You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@helix.apache.org by ka...@apache.org on 2014/01/02 01:14:05 UTC
[05/31] Redesign documentation for 0.6.2, 0.7.0, and trunk
http://git-wip-us.apache.org/repos/asf/incubator-helix/blob/4a4510d1/site-releases/0.7.0-incubating/src/site/markdown/UseCases.md
----------------------------------------------------------------------
diff --git a/site-releases/0.7.0-incubating/src/site/markdown/UseCases.md b/site-releases/0.7.0-incubating/src/site/markdown/UseCases.md
deleted file mode 100644
index 001b012..0000000
--- a/site-releases/0.7.0-incubating/src/site/markdown/UseCases.md
+++ /dev/null
@@ -1,113 +0,0 @@
-<!---
-Licensed to the Apache Software Foundation (ASF) under one
-or more contributor license agreements. See the NOTICE file
-distributed with this work for additional information
-regarding copyright ownership. The ASF licenses this file
-to you under the Apache License, Version 2.0 (the
-"License"); you may not use this file except in compliance
-with the License. You may obtain a copy of the License at
-
- http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing,
-software distributed under the License is distributed on an
-"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
-KIND, either express or implied. See the License for the
-specific language governing permissions and limitations
-under the License.
--->
-
-<head>
- <title>Use Cases</title>
-</head>
-
-
-# Use cases at LinkedIn
-
-At LinkedIn Helix framework is used to manage 3 distributed data systems which are quite different from each other.
-
-* Espresso
-* Databus
-* Search As A Service
-
-## Espresso
-
-Espresso is a distributed, timeline consistent, scal- able, document store that supports local secondary indexing and local transactions.
-Espresso databases are horizontally partitioned into a number of partitions, with each partition having a certain number of replicas
-distributed across the storage nodes.
-Espresso designates one replica of each partition as master and the rest as slaves; only one master may exist for each partition at any time.
-Espresso enforces timeline consistency where only the master of a partition can accept writes to its records, and all slaves receive and
-apply the same writes through a replication stream.
-For load balancing, both master and slave partitions are assigned evenly across all storage nodes.
-For fault tolerance, it adds the constraint that no two replicas of the same partition may be located on the same node.
-
-### State model
-Espresso follows a Master-Slave state model. A replica can be in Offline,Slave or Master state.
-The state machine table describes the next state given the Current State, Final State
-
-```
- OFFLINE | SLAVE | MASTER
- _____________________________
- | | | |
-OFFLINE | N/A | SLAVE | SLAVE |
- |__________|________|_________|
- | | | |
-SLAVE | OFFLINE | N/A | MASTER |
- |__________|________|_________|
- | | | |
-MASTER | SLAVE | SLAVE | N/A |
- |__________|________|_________|
-
-```
-
-### Constraints
-* Max number of replicas in Master state:1
-* Execution mode AUTO. i.e on node failure no new replicas will be created. Only the State of remaining replicas will be changed.
-* Number of mastered partitions on each node must be approximately same.
-* The above constraint must be satisfied when a node fails or a new node is added.
-* When new nodes are added the number of partitions moved must be minimized.
-* When new nodes are added the max number of OFFLINE-SLAVE transitions that can happen concurrently on new node is X.
-
-## Databus
-
-Databus is a change data capture (CDC) system that provides a common pipeline for transporting events
-from LinkedIn primary databases to caches within various applications.
-Databus deploys a cluster of relays that pull the change log from multiple databases and
-let consumers subscribe to the change log stream. Each Databus relay connects to one or more database servers and
-hosts a certain subset of databases (and partitions) from those database servers.
-
-For a large partitioned database (e.g. Espresso), the change log is consumed by a bank of consumers.
-Each databus partition is assigned to a consumer such that partitions are evenly distributed across consumers and each partition is
-assigned to exactly one consumer at a time. The set of consumers may grow over time, and consumers may leave the group due to planned or unplanned
-outages. In these cases, partitions must be reassigned, while maintaining balance and the single consumer-per-partition invariant.
-
-### State model
-Databus consumers follow a simple Offline-Online state model.
-The state machine table describes the next state given the Current State, Final State
-
-<pre><code>
- OFFLINE | ONLINE |
- ___________________|
- | | |
-OFFLINE | N/A | ONLINE |
- |__________|________|
- | | |
-ONLINE | OFFLINE | N/A |
- |__________|________|
-
-
-</code></pre>
-
-
-## Search As A Service
-
-LinkedIn�s Search-as-a-service lets internal customers define custom indexes on a chosen dataset
-and then makes those indexes searchable via a service API. The index service runs on a cluster of machines.
-The index is broken into partitions and each partition has a configured number of replicas.
-Each cluster server runs an instance of the Sensei system (an online index store) and hosts index partitions.
-Each new indexing service gets assigned to a set of servers, and the partition replicas must be evenly distributed across those servers.
-
-### State model
-
-
-
http://git-wip-us.apache.org/repos/asf/incubator-helix/blob/4a4510d1/site-releases/0.7.0-incubating/src/site/markdown/index.md
----------------------------------------------------------------------
diff --git a/site-releases/0.7.0-incubating/src/site/markdown/index.md b/site-releases/0.7.0-incubating/src/site/markdown/index.md
index 333110b..b625fd7 100644
--- a/site-releases/0.7.0-incubating/src/site/markdown/index.md
+++ b/site-releases/0.7.0-incubating/src/site/markdown/index.md
@@ -18,23 +18,18 @@ under the License.
-->
<head>
- <title>Home</title>
+ <title>Helix 0.7.0-incubating Documentation [ALPHA]</title>
</head>
-__Alpha!__ This release contains many new features, but things might not work just right, and some APIs are still in the process of being developed.
+### Get Helix
-Navigating the Documentation
-----------------------------
+[Download](./download.html)
-### Conceptual Understanding
+[Building](./Building.html)
-[Concepts / Terminology](./Concepts.html)
+[Release Notes](./releasenotes/release-0.7.0-incubating.html)
-[Architecture](./Architecture.html)
-
-### Hands-on Helix
-
-[Getting Helix](./Building.html)
+### Hands-On
[Quickstart](./Quickstart.html)
@@ -52,11 +47,12 @@ Navigating the Documentation
[Service discovery](./recipes/service_discovery.html)
-[Distributed Task DAG Execution](./recipes/task_dag_execution.html)
+[Distributed task DAG execution](./recipes/task_dag_execution.html)
[User-Defined Rebalancer Example](./recipes/user_def_rebalancer.html)
-### Download
+---
-[0.7.0-incubating](./download.html)
+### Alpha!
+This release contains many new features, but things might not work just right, and some APIs are still in the process of being developed.
http://git-wip-us.apache.org/repos/asf/incubator-helix/blob/4a4510d1/site-releases/0.7.0-incubating/src/site/markdown/recipes/lock_manager.md
----------------------------------------------------------------------
diff --git a/site-releases/0.7.0-incubating/src/site/markdown/recipes/lock_manager.md b/site-releases/0.7.0-incubating/src/site/markdown/recipes/lock_manager.md
index 252ace7..34950d2 100644
--- a/site-releases/0.7.0-incubating/src/site/markdown/recipes/lock_manager.md
+++ b/site-releases/0.7.0-incubating/src/site/markdown/recipes/lock_manager.md
@@ -16,21 +16,21 @@ KIND, either express or implied. See the License for the
specific language governing permissions and limitations
under the License.
-->
-Distributed lock manager
+Distributed Lock Manager
------------------------
-Distributed locks are used to synchronize accesses shared resources. Most applications use Zookeeper to model the distributed locks.
+Distributed locks are used to synchronize accesses shared resources. Most applications today use ZooKeeper to model distributed locks.
-The simplest way to model a lock using zookeeper is (See Zookeeper leader recipe for an exact and more advanced solution)
+The simplest way to model a lock using ZooKeeper is (See ZooKeeper leader recipe for an exact and more advanced solution)
-* Each process tries to create an emphemeral node.
-* If can successfully create it then, it acquires the lock
-* Else it will watch on the znode and try to acquire the lock again if the current lock holder disappears
+* Each process tries to create an emphemeral node
+* If the node is successfully created, the process acquires the lock
+* Otherwise, it will watch the ZNode and try to acquire the lock again if the current lock holder disappears
-This is good enough if there is only one lock. But in practice, an application will need many such locks. Distributing and managing the locks among difference process becomes challenging. Extending such a solution to many locks will result in
+This is good enough if there is only one lock. But in practice, an application will need many such locks. Distributing and managing the locks among difference process becomes challenging. Extending such a solution to many locks will result in:
-* Uneven distribution of locks among nodes, the node that starts first will acquire all the lock. Nodes that start later will be idle.
-* When a node fails, how the locks will be distributed among remaining nodes is not predicable.
-* When new nodes are added the current nodes dont relinquish the locks so that new nodes can acquire some locks
+* Uneven distribution of locks among nodes; the node that starts first will acquire all the locks. Nodes that start later will be idle.
+* When a node fails, how the locks will be distributed among remaining nodes is not predicable.
+* When new nodes are added the current nodes don\'t relinquish the locks so that new nodes can acquire some locks
In other words we want a system to satisfy the following requirements.
@@ -38,28 +38,29 @@ In other words we want a system to satisfy the following requirements.
* If a node fails, the locks that were acquired by that node should be evenly distributed among other nodes
* If nodes are added, locks must be evenly re-distributed among nodes.
-Helix provides a simple and elegant solution to this problem. Simply specify the number of locks and Helix will ensure that above constraints are satisfied.
+Helix provides a simple and elegant solution to this problem. Simply specify the number of locks and Helix will ensure that above constraints are satisfied.
-To quickly see this working run the lock-manager-demo script where 12 locks are evenly distributed among three nodes, and when a node fails, the locks get re-distributed among remaining two nodes. Note that Helix does not re-shuffle the locks completely, instead it simply distributes the locks relinquished by dead node among 2 remaining nodes evenly.
+To quickly see this working run the `lock-manager-demo` script where 12 locks are evenly distributed among three nodes, and when a node fails, the locks get re-distributed among remaining two nodes. Note that Helix does not re-shuffle the locks completely, instead it simply distributes the locks relinquished by dead node among 2 remaining nodes evenly.
----------------------------------------------------------------------------------------
-#### Short version
- This version starts multiple threads with in same process to simulate a multi node deployment. Try the long version to get a better idea of how it works.
-
+### Short Version
+This version starts multiple threads within the same process to simulate a multi node deployment. Try the long version to get a better idea of how it works.
+
```
git clone https://git-wip-us.apache.org/repos/asf/incubator-helix.git
cd incubator-helix
+git checkout tags/helix-0.7.0-incubating
mvn clean install package -DskipTests
cd recipes/distributed-lock-manager/target/distributed-lock-manager-pkg/bin
chmod +x *
./lock-manager-demo
```
-##### Output
+#### Output
```
-./lock-manager-demo
+./lock-manager-demo
STARTING localhost_12000
STARTING localhost_12002
STARTING localhost_12001
@@ -117,83 +118,74 @@ lock-group_9 localhost_12001
----------------------------------------------------------------------------------------
-#### Long version
+### Long version
This provides more details on how to setup the cluster and where to plugin application code.
-##### start zookeeper
+#### Start ZooKeeper
```
./start-standalone-zookeeper 2199
```
-##### Create a cluster
+#### Create a Cluster
```
./helix-admin --zkSvr localhost:2199 --addCluster lock-manager-demo
```
-##### Create a lock group
+#### Create a Lock Group
-Create a lock group and specify the number of locks in the lock group.
+Create a lock group and specify the number of locks in the lock group.
```
-./helix-admin --zkSvr localhost:2199 --addResource lock-manager-demo lock-group 6 OnlineOffline FULL_AUTO
+./helix-admin --zkSvr localhost:2199 --addResource lock-manager-demo lock-group 6 OnlineOffline AUTO_REBALANCE
```
-##### Start the nodes
+#### Start the Nodes
-Create a Lock class that handles the callbacks.
+Create a Lock class that handles the callbacks.
```
-
-public class Lock extends StateModel
-{
+public class Lock extends StateModel {
private String lockName;
- public Lock(String lockName)
- {
+ public Lock(String lockName) {
this.lockName = lockName;
}
- public void lock(Message m, NotificationContext context)
- {
+ public void lock(Message m, NotificationContext context) {
System.out.println(" acquired lock:"+ lockName );
}
- public void release(Message m, NotificationContext context)
- {
+ public void release(Message m, NotificationContext context) {
System.out.println(" releasing lock:"+ lockName );
}
}
-
```
-LockFactory that creates the lock
-
+and a LockFactory that creates Locks
+
```
-public class LockFactory extends StateModelFactory<Lock>{
-
- /* Instantiates the lock handler, one per lockName*/
- public Lock create(String lockName)
- {
+public class LockFactory extends StateModelFactory<Lock> {
+ /* Instantiates the lock handler, one per lockName */
+ public Lock create(String lockName) {
return new Lock(lockName);
- }
+ }
}
```
-At node start up, simply join the cluster and helix will invoke the appropriate callbacks on Lock instance. One can start any number of nodes and Helix detects that a new node has joined the cluster and re-distributes the locks automatically.
+At node start up, simply join the cluster and Helix will invoke the appropriate callbacks on the appropriate Lock instance. One can start any number of nodes and Helix detects that a new node has joined the cluster and re-distributes the locks automatically.
```
-public class LockProcess{
-
- public static void main(String args){
+public class LockProcess {
+ public static void main(String args) {
String zkAddress= "localhost:2199";
String clusterName = "lock-manager-demo";
//Give a unique id to each process, most commonly used format hostname_port
String instanceName ="localhost_12000";
ZKHelixAdmin helixAdmin = new ZKHelixAdmin(zkAddress);
- //configure the instance and provide some metadata
+ //configure the instance and provide some metadata
InstanceConfig config = new InstanceConfig(instanceName);
config.setHostName("localhost");
config.setPort("12000");
@@ -207,47 +199,38 @@ public class LockProcess{
manager.getStateMachineEngine().registerStateModelFactory("OnlineOffline", modelFactory);
manager.connect();
Thread.currentThread.join();
- }
-
+ }
}
```
-##### Start the controller
+#### Start the Controller
-Controller can be started either as a separate process or can be embedded within each node process
+The controller can be started either as a separate process or can be embedded within each node process
-###### Separate process
-This is recommended when number of nodes in the cluster >100. For fault tolerance, you can run multiple controllers on different boxes.
+##### Separate Process
+This is recommended when number of nodes in the cluster \> 100. For fault tolerance, you can run multiple controllers on different boxes.
```
./run-helix-controller --zkSvr localhost:2199 --cluster lock-manager-demo 2>&1 > /tmp/controller.log &
```
-###### Embedded within the node process
+##### Embedded Within the Node Process
This is recommended when the number of nodes in the cluster is less than 100. To start a controller from each process, simply add the following lines to MyClass
```
-public class LockProcess{
-
- public static void main(String args){
+public class LockProcess {
+ public static void main(String args) {
String zkAddress= "localhost:2199";
String clusterName = "lock-manager-demo";
- .
- .
+ // .
+ // .
manager.connect();
HelixManager controller;
- controller = HelixControllerMain.startHelixController(zkAddress,
+ controller = HelixControllerMain.startHelixController(zkAddress,
clusterName,
- "controller",
+ "controller",
HelixControllerMain.STANDALONE);
Thread.currentThread.join();
}
}
```
-
-----------------------------------------------------------------------------------------
-
-
-
-
-
http://git-wip-us.apache.org/repos/asf/incubator-helix/blob/4a4510d1/site-releases/0.7.0-incubating/src/site/markdown/recipes/rabbitmq_consumer_group.md
----------------------------------------------------------------------
diff --git a/site-releases/0.7.0-incubating/src/site/markdown/recipes/rabbitmq_consumer_group.md b/site-releases/0.7.0-incubating/src/site/markdown/recipes/rabbitmq_consumer_group.md
index 9edc2cb..58f4a9e 100644
--- a/site-releases/0.7.0-incubating/src/site/markdown/recipes/rabbitmq_consumer_group.md
+++ b/site-releases/0.7.0-incubating/src/site/markdown/recipes/rabbitmq_consumer_group.md
@@ -19,40 +19,40 @@ under the License.
RabbitMQ Consumer Group
-=======================
+-----------------------
-[RabbitMQ](http://www.rabbitmq.com/) is a well known Open source software the provides robust messaging for applications.
+[RabbitMQ](http://www.rabbitmq.com/) is well-known open source software the provides robust messaging for applications.
-One of the commonly implemented recipes using this software is a work queue. http://www.rabbitmq.com/tutorials/tutorial-four-java.html describes the use case where
+One of the commonly implemented recipes using this software is a work queue. [http://www.rabbitmq.com/tutorials/tutorial-four-java.html](http://www.rabbitmq.com/tutorials/tutorial-four-java.html) describes the use case where
-* A producer sends a message with a routing key.
-* The message is routed to the queue whose binding key exactly matches the routing key of the message.
+* A producer sends a message with a routing key
+* The message is routed to the queue whose binding key exactly matches the routing key of the message
* There are multiple consumers and each consumer is interested in processing only a subset of the messages by binding to the interested keys
The example provided [here](http://www.rabbitmq.com/tutorials/tutorial-four-java.html) describes how multiple consumers can be started to process all the messages.
-While this works, in production systems one needs the following
+While this works, in production systems one needs the following:
-* Ability to handle failures: when a consumers fails another consumer must be started or the other consumers must start processing these messages that should have been processed by the failed consumer.
-* When the existing consumers cannot keep up with the task generation rate, new consumers will be added. The tasks must be redistributed among all the consumers.
+* Ability to handle failures: when a consumer fails, another consumer must be started or the other consumers must start processing these messages that should have been processed by the failed consumer
+* When the existing consumers cannot keep up with the task generation rate, new consumers will be added. The tasks must be redistributed among all the consumers
In this recipe, we demonstrate handling of consumer failures and new consumer additions using Helix.
-Mapping this usecase to Helix is pretty easy as the binding key/routing key is equivalent to a partition.
+Mapping this usecase to Helix is pretty easy as the binding key/routing key is equivalent to a partition.
-Let's take an example. Lets say the queue has 6 partitions, and we have 2 consumers to process all the queues.
-What we want is all 6 queues to be evenly divided among 2 consumers.
+Let's take an example. Lets say the queue has 6 partitions, and we have 2 consumers to process all the queues.
+What we want is all 6 queues to be evenly divided among 2 consumers.
Eventually when the system scales, we add more consumers to keep up. This will make each consumer process tasks from 2 queues.
-Now let's say that a consumer failed which reduces the number of active consumers to 2. This means each consumer must process 3 queues.
+Now let's say that a consumer failed, reducing the number of active consumers to 2. This means each consumer must process 3 queues.
-We showcase how such a dynamic App can be developed using Helix. Even though we use rabbitmq as the pub/sub system one can extend this solution to other pub/sub systems.
+We showcase how such a dynamic application can be developed using Helix. Even though we use RabbitMQ as the pub/sub system one can extend this solution to other pub/sub systems.
-Try it
-======
+### Try It
```
git clone https://git-wip-us.apache.org/repos/asf/incubator-helix.git
cd incubator-helix
+git checkout tags/helix-0.7.0-incubating
mvn clean install package -DskipTests
cd recipes/rabbitmq-consumer-group/bin
chmod +x *
@@ -62,63 +62,60 @@ chmod +x $HELIX_PKG_ROOT/bin/*
chmod +x $HELIX_RABBITMQ_ROOT/bin/*
```
-
-Install Rabbit MQ
-----------------
+#### Install RabbitMQ
Setting up RabbitMQ on a local box is straightforward. You can find the instructions here
http://www.rabbitmq.com/download.html
-Start ZK
---------
-Start zookeeper at port 2199
+#### Start ZK
+
+Start ZooKeeper at port 2199
```
$HELIX_PKG_ROOT/bin/start-standalone-zookeeper 2199
```
-Setup the consumer group cluster
---------------------------------
-This will setup the cluster by creating a "rabbitmq-consumer-group" cluster and adds a "topic" with "6" queues.
+#### Setup the Consumer Group Cluster
+
+This will setup the cluster by creating a "rabbitmq-consumer-group" cluster and adds a "topic" with "6" queues.
```
-$HELIX_RABBITMQ_ROOT/bin/setup-cluster.sh localhost:2199
+$HELIX_RABBITMQ_ROOT/bin/setup-cluster.sh localhost:2199
```
-Add consumers
--------------
-Start 2 consumers in 2 different terminals. Each consumer is given a unique id.
+#### Add Consumers
+
+Start 2 consumers in 2 different terminals. Each consumer is given a unique ID.
```
//start-consumer.sh zookeeperAddress (e.g. localhost:2181) consumerId , rabbitmqServer (e.g. localhost)
-$HELIX_RABBITMQ_ROOT/bin/start-consumer.sh localhost:2199 0 localhost
-$HELIX_RABBITMQ_ROOT/bin/start-consumer.sh localhost:2199 1 localhost
+$HELIX_RABBITMQ_ROOT/bin/start-consumer.sh localhost:2199 0 localhost
+$HELIX_RABBITMQ_ROOT/bin/start-consumer.sh localhost:2199 1 localhost
```
-Start HelixController
---------------------
+#### Start the Helix Controller
+
Now start a Helix controller that starts managing the "rabbitmq-consumer-group" cluster.
```
$HELIX_RABBITMQ_ROOT/bin/start-cluster-manager.sh localhost:2199
```
-Send messages to the Topic
---------------------------
+#### Send Messages to the Topic
-Start sending messages to the topic. This script randomly selects a routing key (1-6) and sends the message to topic.
+Start sending messages to the topic. This script randomly selects a routing key (1-6) and sends the message to topic.
Based on the key, messages gets routed to the appropriate queue.
```
$HELIX_RABBITMQ_ROOT/bin/send-message.sh localhost 20
```
-After running this, you should see all 20 messages being processed by 2 consumers.
+After running this, you should see all 20 messages being processed by 2 consumers.
-Add another consumer
---------------------
-Once a new consumer is started, helix detects it. In order to balance the load between 3 consumers, it deallocates 1 partition from the existing consumers and allocates it to the new consumer. We see that
+#### Add Another Consumer
+
+Once a new consumer is started, Helix detects it. In order to balance the load between 3 consumers, it deallocates 1 partition from the existing consumers and allocates it to the new consumer. We see that
each consumer is now processing only 2 queues.
Helix makes sure that old nodes are asked to stop consuming before the new consumer is asked to start consuming for a given partition. But the transitions for each partition can happen in parallel.
@@ -126,7 +123,7 @@ Helix makes sure that old nodes are asked to stop consuming before the new consu
$HELIX_RABBITMQ_ROOT/bin/start-consumer.sh localhost:2199 2 localhost
```
-Send messages again to the topic.
+Send messages again to the topic
```
$HELIX_RABBITMQ_ROOT/bin/send-message.sh localhost 100
@@ -134,94 +131,83 @@ $HELIX_RABBITMQ_ROOT/bin/send-message.sh localhost 100
You should see that messages are now received by all 3 consumers.
-Stop a consumer
----------------
+#### Stop a Consumer
+
In any terminal press CTRL^C and notice that Helix detects the consumer failure and distributes the 2 partitions that were processed by failed consumer to the remaining 2 active consumers.
-How does it work
-================
+### How does this work?
-Find the entire code [here](https://git-wip-us.apache.org/repos/asf?p=incubator-helix.git;a=tree;f=recipes/rabbitmq-consumer-group/src/main/java/org/apache/helix/recipes/rabbitmq).
-
-Cluster setup
--------------
-This step creates znode on zookeeper for the cluster and adds the state model. We use online offline state model since there is no need for other states. The consumer is either processing a queue or it is not.
+Find the entire code [here](https://git-wip-us.apache.org/repos/asf?p=incubator-helix.git;a=tree;f=recipes/rabbitmq-consumer-group/src/main/java/org/apache/helix/recipes/rabbitmq).
-It creates a resource called "rabbitmq-consumer-group" with 6 partitions. The execution mode is set to FULL_AUTO. This means that the Helix controls the assignment of partition to consumers and automatically distributes the partitions evenly among the active consumers. When a consumer is added or removed, it ensures that a minimum number of partitions are shuffled.
+#### Cluster Setup
-```
- zkclient = new ZkClient(zkAddr, ZkClient.DEFAULT_SESSION_TIMEOUT,
- ZkClient.DEFAULT_CONNECTION_TIMEOUT, new ZNRecordSerializer());
- ZKHelixAdmin admin = new ZKHelixAdmin(zkclient);
-
- // add cluster
- admin.addCluster(clusterName, true);
+This step creates ZNode on ZooKeeper for the cluster and adds the state model. We use online offline state model since there is no need for other states. The consumer is either processing a queue or it is not.
- // add state model definition
- StateModelConfigGenerator generator = new StateModelConfigGenerator();
- admin.addStateModelDef(clusterName, "OnlineOffline",
- new StateModelDefinition(generator.generateConfigForOnlineOffline()));
+It creates a resource called "rabbitmq-consumer-group" with 6 partitions. The execution mode is set to AUTO_REBALANCE. This means that the Helix controls the assignment of partition to consumers and automatically distributes the partitions evenly among the active consumers. When a consumer is added or removed, it ensures that a minimum number of partitions are shuffled.
- // add resource "topic" which has 6 partitions
- String resourceName = "rabbitmq-consumer-group";
- admin.addResource(clusterName, resourceName, 6, "OnlineOffline", "FULL_AUTO");
```
+zkclient = new ZkClient(zkAddr, ZkClient.DEFAULT_SESSION_TIMEOUT,
+ ZkClient.DEFAULT_CONNECTION_TIMEOUT, new ZNRecordSerializer());
+ZKHelixAdmin admin = new ZKHelixAdmin(zkclient);
+
+// add cluster
+admin.addCluster(clusterName, true);
-Starting the consumers
-----------------------
-The only thing consumers need to know is the zkaddress, cluster name and consumer id. It does not need to know anything else.
+// add state model definition
+StateModelConfigGenerator generator = new StateModelConfigGenerator();
+admin.addStateModelDef(clusterName, "OnlineOffline",
+ new StateModelDefinition(generator.generateConfigForOnlineOffline()));
+// add resource "topic" which has 6 partitions
+String resourceName = "rabbitmq-consumer-group";
+admin.addResource(clusterName, resourceName, 6, "OnlineOffline", "AUTO_REBALANCE");
```
- _manager =
- HelixManagerFactory.getZKHelixManager(_clusterName,
- _consumerId,
- InstanceType.PARTICIPANT,
- _zkAddr);
- StateMachineEngine stateMach = _manager.getStateMachineEngine();
- ConsumerStateModelFactory modelFactory =
- new ConsumerStateModelFactory(_consumerId, _mqServer);
- stateMach.registerStateModelFactory("OnlineOffline", modelFactory);
+### Starting the Consumers
- _manager.connect();
+The only thing consumers need to know is the ZooKeeper address, cluster name and consumer ID. It does not need to know anything else.
```
+_manager = HelixManagerFactory.getZKHelixManager(_clusterName,
+ _consumerId,
+ InstanceType.PARTICIPANT,
+ _zkAddr);
-Once the consumer has registered the statemodel and the controller is started, the consumer starts getting callbacks (onBecomeOnlineFromOffline) for the partition it needs to host. All it needs to do as part of the callback is to start consuming messages from the appropriate queue. Similarly, when the controller deallocates a partitions from a consumer, it fires onBecomeOfflineFromOnline for the same partition.
-As a part of this transition, the consumer will stop consuming from a that queue.
+StateMachineEngine stateMach = _manager.getStateMachineEngine();
+ConsumerStateModelFactory modelFactory =
+ new ConsumerStateModelFactory(_consumerId, _mqServer);
+stateMach.registerStateModelFactory("OnlineOffline", modelFactory);
+_manager.connect();
```
- @Transition(to = "ONLINE", from = "OFFLINE")
- public void onBecomeOnlineFromOffline(Message message, NotificationContext context)
- {
- LOG.debug(_consumerId + " becomes ONLINE from OFFLINE for " + _partition);
-
- if (_thread == null)
- {
- LOG.debug("Starting ConsumerThread for " + _partition + "...");
- _thread = new ConsumerThread(_partition, _mqServer, _consumerId);
- _thread.start();
- LOG.debug("Starting ConsumerThread for " + _partition + " done");
-
- }
- }
-
- @Transition(to = "OFFLINE", from = "ONLINE")
- public void onBecomeOfflineFromOnline(Message message, NotificationContext context)
- throws InterruptedException
- {
- LOG.debug(_consumerId + " becomes OFFLINE from ONLINE for " + _partition);
- if (_thread != null)
- {
- LOG.debug("Stopping " + _consumerId + " for " + _partition + "...");
+Once the consumer has registered the state model and the controller is started, the consumer starts getting callbacks (onBecomeOnlineFromOffline) for the partition it needs to host. All it needs to do as part of the callback is to start consuming messages from the appropriate queue. Similarly, when the controller deallocates a partitions from a consumer, it fires onBecomeOfflineFromOnline for the same partition.
+As a part of this transition, the consumer will stop consuming from a that queue.
- _thread.interrupt();
- _thread.join(2000);
- _thread = null;
- LOG.debug("Stopping " + _consumerId + " for " + _partition + " done");
+```
+@Transition(to = "ONLINE", from = "OFFLINE")
+public void onBecomeOnlineFromOffline(Message message, NotificationContext context) {
+ LOG.debug(_consumerId + " becomes ONLINE from OFFLINE for " + _partition);
+ if (_thread == null) {
+ LOG.debug("Starting ConsumerThread for " + _partition + "...");
+ _thread = new ConsumerThread(_partition, _mqServer, _consumerId);
+ _thread.start();
+ LOG.debug("Starting ConsumerThread for " + _partition + " done");
- }
}
-```
\ No newline at end of file
+}
+
+@Transition(to = "OFFLINE", from = "ONLINE")
+public void onBecomeOfflineFromOnline(Message message, NotificationContext context)
+ throws InterruptedException {
+ LOG.debug(_consumerId + " becomes OFFLINE from ONLINE for " + _partition);
+ if (_thread != null) {
+ LOG.debug("Stopping " + _consumerId + " for " + _partition + "...");
+ _thread.interrupt();
+ _thread.join(2000);
+ _thread = null;
+ LOG.debug("Stopping " + _consumerId + " for " + _partition + " done");
+ }
+}
+```
http://git-wip-us.apache.org/repos/asf/incubator-helix/blob/4a4510d1/site-releases/0.7.0-incubating/src/site/markdown/recipes/rsync_replicated_file_store.md
----------------------------------------------------------------------
diff --git a/site-releases/0.7.0-incubating/src/site/markdown/recipes/rsync_replicated_file_store.md b/site-releases/0.7.0-incubating/src/site/markdown/recipes/rsync_replicated_file_store.md
index f8a74a0..9e26f96 100644
--- a/site-releases/0.7.0-incubating/src/site/markdown/recipes/rsync_replicated_file_store.md
+++ b/site-releases/0.7.0-incubating/src/site/markdown/recipes/rsync_replicated_file_store.md
@@ -17,25 +17,26 @@ specific language governing permissions and limitations
under the License.
-->
-Near real time rsync replicated file system
-===========================================
+Near-Realtime Rsync Replicated File System
+------------------------------------------
-Quickdemo
----------
+### Quick Demo
* This demo starts 3 instances with id's as ```localhost_12001, localhost_12002, localhost_12003```
* Each instance stores its files under ```/tmp/<id>/filestore```
-* ``` localhost_12001 ``` is designated as the master and ``` localhost_12002 and localhost_12003``` are the slaves.
-* Files written to master are replicated to the slaves automatically. In this demo, a.txt and b.txt are written to ```/tmp/localhost_12001/filestore``` and it gets replicated to other folders.
-* When the master is stopped, ```localhost_12002``` is promoted to master.
+* ```localhost_12001``` is designated as the master, and ```localhost_12002``` and ```localhost_12003``` are the slaves
+* Files written to the master are replicated to the slaves automatically. In this demo, a.txt and b.txt are written to ```/tmp/localhost_12001/filestore``` and they get replicated to other folders.
+* When the master is stopped, ```localhost_12002``` is promoted to master.
* The other slave ```localhost_12003``` stops replicating from ```localhost_12001``` and starts replicating from new master ```localhost_12002```
* Files written to new master ```localhost_12002``` are replicated to ```localhost_12003```
-* In the end state of this quick demo, ```localhost_12002``` is the master and ```localhost_12003``` is the slave. Manually create files under ```/tmp/localhost_12002/filestore``` and see that appears in ```/tmp/localhost_12003/filestore```
-* Ignore the interrupted exceptions on the console :-).
+* In the end state of this quick demo, ```localhost_12002``` is the master and ```localhost_12003``` is the slave. Manually create files under ```/tmp/localhost_12002/filestore``` and see that appear in ```/tmp/localhost_12003/filestore```
+* Ignore the interrupted exceptions on the console :-)
```
git clone https://git-wip-us.apache.org/repos/asf/incubator-helix.git
+cd incubator-helix
+git checkout tags/helix-0.7.0-incubating
cd recipes/rsync-replicated-file-system/
mvn clean install package -DskipTests
cd target/rsync-replicated-file-system-pkg/bin
@@ -44,103 +45,99 @@ chmod +x *
```
-Overview
---------
+### Overview
-There are many applications that require storage for storing large number of relatively small data files. Examples include media stores to store small videos, images, mail attachments etc. Each of these objects is typically kilobytes, often no larger than a few megabytes. An additional distinguishing feature of these usecases is also that files are typically only added or deleted, rarely updated. When there are updates, they are rare and do not have any concurrency requirements.
+There are many applications that require storage for storing large number of relatively small data files. Examples include media stores to store small videos, images, mail attachments etc. Each of these objects is typically kilobytes, often no larger than a few megabytes. An additional distinguishing feature of these use cases is that files are typically only added or deleted, rarely updated. When there are updates, they do not have any concurrency requirements.
+
+These are much simpler requirements than what general purpose distributed file system have to satisfy; these would include concurrent access to files, random access for reads and updates, posix compliance, and others. To satisfy those requirements, general DFSs are also pretty complex that are expensive to build and maintain.
-These are much simpler requirements than what general purpose distributed file system have to satisfy including concurrent access to files, random access for reads and updates, posix compliance etc. To satisfy those requirements, general DFSs are also pretty complex that are expensive to build and maintain.
-
A different implementation of a distributed file system includes HDFS which is inspired by Google's GFS. This is one of the most widely used distributed file system that forms the main data storage platform for Hadoop. HDFS is primary aimed at processing very large data sets and distributes files across a cluster of commodity servers by splitting up files in fixed size chunks. HDFS is not particularly well suited for storing a very large number of relatively tiny files.
### File Store
It's possible to build a vastly simpler system for the class of applications that have simpler requirements as we have pointed out.
-* Large number of files but each file is relatively small.
-* Access is limited to create, delete and get entire files.
-* No updates to files that are already created (or it's feasible to delete the old file and create a new one).
-
+* Large number of files but each file is relatively small
+* Access is limited to create, delete and get entire files
+* No updates to files that are already created (or it's feasible to delete the old file and create a new one)
+
We call this system a Partitioned File Store (PFS) to distinguish it from other distributed file systems. This system needs to provide the following features:
* CRD access to large number of small files
-* Scalability: Files should be distributed across a large number of commodity servers based on the storage requirement.
-* Fault-tolerance: Each file should be replicated on multiple servers so that individual server failures do not reduce availability.
-* Elasticity: It should be possible to add capacity to the cluster easily.
-
+* Scalability: Files should be distributed across a large number of commodity servers based on the storage requirement
+* Fault-tolerance: Each file should be replicated on multiple servers so that individual server failures do not reduce availability
+* Elasticity: It should be possible to add capacity to the cluster easily
-Apache Helix is a generic cluster management framework that makes it very easy to provide the scalability, fault-tolerance and elasticity features.
-Rsync can be easily used as a replication channel between servers so that each file gets replicated on multiple servers.
-Design
-------
+Apache Helix is a generic cluster management framework that makes it very easy to provide scalability, fault-tolerance and elasticity features.
+rsync can be easily used as a replication channel between servers so that each file gets replicated on multiple servers.
-High level
+### Design
-* Partition the file system based on the file name.
-* At any time a single writer can write, we call this a master.
-* For redundancy, we need to have additional replicas called slave. Slaves can optionally serve reads.
-* Slave replicates data from the master.
-* When a master fails, slave gets promoted to master.
+#### High Level
-### Transaction log
+* Partition the file system based on the file name
+* At any time a single writer can write, we call this a master
+* For redundancy, we need to have additional replicas called slave. Slaves can optionally serve reads
+* Slave replicates data from the master
+* When a master fails, a slave gets promoted to master
-Every write on the master will result in creation/deletion of one or more files. In order to maintain timeline consistency slaves need to apply the changes in the same order.
-To facilitate this, the master logs each transaction in a file and each transaction is associated with an 64 bit id in which the 32 LSB represents a sequence number and MSB represents the generation number.
-Sequence gets incremented on every transaction and and generation is increment when a new master is elected.
+#### Transaction Log
-### Replication
+Every write on the master will result in creation/deletion of one or more files. In order to maintain timeline consistency slaves need to apply the changes in the same order
+To facilitate this, the master logs each transaction in a file and each transaction is associated with an 64 bit ID in which the 32 LSB represents a sequence number and MSB represents the generation number
+The sequence number gets incremented on every transaction and the generation is incremented when a new master is elected
-Replication is required to slave to keep up with the changes on the master. Every time the slave applies a change it checkpoints the last applied transaction id.
-During restarts, this allows the slave to pull changes from the last checkpointed id. Similar to master, the slave logs each transaction to the transaction logs but instead of generating new transaction id, it uses the same id generated by the master.
+#### Replication
+Replication is required for slaves to keep up with changes on the master. Every time the slave applies a change it checkpoints the last applied transaction ID.
+During restarts, this allows the slave to pull changes from the last checkpointed ID. Similar to master, the slave logs each transaction to the transaction logs but instead of generating new transaction ID, it uses the same ID generated by the master.
-### Fail over
-When a master fails, a new slave will be promoted to master. If the prev master node is reachable, then the new master will flush all the
-changes from previous master before taking up mastership. The new master will record the end transaction id of the current generation and then starts new generation
-with sequence starting from 1. After this the master will begin accepting writes.
+#### Failover
+When a master fails, a new slave will be promoted to master. If the previous master node is reachable, then the new master will flush all the
+changes from previous the master before taking up mastership. The new master will record the end transaction ID of the current generation and then start a new generation
+with sequence starting from 1. After this the master will begin accepting writes.
-Rsync based solution
--------------------
+### Rsync-based Solution
-This application demonstrate a file store that uses rsync as the replication mechanism. One can envision a similar system where instead of using rsync,
+This application demonstrates a file store that uses rsync as the replication mechanism. One can envision a similar system where instead of using rsync, one
can implement a custom solution to notify the slave of the changes and also provide an api to pull the change files.
-#### Concept
-* file_store_dir: Root directory for the actual data files
-* change_log_dir: The transaction logs are generated under this folder.
-* check_point_dir: The slave stores the check points ( last processed transaction) here.
+
+#### Concepts
+* file_store_dir: Root directory for the actual data files
+* change_log_dir: The transaction logs are generated under this folder
+* check_point_dir: The slave stores the check points ( last processed transaction) here
#### Master
-* File server: This component support file uploads and downloads and writes the files to ```file_store_dir```. This is not included in this application. Idea is that most applications have different ways of implementing this component and has some business logic associated with it. It is not hard to come up with such a component if needed.
-* File store watcher: This component watches the ```file_store_dir``` directory on the local file system for any changes and notifies the registered listeners of the changes.
-* Change Log Generator: This registers as a listener of File System Watcher and on each notification logs the changes into a file under ```change_log_dir```.
+* File server: This component supports file uploads and downloads and writes the files to ```file_store_dir```. This is not included in this application. The idea is that most applications have different ways of implementing this component and have some associated business logic. It is not hard to come up with such a component if needed.
+* File store watcher: This component watches the ```file_store_dir``` directory on the local file system for any changes and notifies the registered listeners of the changes
+* Change log generator: This registers as a listener of the file store watcher and on each notification logs the changes into a file under ```change_log_dir```
-####Slave
-* File server: This component on the slave will only support reads.
-* Cluster state observer: Slave observes the cluster state and is able to know who is the current master.
+#### Slave
+* File server: This component on the slave will only support reads
+* Cluster state observer: Slave observes the cluster state and is able to know who is the current master
* Replicator: This has two subcomponents
- Periodic rsync of change log: This is a background process that periodically rsyncs the ```change_log_dir``` of the master to its local directory
- Change Log Watcher: This watches the ```change_log_dir``` for changes and notifies the registered listeners of the change
- - On demand rsync invoker: This is registered as a listener to change log watcher and on every change invokes rsync to sync only the changed file.
-
+ - On demand rsync invoker: This is registered as a listener to change log watcher and on every change invokes rsync to sync only the changed file
#### Coordination
The coordination between nodes is done by Helix. Helix does the partition management and assigns the partition to multiple nodes based on the replication factor. It elects one the nodes as master and designates others as slaves.
-It provides notifications to each node in the form of state transitions ( Offline to Slave, Slave to Master). It also provides notification when there is change is cluster state.
-This allows the slave to stop replicating from current master and start replicating from new master.
+It provides notifications to each node in the form of state transitions (Offline to Slave, Slave to Master). It also provides notifications when there is change is cluster state.
+This allows the slave to stop replicating from current master and start replicating from new master.
-In this application, we have only one partition but its very easy to extend it to support multiple partitions. By partitioning the file store, one can add new nodes and Helix will automatically
+In this application, we have only one partition but its very easy to extend it to support multiple partitions. By partitioning the file store, one can add new nodes and Helix will automatically
re-distribute partitions among the nodes. To summarize, Helix provides partition management, fault tolerance and facilitates automated cluster expansion.
http://git-wip-us.apache.org/repos/asf/incubator-helix/blob/4a4510d1/site-releases/0.7.0-incubating/src/site/markdown/recipes/service_discovery.md
----------------------------------------------------------------------
diff --git a/site-releases/0.7.0-incubating/src/site/markdown/recipes/service_discovery.md b/site-releases/0.7.0-incubating/src/site/markdown/recipes/service_discovery.md
index 8e06ead..ebe44cc 100644
--- a/site-releases/0.7.0-incubating/src/site/markdown/recipes/service_discovery.md
+++ b/site-releases/0.7.0-incubating/src/site/markdown/recipes/service_discovery.md
@@ -19,95 +19,90 @@ under the License.
Service Discovery
-----------------
-One of the common usage of zookeeper is enable service discovery.
-The basic idea is that when a server starts up it advertises its configuration/metadata such as host name port etc on zookeeper.
-This allows clients to dynamically discover the servers that are currently active. One can think of this like a service registry to which a server registers when it starts and
-is automatically deregistered when it shutdowns or crashes. In many cases it serves as an alternative to vips.
+One of the common usage of ZooKeeper is to enable service discovery.
+The basic idea is that when a server starts up it advertises its configuration/metadata such as its hostname and port on ZooKeeper.
+This allows clients to dynamically discover the servers that are currently active. One can think of this like a service registry to which a server registers when it starts and
+is automatically deregistered when it shutdowns or crashes. In many cases it serves as an alternative to VIPs.
-The core idea behind this is to use zookeeper ephemeral nodes. The ephemeral nodes are created when the server registers and all its metadata is put into a znode.
-When the server shutdowns, zookeeper automatically removes this znode.
+The core idea behind this is to use ZooKeeper ephemeral nodes. The ephemeral nodes are created when the server registers and all its metadata is put into a ZNode.
+When the server shutdowns, ZooKeeper automatically removes this ZNode.
-There are two ways the clients can dynamically discover the active servers
+There are two ways the clients can dynamically discover the active servers:
-#### ZOOKEEPER WATCH
+### ZooKeeper Watch
-Clients can set a child watch under specific path on zookeeper.
-When a new service is registered/deregistered, zookeeper notifies the client via watchevent and the client can read the list of services. Even though this looks trivial,
-there are lot of things one needs to keep in mind like ensuring that you first set the watch back on zookeeper before reading data from zookeeper.
+Clients can set a child watch under specific path on ZooKeeper.
+When a new service is registered/deregistered, ZooKeeper notifies the client via a watch event and the client can read the list of services. Even though this looks trivial,
+there are lot of things one needs to keep in mind like ensuring that you first set the watch back on ZooKeeper before reading data.
-#### POLL
+### Poll
-Another approach is for the client to periodically read the zookeeper path and get the list of services.
+Another approach is for the client to periodically read the ZooKeeper path and get the list of services.
+Both approaches have pros and cons, for example setting a watch might trigger herd effect if there are large number of clients. This is problematic, especially when servers are starting up.
+But the advantage to setting watches is that clients are immediately notified of a change which is not true in case of polling.
+In some cases, having both watches and polls makes sense; watch allows one to get notifications as soon as possible while poll provides a safety net if a watch event is missed because of code bug or ZooKeeper fails to notify.
-Both approaches have pros and cons, for example setting a watch might trigger herd effect if there are large number of clients. This is worst especially when servers are starting up.
-But good thing about setting watch is that clients are immediately notified of a change which is not true in case of polling.
-In some cases, having both WATCH and POLL makes sense, WATCH allows one to get notifications as soon as possible while POLL provides a safety net if a watch event is missed because of code bug or zookeeper fails to notify.
+### Other Developer Considerations
+* What happens when the ZooKeeper session expires? All the watches and ephemeral nodes previously added or created by this server are lost. One needs to add the watches again, recreate the ephemeral nodes, and so on.
+* Due to network issues or Java GC pauses session expiry might happen again and again; this phenomenon is known as flapping. It\'s important for the server to detect this and deregister itself.
-##### Other important scenarios to take care of
-* What happens when zookeeper session expires. All the watches/ephemeral nodes previously added/created by this server are lost.
-One needs to add the watches again , recreate the ephemeral nodes etc.
-* Due to network issues or java GC pauses session expiry might happen again and again also known as flapping. Its important for the server to detect this and deregister itself.
+### Other Operational Considerations
+* What if the node is behaving badly? One might kill the server, but it will lose the ability to debug. It would be nice to have the ability to mark a server as disabled and clients know that a node is disabled and will not contact that node.
-##### Other operational things to consider
-* What if the node is behaving badly, one might kill the server but will lose the ability to debug.
-It would be nice to have the ability to mark a server as disabled and clients know that a node is disabled and will not contact that node.
-
-#### Configuration ownership
+### Configuration Ownership
-This is an important aspect that is often ignored in the initial stages of your development. In common, service discovery pattern means that servers start up with some configuration and then simply puts its configuration/metadata in zookeeper. While this works well in the beginning,
-configuration management becomes very difficult since the servers themselves are statically configured. Any change in server configuration implies restarting of the server. Ideally, it will be nice to have the ability to change configuration dynamically without having to restart a server.
+This is an important aspect that is often ignored in the initial stages of your development. Typically, the service discovery pattern means that servers start up with some configuration which it simply puts into ZooKeeper. While this works well in the beginning, configuration management becomes very difficult since the servers themselves are statically configured. Any change in server configuration implies restarting the server. Ideally, it will be nice to have the ability to change configuration dynamically without having to restart a server.
-Ideally you want a hybrid solution, a node starts with minimal configuration and gets the rest of configuration from zookeeper.
+Ideally you want a hybrid solution, a node starts with minimal configuration and gets the rest of configuration from ZooKeeper.
-h3. How to use Helix to achieve this
+### Using Helix for Service Discovery
-Even though Helix has higher level abstraction in terms of statemachine, constraints and objectives,
-service discovery is one of things that existed since we started.
-The controller uses the exact mechanism we described above to discover when new servers join the cluster.
-We create these znodes under /CLUSTERNAME/LIVEINSTANCES.
-Since at any time there is only one controller, we use ZK watch to track the liveness of a server.
+Even though Helix has a higher-level abstraction in terms of state machines, constraints and objectives, service discovery is one of things has been a prevalent use case from the start.
+The controller uses the exact mechanism we described above to discover when new servers join the cluster. We create these ZNodes under /CLUSTERNAME/LIVEINSTANCES.
+Since at any time there is only one controller, we use a ZK watch to track the liveness of a server.
-This recipe, simply demonstrate how one can re-use that part for implementing service discovery. This demonstrates multiple MODE's of service discovery
+This recipe simply demonstrates how one can re-use that part for implementing service discovery. This demonstrates multiple modes of service discovery:
* POLL: The client reads from zookeeper at regular intervals 30 seconds. Use this if you have 100's of clients
-* WATCH: The client sets up watcher and gets notified of the changes. Use this if you have 10's of clients.
-* NONE: This does neither of the above, but reads directly from zookeeper when ever needed.
+* WATCH: The client sets up watcher and gets notified of the changes. Use this if you have 10's of clients
+* NONE: This does neither of the above, but reads directly from zookeeper when ever needed
-Helix provides these additional features compared to other implementations available else where
+Helix provides these additional features compared to other implementations available elsewhere:
-* It has the concept of disabling a node which means that a badly behaving node, can be disabled using helix admin api.
-* It automatically detects if a node connects/disconnects from zookeeper repeatedly and disables the node.
-* Configuration management
- * Allows one to set configuration via admin api at various granulaties like cluster, instance, resource, partition
- * Configuration can be dynamically changed.
- * Notifies the server when configuration changes.
+* It has the concept of disabling a node which means that a badly behaving node can be disabled using the Helix admin API
+* It automatically detects if a node connects/disconnects from zookeeper repeatedly and disables the node
+* Configuration management
+ * Allows one to set configuration via the admin API at various granulaties like cluster, instance, resource, partition
+ * Configurations can be dynamically changed
+ * The server is notified when configurations change
-##### checkout and build
+### Checkout and Build
```
git clone https://git-wip-us.apache.org/repos/asf/incubator-helix.git
cd incubator-helix
+git checkout tags/helix-0.7.0-incubating
mvn clean install package -DskipTests
cd recipes/service-discovery/target/service-discovery-pkg/bin
chmod +x *
```
-##### start zookeeper
+### Start ZooKeeper
```
./start-standalone-zookeeper 2199
```
-#### Run the demo
+### Run the Demo
```
./service-discovery-demo.sh
```
-#### Output
+### Output
```
START:Service discovery demo mode:WATCH
@@ -186,6 +181,4 @@ START:Service discovery demo mode:NONE
Registering service:host.x.y.z_12000
END:Service discovery demo mode:NONE
=============================================
-
```
-
http://git-wip-us.apache.org/repos/asf/incubator-helix/blob/4a4510d1/site-releases/0.7.0-incubating/src/site/markdown/recipes/task_dag_execution.md
----------------------------------------------------------------------
diff --git a/site-releases/0.7.0-incubating/src/site/markdown/recipes/task_dag_execution.md b/site-releases/0.7.0-incubating/src/site/markdown/recipes/task_dag_execution.md
index f0474e4..15c46a3 100644
--- a/site-releases/0.7.0-incubating/src/site/markdown/recipes/task_dag_execution.md
+++ b/site-releases/0.7.0-incubating/src/site/markdown/recipes/task_dag_execution.md
@@ -17,20 +17,18 @@ specific language governing permissions and limitations
under the License.
-->
-# Distributed task execution
+Distributed Task Execution
+--------------------------
-
-This recipe is intended to demonstrate how task dependencies can be modeled using primitives provided by Helix. A given task can be run with desired parallelism and will start only when up-stream dependencies are met. The demo executes the task DAG described below using 10 workers. Although the demo starts the workers as threads, there is no requirement that all the workers need to run in the same process. In reality, these workers run on many different boxes on a cluster. When worker fails, Helix takes care of
-re-assigning a failed task partition to a new worker.
+This recipe is intended to demonstrate how task dependencies can be modeled using primitives provided by Helix. A given task can be run with the desired amount of parallelism and will start only when upstream dependencies are met. The demo executes the task DAG described below using 10 workers. Although the demo starts the workers as threads, there is no requirement that all the workers need to run in the same process. In reality, these workers run on many different boxes on a cluster. When worker fails, Helix takes care of re-assigning a failed task partition to a new worker.
Redis is used as a result store. Any other suitable implementation for TaskResultStore can be plugged in.
-### Workflow
-
+### Workflow
-#### Input
+#### Input
-10000 impression events and around 100 click events are pre-populated in task result store (redis).
+10000 impression events and around 100 click events are pre-populated in task result store (redis).
* **ImpEvent**: format: id,isFraudulent,country,gender
@@ -55,45 +53,45 @@ Redis is used as a result store. Any other suitable implementation for TaskResul
+ **report**: Reads from all aggregates generated by previous stages and prints them. Depends on: **impCountsByGender, impCountsByCountry, clickCountsByGender,clickCountsByGender**
-### Creating DAG
+### Creating a DAG
-Each stage is represented as a Node along with the upstream dependency and desired parallelism. Each stage is modelled as a resource in Helix using OnlineOffline state model. As part of Offline to Online transition, we watch the external view of upstream resources and wait for them to transition to online state. See Task.java for additional info.
+Each stage is represented as a Node along with the upstream dependency and desired parallelism. Each stage is modeled as a resource in Helix using OnlineOffline state model. As part of an Offline to Online transition, we watch the external view of upstream resources and wait for them to transition to the online state. See Task.java for additional info.
```
-
- Dag dag = new Dag();
- dag.addNode(new Node("filterImps", 10, ""));
- dag.addNode(new Node("filterClicks", 5, ""));
- dag.addNode(new Node("impClickJoin", 10, "filterImps,filterClicks"));
- dag.addNode(new Node("impCountsByGender", 10, "filterImps"));
- dag.addNode(new Node("impCountsByCountry", 10, "filterImps"));
- dag.addNode(new Node("clickCountsByGender", 5, "impClickJoin"));
- dag.addNode(new Node("clickCountsByCountry", 5, "impClickJoin"));
- dag.addNode(new Node("report",1,"impCountsByGender,impCountsByCountry,clickCountsByGender,clickCountsByCountry"));
-
-
+Dag dag = new Dag();
+dag.addNode(new Node("filterImps", 10, ""));
+dag.addNode(new Node("filterClicks", 5, ""));
+dag.addNode(new Node("impClickJoin", 10, "filterImps,filterClicks"));
+dag.addNode(new Node("impCountsByGender", 10, "filterImps"));
+dag.addNode(new Node("impCountsByCountry", 10, "filterImps"));
+dag.addNode(new Node("clickCountsByGender", 5, "impClickJoin"));
+dag.addNode(new Node("clickCountsByCountry", 5, "impClickJoin"));
+dag.addNode(new Node("report",1,"impCountsByGender,impCountsByCountry,clickCountsByGender,clickCountsByCountry"));
```
-### DEMO
+### Demo
In order to run the demo, use the following steps
See http://redis.io/topics/quickstart on how to install redis server
```
-
Start redis e.g:
./redis-server --port 6379
git clone https://git-wip-us.apache.org/repos/asf/incubator-helix.git
+cd incubator-helix
+git checkout helix-0.7.0-incubating
cd recipes/task-execution
mvn clean install package -DskipTests
cd target/task-execution-pkg/bin
chmod +x task-execution-demo.sh
-./task-execution-demo.sh 2181 localhost 6379
+./task-execution-demo.sh 2181 localhost 6379
```
+Here\'s a visual representation of the DAG.
+
```
@@ -130,7 +128,7 @@ chmod +x task-execution-demo.sh
(credit for above ascii art: http://www.asciiflow.com)
-### OUTPUT
+#### Output
```
Done populating dummy data
@@ -198,7 +196,4 @@ Impression counts per gender
{F=3325, UNKNOWN=3259, M=3296}
Click counts per gender
{F=33, UNKNOWN=32, M=35}
-
-
```
-
http://git-wip-us.apache.org/repos/asf/incubator-helix/blob/4a4510d1/site-releases/0.7.0-incubating/src/site/markdown/recipes/user_def_rebalancer.md
----------------------------------------------------------------------
diff --git a/site-releases/0.7.0-incubating/src/site/markdown/recipes/user_def_rebalancer.md b/site-releases/0.7.0-incubating/src/site/markdown/recipes/user_def_rebalancer.md
index 68fd954..3cabb2f 100644
--- a/site-releases/0.7.0-incubating/src/site/markdown/recipes/user_def_rebalancer.md
+++ b/site-releases/0.7.0-incubating/src/site/markdown/recipes/user_def_rebalancer.md
@@ -20,13 +20,13 @@ Lock Manager with a User-Defined Rebalancer
-------------------------------------------
Helix is able to compute node preferences and state assignments automatically using general-purpose algorithms. In many cases, a distributed system implementer may choose to instead define a customized approach to computing the location of replicas, the state mapping, or both in response to the addition or removal of participants. The following is an implementation of the [Distributed Lock Manager](./lock_manager.html) that includes a user-defined rebalancer.
-### Define the cluster and locks
+### Define the Cluster and Resource
The YAML file below fully defines the cluster and the locks. A lock can be in one of two states: locked and unlocked. Transitions can happen in either direction, and the locked is preferred. A resource in this example is the entire collection of locks to distribute. A partition is mapped to a lock; in this case that means there are 12 locks. These 12 locks will be distributed across 3 nodes. The constraints indicate that only one replica of a lock can be in the locked state at any given time. These locks can each only have a single holder, defined by a replica count of 1.
Notice the rebalancer section of the definition. The mode is set to USER_DEFINED and the class name refers to the plugged-in rebalancer implementation that inherits from [HelixRebalancer](http://helix.incubator.apache.org/javadocs/0.7.0-incubating/reference/org/apache/helix/controller/rebalancer/HelixRebalancer.html). This implementation is called whenever the state of the cluster changes, as is the case when participants are added or removed from the system.
-Location: incubator-helix/recipes/user-rebalanced-lock-manager/src/main/resources/lock-manager-config.yaml
+Location: `incubator-helix/recipes/user-defined-rebalancer/src/main/resources/lock-manager-config.yaml`
```
clusterName: lock-manager-custom-rebalancer # unique name for the cluster
@@ -92,16 +92,16 @@ InputStream input =
YAMLClusterSetup.YAMLClusterConfig config = setup.setupCluster(input);
```
-### Write a rebalancer
+### Write a Rebalancer
Below is a full implementation of a rebalancer that extends [HelixRebalancer](http://helix.incubator.apache.org/javadocs/0.7.0-incubating/reference/org/apache/helix/controller/rebalancer/HelixRebalancer.html). In this case, it simply throws out the previous resource assignment, computes the target node for as many partition replicas as can hold a lock in the LOCKED state (in this example, one), and assigns them the LOCKED state (which is at the head of the state preference list). Clearly a more robust implementation would likely examine the current ideal state to maintain current assignments, and the full state list to handle models more complicated than this one. However, for a simple lock holder implementation, this is sufficient.
-Location: incubator-helix/recipes/user-rebalanced-lock-manager/src/main/java/org/apache/helix/userdefinedrebalancer/LockManagerRebalancer.java
+Location: `incubator-helix/recipes/user-rebalanced-lock-manager/src/main/java/org/apache/helix/userdefinedrebalancer/LockManagerRebalancer.java`
```
@Override
public ResourceAssignment computeResourceMapping(RebalancerConfig rebalancerConfig, Cluster cluster,
ResourceCurrentState currentState) {
- // Get the rebalcancer context (a basic partitioned one)
+ // Get the rebalancer context (a basic partitioned one)
PartitionedRebalancerContext context = rebalancerConfig.getRebalancerContext(
PartitionedRebalancerContext.class);
@@ -156,10 +156,10 @@ public ResourceAssignment computeResourceMapping(RebalancerConfig rebalancerConf
}
```
-### Start up the participants
+### Start up the Participants
Here is a lock class based on the newly defined lock-unlock state model so that the participant can receive callbacks on state transitions.
-Location: incubator-helix/recipes/user-rebalanced-lock-manager/src/main/java/org/apache/helix/userdefinedrebalancer/Lock.java
+Location: `incubator-helix/recipes/user-rebalanced-lock-manager/src/main/java/org/apache/helix/userdefinedrebalancer/Lock.java`
```
public class Lock extends StateModel {
@@ -183,7 +183,7 @@ public class Lock extends StateModel {
Here is the factory to make the Lock class accessible.
-Location: incubator-helix/recipes/user-rebalanced-lock-manager/src/main/java/org/apache/helix/userdefinedrebalancer/LockFactory.java
+Location: `incubator-helix/recipes/user-rebalanced-lock-manager/src/main/java/org/apache/helix/userdefinedrebalancer/LockFactory.java`
```
public class LockFactory extends StateModelFactory<Lock> {
@@ -205,7 +205,7 @@ participantManager.getStateMachineEngine().registerStateModelFactory(stateModelN
participantManager.connect();
```
-### Start up the controller
+### Start up the Controller
```
controllerManager =
@@ -213,13 +213,14 @@ controllerManager =
HelixControllerMain.STANDALONE);
```
-### Try it out
-#### Building
+### Try It Out
+
```
git clone https://git-wip-us.apache.org/repos/asf/incubator-helix.git
cd incubator-helix
+git checkout tags/helix-0.7.0-incubating
mvn clean install package -DskipTests
-cd recipes/user-rebalanced-lock-manager/target/user-rebalanced-lock-manager-pkg/bin
+cd recipes/user-defined-rebalancer/target/user-defined-rebalancer-pkg/bin
chmod +x *
./lock-manager-demo.sh
```
@@ -227,7 +228,7 @@ chmod +x *
#### Output
```
-./lock-manager-demo
+./lock-manager-demo
STARTING localhost_12002
STARTING localhost_12001
STARTING localhost_12003
@@ -282,4 +283,4 @@ lock-group_8 localhost_12002
lock-group_9 localhost_12002
```
-Notice that the lock assignment directly follows the assignment generated by the user-defined rebalancer both initially and after a participant is removed from the system.
\ No newline at end of file
+Notice that the lock assignment directly follows the assignment generated by the user-defined rebalancer both initially and after a participant is removed from the system.
http://git-wip-us.apache.org/repos/asf/incubator-helix/blob/4a4510d1/site-releases/0.7.0-incubating/src/site/markdown/tutorial_accessors.md
----------------------------------------------------------------------
diff --git a/site-releases/0.7.0-incubating/src/site/markdown/tutorial_accessors.md b/site-releases/0.7.0-incubating/src/site/markdown/tutorial_accessors.md
index b431710..ee01cfb 100644
--- a/site-releases/0.7.0-incubating/src/site/markdown/tutorial_accessors.md
+++ b/site-releases/0.7.0-incubating/src/site/markdown/tutorial_accessors.md
@@ -21,7 +21,7 @@ under the License.
<title>Tutorial - Logical Accessors</title>
</head>
-# [Helix Tutorial](./Tutorial.html): Logical Accessors
+## [Helix Tutorial](./Tutorial.html): Logical Accessors
Helix constructs follow a logical hierarchy. A cluster contains participants, and serve logical resources. Each resource can be divided into partitions, which themselves can be replicated. Helix now supports configuring and modifying clusters programmatically in a hierarchical way using logical accessors.
@@ -122,4 +122,4 @@ Cluster cluster = clusterAccessor.readCluster();
### Atomic Accessors
-Helix also includes versions of ClusterAccessor, ParticipantAccessor, and ResourceAccessor that can complete operations atomically relative to one another. The specific semantics of the atomic operations are included in the Javadocs. These atomic classes should be used sparingly and only in cases where contention can adversely affect the correctness of a Helix-based cluster. For most deployments, this is not the case, and using these classes will cause a degradation in performance. However, the interface for all atomic accessors mirrors that of the non-atomic accessors.
\ No newline at end of file
+Helix also includes versions of ClusterAccessor, ParticipantAccessor, and ResourceAccessor that can complete operations atomically relative to one another. The specific semantics of the atomic operations are included in the Javadocs. These atomic classes should be used sparingly and only in cases where contention can adversely affect the correctness of a Helix-based cluster. For most deployments, this is not the case, and using these classes will cause a degradation in performance. However, the interface for all atomic accessors mirrors that of the non-atomic accessors.