Continuous deployment over a cluster of servers

Question

I am currently working with a startup that is in a transitional phase.

We have a PHP web application and utilise continuous integration with the standard unit and regression tests (selenium) run over jenkins. We have a development server which hosts newly committed code and a staging server that holds the build ready for deployment to the production server. The way we deploy to the production server is through a rudimentary script that pulls the latest svn copy and overwrites the changes in the htdocs directory. Any SQL changes are applied via the sync feature from MySQL Workbench.

This setup works fine for a very basic environment but we are now in a transition from single server setups to clusters due to high traffic and I have come up against a conundrum.

• My main concern is how exactly do we switch deployment from a single server to a cluster of servers ? Each server will have its own htdocs and SQL database and under the current setup I would need to execute the script on every server which sounds like an abhorrent thing to do. I was looking into puppet which can be used to automate sysadmin tasks but I am not sure whether it is a formidable approach for deploying new builds to a cluster.

• My second problem is to do with the database. Now my assumption is the code changes will be applied immediately, but since we will have a db master/slave replication my concern is the database changes will take longer to propagate and thus introduce inconsistencies during deployment. How can the code AND database be synchronised at the same time ?

• My third problem is related to automation of database changes. Does anyone know of any way I can automate the process of updating a DB schema without manually having to run the synchronisation ? At the moment I have to manually run the workbench sync tool, whereas I am really looking for a commit and forget approach. I commit it and DB changes are auto synchronised across the dev and QA setups.

Answer 1

I am running a similar scenario, but I am using a Cloud Provider for my production environment, in order that I do not need to care about replication of DB, multi server instances etc. (I am Using pagodabox, but AWS would also work perfectly fine).

I would recommend you to create real migrations for Database Migrations, in order to track those via svn or something else. In that case, you can also provide information, how to roll back. I am using https://github.com/doctrine/migrations, but mainly because I use doctrine as ORM.

If you have a migration tool, you can easily add a command in your deployment script to run those migrations after deployment.

I don't think that the database synchronisation is a big issue during deployment. That might depend on the actual infrastructure youre using. The cloud providers like pagoda or aws take care of it for you.