Publishing WebJobs with Azure Pipelines

Azure WebJobs are a great way to execute background process.

Scenario

In last days I’ve created a Azure Pipelines that publish WebJobs, but in our specific case our WebJobs were not being published to /site/wwwroot/App_Data/jobs/Continuous/<webjobs name>, because “reasons” our root web app is using ‘/site/www’ as folder. So, you should expect that the new folder to deploy WebJobs is /site/www/App_Data/jobs/Continuous/<webjobs name>, but NO, the new folder is /site/jobs/Continuous/<webjobs name>.

The big problem with the above behaviour is that WebDeploy via VS or AzureRmWebAppDeployment task via Azure Pipeline will still try to deploy the webjobs to the folder inside /site/www/

My best suspects are these lines on AzureRmWebAppDeployment@3

The AzureRmWebAppDeployment just check the physical path configured on Azure when we defined a VirtualApplicatoin parameter, but in our case it is not a virtual application, but it is a different folder than /site/wwwroot/.

in Kudu wiki there is some explanation about the webjobs folders.

Solution

I used the FtpUpload task to publish the WebJobs binaries to the /site/jobs/Continuous/<webjobs name> and two instances of AzureAppServiceManage task to stop and start the jobs:

you’ll need to replace the values between <..> in the variables section to specific values of your project

I omitted some tasks from the original azure-pipelines.yml for simplicity. The original has a lot of others taks, as unit tests and web api publishing

Extra points (graceful shutdown)

Now that you’ve automate your WebJobs publishing, is a good idea improve your jobs to check if Azure has requested a shutdown. You can do this taks through the CancellationToken parameter passed to the job.

A function can accept a CancellationToken parameter, which enables the operating system to notify your code when the function is about to be terminated. You can use this notification to make sure the function doesn’t terminate unexpectedly in a way that leaves data in an inconsistent state.

Further reading

Follow @giacomelli
Tutorials

News

Articles

Labs