Friday, September 18, 2009

Upgrading Your Service in Windows Azure

Windows Azure has been in CTP since PDC 08 in October of last year.  Since that time, we have had a fairly simple, yet powerful concept for how to upgrade your application.  Essentially, we have two environments: staging and production.


The difference between these two environments is only in the URI that points to any web exposed services.  In staging, we give you an opaque GUID-like URI (e.g. <guidvalue> that is hard to publically discover and in production, we give you the URI that you chose when you created the hosted service (e.g. <yourservice>

VIP Swaps, Deploys, and Upgrades

When you wanted to upgrade your service, you needed to deploy the updated service package containing all your roles into one of the environments.  Typically, this was in the staging environment.  Whenever you were ready, you would then click the big button in the middle to swap environments.  This re-programmed the load balancers and suddenly staging was production and vice versa.  If anything went wrong in your upgrade, you could hit the button again and you were back to the original deployment in seconds.  We called this model a "VIP Swap" and it is easy to understand and very powerful.

We heard from some customers that they wanted more flexibility to upgrade an individual role without redeploying the entire service.  Typically, this can be because there might be some state or caching going on in one of the other roles that a VIP swap would cause to be lost.

The good news is that now you can upgrade individual roles (or even the whole service) using the in place upgrade.  When you click the new 'Upgrade' button on the portal, you will see a screen very similar to the 'Deploy' screen that you would be used to from before, but this time you have two new options.


Upgrade Domains

The first new option allow you to choose if you want the upgrade to be 'Automatic' or 'Manual' across the upgrade domains.  To understand this option, you would probably want to understand what an 'Upgrade Domain' is all about.  You can think of upgrade domains as vertical slices of your application, crossing roles.  So, if I had a service with a single web role using 10 instances with 2 worker roles, each with 4 instances, then with 2 upgrade domains, I would have a 5 web role instance, and 2 + 2 worker roles instances in each upgrade domain.  Illustrated:


If I choose 'Automatic', it simply means that each upgrade domain will be sequentially be brought down and upgraded in turn.  If I choose 'Manual', then I need to click another button between each upgrade domain update in order to proceed.

Note:  in the CTP today, 2 upgrade domains are automatically defined and set.  In the future, you will be able to specify how many upgrade domains you would like to have.

Role Upgrades

Next, we have a radio button that specifies if you want to update the whole service, or a specific role with in the service.  Most folks will likely use the role specific update.

It is important to note that these upgrades are for the services where the topology has not changed.  That is, you cannot update the Service Definition (e.g. adding, removing roles or configuration options).  If you want to change the topology, you would need to use the more familiar VIP swap model.

Once you click Deploy, the selected role will be upgraded according to the upgrade mode you specified.

More information about in-place upgrades and update domains can be found here.  Lastly, you can of course eschew the portal and perform all of these actions using the new Service Management API.  Happy upgrading!