On Rabbits and Elephants

Gavin M. Roy @crad - pgConf.eu 2011

Data Distribution with pg_amqp and RabbitMQ

On Rabbits and Elephants



myYearbook.com• #1 Comscore Teen Destination, Top Trafficked Sites in

US who is hiring

• Use of PostgreSQL predates my joining as CTO in 2007 and we’re hiring

• Use of RabbitMQ goes back to 2009, we were hiring then too

• Awesome place to work, have been known to sponsor Europeans for work in the US and we’re hiring


Don't try this at home.(try it at work or at least on your laptop)


Audience Participation

• You’ll need PostgreSQL 9.1rc1 with pgbench installed

• Python 2.6 or Python 2.7

• Tarball from http://192.168.103.135/~gmr/

• Untar and then ./setup-client.py


Use Cases• Adding complexity to your environment

• Ensuring data inconsistency between canonical data sources and data warehouses

• Impressing your hipster Ruby programmer friends

• Replace your Slony replication setup


Use Cases• Write distributed map-reduce functions in

Erlang


What is PostgreSQL?(just kidding)


What is RabbitMQ?(don’t let the cloud based marketing deter you)


About AMPQ• Platform, Vendor Neutral Messaging format

• 0.9.1 finalized in November 2008

• JP Morgan Chase, Red Hat, Rabbit Technologies, iMatrix, IONA Technologies, Cisco Systems, Envoy Technologies, Twist Process Innovations

• Members now include Bank of America, Barclays Banks, Microsoft Corporation, Novell, VMWare

• 1.0 Final: October 2011 “Changes some stuff”


Going to only scratch the surface

RabbitMQ & AMQPConcepts


Management Plugin


Publishers


Message Broker


Consumers


Message RoutingIt’s not just about Queues.


Routing KeyHelps Determine Message Destination

(in most cases)


Queues• Durability: Queue survives reboot

• Auto-Delete: Delete the queue when consumer goes away

• Exclusive: Only one consumer may consume

• x-expires: Auto-Delete after given duration

• x-message-ttl: Auto-discard a message after ttl


Declaring a Queue


Exchanges(The ones we care about for this talk)


Direct exchanges

• 1:1 Message Delivery Pattern

• Routing Keys are direct, no wildcarding

• All queues bound to a routing key get the message


Topic exchanges• Pattern Matching in Routing Keys

• Publish one message type to public.pgbench_accounts another public.pgbench_history

• Period delimiter, * stays within the period scope, # does not

• Example patterns: *.pgbench_accounts, public.*


Other Exchange Types• Built-In: Fanout, Headers

• 3rd Party or Plugin Types:

• Consistent Hash Exchange: Round-robin distribution to queues

• External: RPC Exchange Plugin

• Riak

• Last-Value Cache

• Recent History Exchange: Keeps last 20 messages routed

• Global Fanout: Sends to all queues, regardless of routing key


One More thing...Exchange-to-Exchange Bindings


Messages• Have client specified properties:

• Content type, Encoding, Timestamp, App-Id, User-Id, Headers

• Delivery Mode:

• 1: Non-Persistent

• 2: Persistent


Step-By-Step Instructions1. Install your Erlang on the machine you into intend to run your RabbitMQ broker on. My preference is for Erlang R14b2, which almost makes me think they name Erlang releases after Star Wars characters, but they

don't. Erlang is not as fun as Debian is. It doesn't need to be a beefy machine but it should have enough ram that if you don't have an active consumer it won't run out of memory.

2. Install RabbitMQ. This is a multistep process involving black magic. Depending on your distribution you can either download a package, or if your *nix impaired there is a windows installer too. Oh and don't forget to install the management plugin. You can download the management plugin from the RqbbitMQ site pretty easy. After you have downloaded it you put it in the RabbitMQ plugins directory. Kind of makes sense doesn't it?

3. Go to the store and buy RabbitMQ in Action and RabbitMQ in Detail. What do you mean you can't? No pre-order either? Well there are plenty of good blogs on the intarwebs that have good information such as *cough* mine. Seriously though if you are crazy enough to follow my advice and get this far out of actual interest in the subject, make sure you verse yourself in the operation of and the flexibility that is RabbitMQ.

4. Install one of the following: pl/Python, pl/Perl, pl/Ruby, pl/PHP, pl/PGSQL and pg_amqp, or write your own damn addon in c using librabbitmq and then distribute it using the very cool pgxn service. Buoy only need this on the databases your going to be writing triggers on. Are you still reading this? Seriously? I would have thought I had lost you by now. Either that or you are skipping ahead because I should have changed the slide by now. It could be that Selena is operating the slides, in which case I have no control over these slides which is a terrifying thought. You see there is very little content on the slides, a real lack of substance. If I stay on any one slide for too long then you'll realize I really don't know what I am talking about and then I'd not be asked back to give one of the most important talks of the conference.

5. It's almost time to write your publisher, but don't get too far ahead of yourself. You first need to make sure that you have figured out a routing paradigm to use. Don't you like that word? Paradigm. I hear it in my head as para dig'um. Anyway, back to routing paradigms. Depending on your use case you will have the ability to broadcast your events to a whole slew of databases that care about what you have to send, or just one. You can implement a 1-to-1 type of scenario where a "direct" exchange is setup and your data will be queued in a way that no message is duplicated to your client and there is transactional guarantees in RabbitMQ. You probably don't want that thou, as you're a fly -by-the-seat-of-your-pants type of person, aren't you? For that you'll want to turn on noAck mode in your consumers. Rabbit will just fling your messages to your consumers as fast as possible, just like an ape at a zoo and his favorite projectile.

6. This is where things get fun,. We are going to quite. Stored procedure in the language of your choice and have it publish to RabbitMQ using an exchange and a routing key, now at this point if you are still reading this, I have to question your sanity. This was a one off gag slide to make people at the conference think that I wax going to read all of this. Anyway, wirte your function that publishes and you are almost done. No seriously. Well not quite done because you still need the part that calls this function and then the consumer app that reads the data from RabbitMQ and then doe the actions in PostgreSQL that you want it to do.

7. Now, basically you're going to add after insert, update, delete triggers to the tables you care about. I'm. Not going to go into detail about how to do that since you are at a PostgreSQL. Conference and should know how to do that already. Instead I am going to assume you know what I am writing about and say that in these triggers you will want to call the function we wrote in step 6 from these trigger events,

8. Go to http://github.com/gmr/On-Rabbits-and-Elephants/ and download the sample code and use that instead of all of this nonsense.

http://github.com/gmr/On-Rabbits-and-Elephants/

http://github.com/gmr/On-Rabbits-and-Elephants/


pg_amqp• Is a publisher only PostgreSQL extension

• Transactional if you use amqp.publish

• Not if you use amqp.autonomous_publish

• Needs some <3 for more advanced features like delivery mode changing and message properties.


Use a Trigger(benefit from the value of a trigger?)


Not that kind of trigger.


Maybe this kind.


Really this kind.(pretty boring, I know)


Stop. Demo Time.


Demo• Using pgbench

• PostgreSQL 9.1rc1

• pg_amqp and plpythonu

• RabbitMQ 2.6.1

• Python 2.7 (Should work in 2.5 or 2.6 too)


Generic Trigger

• Publishes to schema.table_name

• Wraps the entire trigger data set into a JSON payload


consumer.pyDynamically builds SQL based upon event type and

executes on a remote PostgreSQL database

./consumer.py 192.168.103.135


Proof!./watch-messages.py 192.168.103.135

./watch-db-client.py


Interactive Demo


Resources• rabbitmq:

http://www.rabbitmq.com

• pg_amqp:easy_install pgxnclientUSE_PGXS=1 pgxn install pg_amqphttp://pgxn.org/dist/pg_amqp/

• example code:https://github.com/gmr/On-Rabbits-and-ElephantspgConf.eu Branch

• follow me on twitter: @crad

Please give feedback:

http://2011.pgconf.eu/feedback



http://pgxn.org/dist/pg_amqp/

http://pgxn.org/dist/pg_amqp/

https://github.com/gmr/On-Rabbits-and-Elephants

https://github.com/gmr/On-Rabbits-and-Elephants

On Rabbits and Elephants

Documents

Transcript of On Rabbits and Elephants