hadoop-common-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Apache Wiki <wikidi...@apache.org>
Subject [Hadoop Wiki] Update of "ZooKeeper/Tutorial" by Flavio Junqueira
Date Thu, 24 Jul 2008 10:53:21 GMT
Dear Wiki user,

You have subscribed to a wiki page or wiki category on "Hadoop Wiki" for change notification.

The following page has been changed by Flavio Junqueira:
http://wiki.apache.org/hadoop/ZooKeeper/Tutorial

New page:
= Programming with ZooKeeper - A quick tutorial =

In this tutorial, we show simple implementations of barriers and producer-consumer queues
using ZooKeeper. We call the respective classes Barrier and Queue. These examples assume that
you have at least one ZooKeeper server running.

Both primitives use the following common excerpt of code:

{{{
    static ZooKeeper zk = null;
    static Integer mutex;

    String root;

    SyncPrimitive(String address) {
        if(zk == null){
            try {
                System.out.println("Starting ZK:");
                zk = new ZooKeeper(address, 3000, this);
                mutex = new Integer(-1);
                System.out.println("Finished starting ZK: " + zk);
            } catch (KeeperException e) {
                System.out.println("Keeper exception when starting new session: "
                        + e.toString());
                zk = null;
            } catch (IOException e) {
                System.out.println(e.toString());
                zk = null;
            }
        }
    }

    synchronized public void process(WatcherEvent event) {
        synchronized (mutex) {
            mutex.notify();
        }
    }

}}}

Both classes extend SyncPrimitive. In this way, we execute steps that are common to all primitives
in the constructor of SyncPrimitive. To keep the examples simple, we create a ZooKeeper object
the first time we instantiate either a barrier object or a queue object, and we declare a
static variable that is a reference to this object. The subsequent instances of Barrier and
Queue check whether a ZooKeeper object exists. Alternatively, we could have the application
creating a ZooKeeper object and passing it to the constructor of Barrier and Queue.

We use the process() method to process notifications triggered due to watches. In the following
discussion, we present code that sets watches. A watch is internal structure that enables
ZooKeeper to notify a client of a change to a node. For example, if a client is waiting for
other clients to leave a barrier, then it can set a watch and wait for modifications to a
particular node, which can indicate that it is the end of the wait. This point becomes clear
once we go over the examples.

== Barriers ==
A barrier is a primitive that enables a group of processes to synchronize the beginning and
the end of a computation. The general idea of this implementation is to have a barrier node
that serves the purpose of being a parent for individual process nodes. Suppose that we call
the barrier node "/b1". Each process "p" then creates a node "/b1/p". Once enough processes
have created their corresponding nodes, joined processes can start the computation.

In this example, each process instantiates a Barrier object, and its constructor takes as
parameters:

    * The address of a ZooKeeper server (e.g., "zoo1.foo.com:2181");
    * The path of the barrier node on ZooKeeper (e.g., "/b1");
    * The size of the group of processes.


The constructor of Barrier passes the address of the Zookeeper server to the constructor of
the parent class. The parent class creates a ZooKeeper instance if one does not exist. The
constructor of Barrier then creates a barrier node on ZooKeeper, which is the parent node
of all process nodes, and we call root (obs: this is not the ZooKeeper root "/").

{{{
      /**
         * Barrier constructor
         *
         * @param address
         * @param name
         * @param size
         */
        Barrier(String address, String name, int size) {
            super(address);
            this.root = name;
            this.size = size;

            // Create barrier node
            if (zk != null) {
                try {
                    Stat s = zk.exists(root, false);
                    if (s == null) {
                        zk.create(root, new byte[0], Ids.OPEN_ACL_UNSAFE, 0);
                    }
                } catch (KeeperException e) {
                    System.out.println("Keeper exception when instantiating queue: " + e.toString());
                } catch (InterruptedException e) {
                    System.out.println("Interrupted exception");
                }
            }

            // My node name
            try {
                name = new String(InetAddress.getLocalHost().getCanonicalHostName().toString());
            } catch (UnknownHostException e) {
                System.out.println(e.toString());
            }

        }
}}}

To enter the barrier, a process calls enter(). The process creates a node under the root to
represent it, using its host name to form the node name. It then wait until enough processes
have entered the barrier. A process does it by checking the number of children the root node
has with "getChildren()", and waiting for notifications in the case it does not have enough.
To receive a notification when there is a change to the root node, a process has to set a
watch, and does it through the call to "getChildren()". In the code, we have that "getChildren()"
has two parameters. The first one states the node to read from, and the second is a boolean
flag that enables the process to set a watch. In the code the flag is true.


{{{
    /**
         * Join barrier
         *
         * @return
         * @throws KeeperException
         * @throws InterruptedException
         */

        boolean enter() throws KeeperException, InterruptedException{
            zk.create(root + "/" + name, new byte[0], Ids.OPEN_ACL_UNSAFE,
                    CreateFlags.EPHEMERAL);
            while (true) {
                synchronized (mutex) {
                    ArrayList<String> list = zk.getChildren(root, true);

                    if (list.size() < size) {
                        mutex.wait();
                    } else {
                        return true;
                    }
                }
            }
        }
}}}

Note that enter() throws both KeeperException and InterruptedException, so it is reponsability
of the application to catch and handle such exceptions.


Once the computation is finished, a process calls leave() to leave the barrier. First it deletes
its corresponding node, and then it gets the children of the root node. If there is at least
one child, then it waits for a notification (obs: note that the second parameter of the call
to getChildren() is true, meaning that ZooKeeper has to set a watch on the the root node).
Upon reception of a notification, it checks once more whether the root node has any child.

{{{
     /**
         * Wait until all reach barrier
         *
         * @return
         * @throws KeeperException
         * @throws InterruptedException
         */

        boolean leave() throws KeeperException, InterruptedException{
            zk.delete(root + "/" + name, 0);
            while (true) {
                synchronized (mutex) {
                    ArrayList<String> list = zk.getChildren(root, true);
                        if (list.size() > 0) {
                            mutex.wait();
                        } else {
                            return true;
                        }
                    }
                }
        }
}}}

== Producer-Consumer Queues ==

A producer-consumer queue is a distributed data estructure thata group of processes use to
generate and consume items. Producer processes create new elements and add them to the queue.
Consumer processes remove elements from the list, and process them. In this implementation,
the elements are simple integers. The queue is represented by a root node, and to add an element
to the queue, a producer process creates a new node, a child of the root node.

The following excerpt of code corresponds to the constructor of the object. As with Barrier
objects, it first calls the constructor of the parent class, SyncPrimitive, that creates a
ZooKeeper object if one doesn't exist. It then verifies if the root node of the queue exists,
and creates if it doesn't.

{{{
     /**
         * Constructor of producer-consumer queue
         *
         * @param address
         * @param name
         */
        Queue(String address, String name) {
            super(address);
            this.root = name;
            // Create ZK node name
            if (zk != null) {
                try {
                    Stat s = zk.exists(root, false);
                    if (s == null) {
                        zk.create(root, new byte[0], Ids.OPEN_ACL_UNSAFE, 0);
                    }
                } catch (KeeperException e) {
                    System.out
                            .println("Keeper exception when instantiating queue: "
                                    + e.toString());
                } catch (InterruptedException e) {
                    System.out.println("Interrupted exception");
                }
            }
        }
}}}

A producer process calls "produce()" to add an element to the queue, and passes an integer
as an argument. To add an element to the queue, the method creates a new node using "create()",
and uses the SEQUENCE flag to instruct ZooKeeper to append the value of the sequencer counter
associated to the root node. In this way, we impose a total order on the elements of the queue,
thus guaranteeing that the oldest element of the queue is the next one consumed.


{{{
    /**
         * Add element to the queue.
         *
         * @param i
         * @return
         */

        boolean produce(int i) throws KeeperException, InterruptedException{
            ByteBuffer b = ByteBuffer.allocate(4);
            byte[] value;

            // Add child with value i
            b.putInt(i);
            value = b.array();
            zk.create(root + "/element", value, Ids.OPEN_ACL_UNSAFE,
                        CreateFlags.SEQUENCE);

            return true;
        }
}}}

To consume an element, a consumer process obtains the children of the root node, reads the
node with smallest counter value, and returns the element. Note that if there is a conflict,
then one of the two contending processes won't be able to delete the node and the delete operation
will throw an exception.

A call to getChildren() returns the list of children in lexicographic order. As lexicographic
order does not necessary follow the numerical order of the counter values, we need to decide
which element is the smallest. To decide which one has the smallest counter value, we traverse
the list, and remove the prefix "element" from each one.

{{{

     /**
         * Remove first element from the queue.
         *
         * @return
         * @throws KeeperException
         * @throws InterruptedException
         */
        int consume() throws KeeperException, InterruptedException{
            int retvalue = -1;
            Stat stat = null;

            // Get the first element available
            while (true) {
                synchronized (mutex) {
                    ArrayList<String> list = zk.getChildren(root, true);
                    if (list.size() == 0) {
                        System.out.println("Going to wait");
                        mutex.wait();
                    } else {
                        Integer min = new Integer(list.get(0).substring(7));
                        for(String s : list){
                            Integer tempValue = new Integer(s.substring(7));
                            if(tempValue < min) min = tempValue;
                        }
                        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;
                    }
                }
            }
        }
}}}

Mime
View raw message