Azure Webjobs vs Azure Functions : How to choose

There are a couple options here within App Service. I won't touch on Logic Apps or Azure Automation, which also touch this space.

Azure WebJobs

This article is honestly the best explanation, but I'll summarize here.

On Demand WebJobs aka. Scheduled WebJobs aka. Triggered WebJobs

Triggered WebJobs are WebJobs which are run once when a URL is called or when the schedule property is present in schedule.job. Scheduled WebJobs are just WebJobs which have had an Azure Scheduler Job created to call our URL on a schedule, but we also support the schedule property, as mentioned previously.

Summary:

• + Executable/Script on demand
• + Scheduled executions
• - Have to trigger via .scm endpoint
• - Scaling is manual
• - VM is always required

Continuous WebJobs (non SDK)

These jobs run forever and we will wake them up when they crash. You need to enable Always On for these to work, which means running them in Basic tier and above.

Summary:

• + Executable/Script always running
• - Requires always on - Basic tier and above
• - VM is always required

Continuous WebJobs with the WebJobs SDK

These aren't anything from a "WebJobs the feature" point of view. Essentially, we have this sweet SDK we wrote targeting WebJobs which lets you execute code based on simple triggers. I'll talk about this more later on.

Summary:

• + Executable/Script always running
• + Richer logging/dashboard
• + Triggers supported along with long running tasks
• - Requires always on - Basic tier and above
• - Scaling is manual to set up
• - Getting started can be a bit tiresome
• - VM is always required

Azure WebJobs SDK

Azure WebJobs SDK is a completely separate SDK from WebJobs the platform feature. It's designed to be run in a WebJob, but can really be run anywhere. We have customers who run them on worker roles and even on prem or other clouds, though support is only best effort.

The SDK is just about making it easy to run some code in reaction to some event and make binding to services/etc. easy. This is honestly best covered in some docs, but the heart of it is that "event" + "code" nature. We've also done some cool extensiblity work, but that's secondary to the core purpose.

Summary:

• Most of these are mentioned above
• + You can extend and run whatever you want. Full control.
• - HTTP stuff is a little wonky, but it works

Azure Functions

Azure Functions is all about taking that core purpose of the WebJobs SDK, hosting it as a service, and making it easy to get started with other languages. We also introduce the "Serverless" concept here because it made a lot of sense to do so - we know how our SDK scales, so we can do intelligent things for you.

Azure Functions is a very heavily managed experience. We aren't supporting bringing your own host. Currently, we don't support custom extensions but its something we're investigating. We're opinionated about what you can and can't do, but for the things we enable, they are slick, and easy to use and manage.

Most of the "framework" things we've done to improve Functions go through the WebJobs SDK, though. For instance, we'll be uploading a new NuGet for WebJobs which really drastically increases the speed of logging, which has huge perf benefits for WebJobs SDK users. In shipping Functions as "WebJobs SDK as a Service" we've really improved a lot of experience issues.

• + Lots of languages supported
• + Fully managed, dynamic scaling
• + Easy to use portal w/ UX for managing connections/etc.
• - Host not customizable (yet)
• ~ Runs in a separate "app" which requires some configuration in your repo, but makes long term maintenance much easier.
• ~ No tooling (yet) Some tooling is now in alpha or preview - https://www.npmjs.com/package/azurefunctions (update Feb 2017: Visual Studio Tools for Azure Functions now available in preview: https://blogs.msdn.microsoft.com/webdev/2016/12/01/visual-studio-tools-for-azure-functions/)

I'm probably biased since Functions is our latest and greatest, but feel free to shoot more cons for Functions my way.

I'll probably end up publishing a blog which elaborates a bit more, but I tried to keep this as succinct as possible for this forum.

Being Azure Functions based on the WebJobs SDK, they provide most of the functionality already available in WebJobs, but with some new cool capabilities.

In terms of triggers, in addition to those already available for WebJobs (e.g. Service Bus, Storage Queues, Storage Blobs, CRON schedules, WebHooks, EventHub, and File Cloud Storage providers), Azure Functions can be triggered as APIs. And HTTP calls don't require kudu credentials, but can be authenticated via Azure AD and third-party identity providers.

In regard to outputs, the only difference is that Functions can return a response when called via HTTP.

Both support a wide variety of languages, including: bash (.sh), batch (.bat / .cmd), C#, F#, Node.Js, PHP, PowerShell, and Python.

Being Functions currently in Preview, tooling is still not ideal. But Microsoft is working on it. Hopefully we get the same flexibility of developing and testing Functions locally as we currently do for WebJobs with Visual Studio.

The most significant and cool advantages brought by Functions is the alternative of having a Dynamic Service Plan with a "Serverless" model, in which we don't need to manage VM instances or scaling; it's all managed for us. Additionally, by not having dedicated instances, we only pay for the resources we actually use.

A more detailed comparison between the two here: https://blog.kloud.com.au/2016/09/14/azure-functions-or-webjobs/

HTH :)

I would like to add two more point to the above long & little bit old posts. if you choose consumption plan in azure functions, below are the limitations

If you want to run any jobs more than 10 minutes choose webjobs. Azure functions, runs only for 5 minutes by default, if your process exceeds 5 minutes, then azure function throws timeout exception. You can increase the timeout to 10 minutes in host.json.

Note: There is no timeout problem if you are using app service plan azure functions.

Another reason to distinguish is. if you use azure function, then your initial start time will be slow because the machines(containers) are created on the fly and destroyed once it is used.

To avoid cold start, azure function app have released premium plan, where one instance will be running all the time and based on the load the function app will start scaling and you will be billed for one instance and other instances based on consumption.