I’ve been monitoring my Azure bills closely of late as the service is now costing real money; in the process I’ve discovered this obvious but potentially nasty issue to be aware of.

Windows Azure Compute Hours are measured in the amount of time your Web Role (or Worker) is running.  That means if it is started, you are being billed even if you have a live site with little or no traffic.  Fine, I understand that - would be happier if it was for actual measured CPU cycles but.. whatever.  However, I was somewhat shocked to see my compute hours consumed for the month to be way higher than I had expected by a long way.  After some investigation and thought, the reason turned out to be obvious.

For each web role, a project can have two running deployments.. one for staging and one for production.  The pattern is typically to deploy your app to staging, test and then rotate live to production (and any existing production back to staging).  The problem is, if you then fail to stop the staging deployment, it will be sitting there cranking away and consuming Compute Hours.  Oops.. I had assumed I would only be billed per hour per role. 

Also, be aware that any other old test projects you have lying around in your account that happen to be started are also costing you since, again,

Lesson(s) learned.