Introduction to ActiveMQ Apollo

Introduc6on to Ac6veMQ Apollo

Bosanac Dejan
May 2011

A Progress So3ware Company
Copyright © 2011 Progress So3ware Corpora6on and/or its subsidiaries or aﬃliates. All rights reserved. FuseSource Conﬁden6al A Progress So3ware Company
1

About me

  Bosanac Dejan
  Senior So3ware Engineer at FUSESource ‐ hNp://fusesource.com
  Apache Ac6veMQ commiNer and PMC member
  Co‐author of Ac6veMQ in Ac6on

2 Copyright © 2011  Progress So3ware Corpora6on and/or its subsidiaries or aﬃliates. All rights reserved.

What we are going to cover?

  Why Apollo?
  HawtDispatch
  Connec6vity
•  Stomp 1.1, MQTT, JMS, ...
  LevelDB Store
  Features
•  REST Based Management
  Future


Why Apollo?

4

Apache Apollo

Goal
  An experiment to beNer u6lize high core counts on modern
processors

Resulted
  A completely new broker core that is much more determinis6c,
stable, and scaleable
  Branched as a new project and a chance for a clean start


Apollo Architecture

  Reactor Based Thread Model
  Scala implementa6on
  Protocol Agnos6c
  REST Based Management


HawtDispatch

7

HawtDispatch ‐ Introduc6on

  Small (less than 100k) thread pooling and NIO event no6ﬁca6on
framework API
  hNp://hawtdispatch.fusesource.org/
  Java clone of Grand Central Dispatch
  Avoid explicit usage of threads and synchroniza6on points in
mul6threaded applica6ons
  Applica6on developer submit tasks to dispatch queues
  Fixed‐sized thread pool execute tasks


HawtDispatch ‐ Dispatch Queues

  Global Dispatch Queue
•  3 global queues shared
•  Non determinis6c order
•  3 priori6es

DispatchQueue queue = getGlobalQueue(HIGH);

  Serial Dispatch Queue
•  Serial FIFO queues
•  Synchronize certain task execu6ons

DispatchQueue queue = createQueue("My queue");


HawtDispatch ‐ Submijng Tasks

  Java queue.execute(new Runnable(){
public void run() {
System.out.println("Hi!");
}
});
  Scala
queue {
}
// or
queue.execute(^{
  Tasks })
•  non‐blocking
•  lock‐free
•  wait‐free


HawtDispatch ‐ Dispatch Sources

  Trigger a task execu6on based on external event
  Ideal for dealing with NIO events

SelectableChannel channel = ...
DispatchQueue queue = createQueue()
DispatchSource source = createSource(channel, OP_READ, queue);
source.setEventHandler(new Runnable(){
public void run() {
ByteBuffer buffer = ByteBuffer.allocate(1024);
int count;
while( (c=channel.read(buffer)) > 0 ) {
// just dump it to the console
System.out.write(buffer.array(), buffer.offset(), buffer.position());
}
}
});
source.resume();


HawtDispatch ‐ conclusion

  Ideal for developing highly concurrent applica6ons
  Ideal for handling NIO events
  Ideal for implemen6ng broker cores
  Impressive performances

  New development paradigm means it couldn t be ﬁNed into exis6ng
Ac6veMQ broker


Connec6vity

13

Message protocols ‐ Overview

  Broker Core is Protocol Agnos6c
  Protocols are Plugins
•  STOMP 1.0/1.1
•  MQTT v3.1
•  Openwire
  Transports are Pluggable too.
•  TCP, WebSockets etc.
  Protocol detec6on (all protocols can use 1 port)


Feature: Mul6ple Transports

  TCP
  SSL
  WebSockets
  Secure WebSockets
  UDP


Stomp ‐ basics

  hNp://stomp.github.com/
  Simple Text Orientated Messaging Protocol
  HTTP for the messaging realm
  Very simple, so it s easy to write clients and servers in prac6cally
any language
  A lot of exis6ng clients in C, Ruby, Pyhton, JS, PHP, etc.
  Provides a way to connect diﬀerent plamorms in asynchronous way
  Also Implemented by Ac6veMQ, RabbitMQ, HornetQ, and many
others.


Stomp ‐ Nutshell

  Based on text frames, similar to
HTTP ones MESSAGE
subscription:0
  Can transport binary bodies message-id:007
  Frame command for every destination:/queue/a
content-type:text/plain
messaging concept, like CONNECT,
MESSAGE, SUBSCRIBE, ACK, etc. hello queue a^@
  New in version 1.1
•  Protocol nego6a6on
•  Heartbeats
•  NACK
•  Virtual hosts


Apollo and Stomp

  Both 1.0 and 1.1 Supported
SEND
  Des6na6on Types destination:/queue/a
•  Queues ‐ /queue/a receipt:001
•  Topics ‐ /topic/b persistent: true
•  Durable Subscrip6ons ‐ /dsub/c
hello queue a
^@
  Reliable messaging
RECEIPT
receipt-id:001

^@


Apollo and Stomp

  More messaging features
•  Message expira6on
•  Topic retained messages
•  Topic durable subscrip6ons
•  Queue browsing
•  Queue message sequences
•  Exclusive subscrip6ons
•  Temporary des6na6ons
•  Des6na6on wildcards
•  Composite des6na6ons
•  Message selectors


Connec6vity ‐ MQTT

  Get at hNps://github.com/fusesource/fuse‐extra/
  Focused on:
•  low bandwidth networks
•  unreliable networks
•  Small footprint / Embedded Devices
  3 QoS Op6ons
  Also Implemented by
WebsphereMQ,
MosquiNo and
others.


Connec6vity – JMS API

  StompJMS is a JMS 1.1 client implemented using the STOMP
protocol.
  Get it at hNps://github.com/fusesource/stompjms
  Client implemented with HawtDispatch
•  Constant number of threads no maNer how many client connec6ons are
established.

import org.fusesource.stomp.jms.*;import javax.jms.*;

StompJmsConnectionFactory factory = new
StompJmsConnectionFactory();factory.setBrokerURI("tcp://localhost:
61613 );
Connection connection = factory.createConnection( admin , password );

Destination example = new StompJmsDestination( /queue/example );


Connec6vity – OpenWire

  OpenWire is the na6ve binary protocol implemented by Ac6veMQ
  API op6ons:
•  JMS 1.1 Client of Ac6veMQ 5.x
•  NMS Client for C# Apps
•  CMS Client for C++ Apps
  Working Features
•  Queues, Topics, Durable Subscrip6ons
•  Temporary Des6na6ons
•  Transac6ons
  Not Yet Supported
•  XA Transac6ons (distributed transac6ons)
•  Network of Brokers style clustering
•  0 sized consumer prefetches


LevelDB Store

23

Message Store ‐ overview

  Message Stores are Plugins
  Ships with 2 Op6ons
•  LevelDB Store
•  BDB Store
  Used to store
•  persistent messages
•  non‐persistent messages that needs to be swapped out of memory
  Non‐persistent messages that get swapped out do not get dropped
on restart
  Delayed Writes


LevelDB ‐ Basics

  What is LevelDB
•  LevelDB is a fast key‐value storage library
•  WriNen at Google
•  Provides an ordered mapping from string keys to string values
•  Based on SSTable and Log Structured Merge (LSM) Trees


LevelDB ‐ Basics

  Designed to eﬃciently store large numbers of key‐value pairs while
op6mizing for high throughput, sequen6al read/write workloads
  Ideal for implemen6ng message store index
  Much beNer performance over tradi6onal B‐Tree indexes (used in
KahaDB)


LevelDB ‐ Basics

  It s a C++ library (no client‐server support)
  Batching writes
  Forward and backward itera6on over data
  Uses Snappy compression


LevelDB Store

  Uses LevelDB to maintain indexes into log ﬁles holding the
messages
  Uses a JNI driver on Linux and OS X, but falls back to a pure Java
version on other plamorms
  Supports replica6on to get High Availability
  Default store in Apollo. Available in Apache Ac6veMQ 5.6.0


LevelDB vs KahaDB

  It maintains fewer index entries per message than KahaDB which
means it has a higher persistent throughput.
  Faster recovery when a broker restarts
  LevelDB based index provide a much beNer performance than the
B‐Tree for sequen6al access
  LevelDB indexes support concurrent read access
  Pauseless data log ﬁle garbage collec6on cycles


LevelDB vs KahaDB

  Uses fewer read IO opera6ons to load stored messages.
  It will only journal the payload of the message once
  Exposes it's status via JMX for monitoring
  Supports replica6on to get High Availability


LevelDB store ‐ HA

  HA version of store works with Hadoop based file systems
  Message log is mirrored to HDFS
  It can sync on HDFS file instead of local file system
  LevelDB indexes are immutable on disk (SSTables)
  On checkpoint .sst files are uploaded to HDFS


LevelDB store ‐ HA Recovery

  On master failure, slave will download
•  message log
•  .sst ﬁles associated with the latest uploaded index
  Con6nue with regular recovery process


LevelDB store – HA locking

  Master elec6on is done externally
  Mul6ple brokers should never use the same HDFS path
  Apache ZooKeeper good op6on for implemen6ng distributed
locking
  FuseFabric (hNp://fuse.fusesource.org/fabric/) can be used as well


BDB store

 Not ASL 2.0! You have to Agree to the BDB
license & download from Oracle.
 Pure Java implementa6on
 Very robust
 The BDB library supports advanced features like
replica6on (not yet exploited)


Features

35

Feature: JAAS Authen6ca6on

 Use any 3rd party JAAS login module
 Ships with security enabled by default
• Default id/password is admin/password
 File based user and group conﬁgura6on
 Supports IP address white and black lists.
 X509 Cer6ﬁcates
 Op6onal guest login support


Feature: Authoriza6on rules

 Fine grained control of who can
• admin, monitor, conﬁgure, connect, create, destroy,
• send, receive, consume
 On broker resources like:
• broker, connector, virtual host, topic, queue or
durable subscrip6ons
<access_rule allow="bartenders" action="send,consume
kind= queue topic id= BAR.* />
<access_rule deny="guests" action="consume"/>


Feature: Config Updates

 All Configura6on files are watch and changes are applied at
run6me:
• Broker config: etc/apollo.xml
• JAAS config files like: etc/user.proper6es
• Logging config: etc/log4j.proper6es
 No need to restart to apply config changes


Feature: REST Management

curl -u "admin:password"  
http://localhost:61680/broker/virtual-hosts/default.json


Where is it going?

41

Where has it been?


Feature diﬀ vs Ac6veMQ

Missing in Apollo Only in Apollo
 Networks of Brokers  REST Management API
 Priority Support  Secure WebSockets
 Message Groups  Message Sequences
 Message Scheduling  Con6nuous Queue
 XA Transac6ons Browsing
 JMX Management API  Run6me Conﬁg Updates
 Per Consumer Store
Prefetch




Future

  Start integra6ng it in Ac6veMQ
•  Apollo should be a new broker engine for 6.0
•  We should try to port all exis6ng features (networks, priority queues, XA, etc.)
  Tons of interes6ng work ahead of us ‐ Join us


Ques6ons?

  Links:
•  Apache Apollo hNp://ac6vemq.apache.org/apollo/
•  STOMP Benchmarks hNp://hiramchirino.com/stomp‐benchmark/
•  MQTT Protocol Plugin for Apollo hNps://github.com/fusesource/fuse‐extra
•  HawtDispatch hNp://hawtdispatch.fusesource.org/
•  StompJMS hNps://github.com/fusesource/stompjms

  Blog:  hNp://www.nighNale.net

  TwiNer:
•  hNp://twiNer.com/dejanb
•  hNp://twiNer.com/fusenews


Introduction to ActiveMQ Apollo

More Related Content

What's hot (20)

Similar to Introduction to ActiveMQ Apollo (20)

More from dejanb (7)

Recently uploaded (20)

Introduction to ActiveMQ Apollo