk.delete(path, -1); path=null; } return true; } public static void main(String []args) throws KeeperException, InterruptedException{ Locks lock1=new Locks("localhost:2181"); if(lock1.getLock()){ System.out.println("T1 Get lock at "+System.currentTimeMillis()); for(int i=0;i<1000;++i) Thread.sleep(5000); lock1.releaseLock(); } Locks lock2=new Locks("localhost:2181"); if(lock2.getLock()){ System.out.println("T2 Get lock at "+System.currentTimeMillis()); lock2.releaseLock(); } } }
读锁(共享锁)和写锁(排斥锁)并存的情况跟单独只有排斥锁的情况有几点不同:
1.当一个线程想施加读锁时就新建一个节点"/root/read-xxxxxx",施加写锁时就新建一个节点"/root/write-xxxxxx";
2.欲施加读锁的线程查看"/root"下有没有“write"开头的节点,如果没有则直接获得读锁;如果有,但是"write"节点的序号比自己刚才创建的"read"节点的序号要大说明是先施加的读锁后施加的写锁,所以依然获得读锁;else,在序号最小的"write"节点上调用exists,等待它被删除。
本文来自博客园,作者:高性能golang,转载请注明原文链接:https://www.cnblogs.com/zhangchaoyang/articles/2536178.html