.Net Foundation On .NET, 21 Dec 2017

Previous Meeting

⏯

youtube image

►

From YouTube: Chris Gillum - Serverless workflows with Durable Functions

Description

In this episode of On .NET, Chris Gillum (@cgillum) from the Azure Functions team stops by to talk to us about Durable Functions. With Durable Functions, you can create stateful workflows in a serverless environment.
00:31 - What does Durable mean?
01:31 - How long has the service been around?
01:55 - What are some of the use cases?
04:19 - Chris shows some demos of Durable Functions
06:00 - What do we really get from using this?
08:30 - What are some of the Orchestration models this enables?
12:58 - What does the API surface look like?
18:51 - What flavors are .NET does Durable Functions support?

Remember to check out the documentation and GitHub repos listed below.
•Durable Functions documentation - https://docs.microsoft.com/en-us/azure/azure-functions/durable-functions-overview
•Azure Functions documentation - https://docs.microsoft.com/en-us/azure/azure-functions/
•Github Repo - https://github.com/Azure/azure-functions-durable-extension
•Durable task framework - https://github.com/Azure/durabletask
Create a Free Account (Azure): https://aka.ms/azft-net

A

Hi,, this is another episode of the on.net show.

A

Today, we're gonna be talking with chris gillum about azure functions.. How about you quickly introduce yourself?.

B

Hi,, I'm chris gillum,, I'm an engineer on the azure functions.

B

Team and I work on a lot of areas.

B

right now. My main focus is on durable functions,.

B

Which is a new extension to azure functions that allows you.

B

To write basically workflows or orchestrations, using c# and.

B

A

Okay,, so what does this term, durable, mean?.

B

Yeah, so durable basically means that, suppose that I'm.

B

Writing a function and didn't use to do multiple things.

B

Maybe you have a process, that's broken up into multiple.

B

Functions. you receive a document.. Maybe you need to do some processing on that document,.

B

Then you need to maybe upload it somewhere. Else.

B

then maybe send an email, some kind of an orchestration.

B

Durable means that for each of those steps, we're going to.

B

Basically, checkpoint your state automatically., so that if the vm goes down or somehow the process gets.

B

Interrupted we're able to kind of bring it back up and.

B

Resume from where we left off., I see,- I don't know if you're familiar with this term,, but.

A

There's this computing concept called coroutines.

C

This sounds a little bit like cloud coroutines to me.

A

Maybe that'll be helpful for some of our listeners.

A

Okay,, so how long has this service existed in azure?.

B

Yeah,, so azure functions has been around for a while., durable functions,. This extension, particularly,.

B

We just launched this in preview in october.

A

Okay, so it's pretty new?.

B

It's fairly new, yeah,, some people may have seen we had like.

B

A kind of a public preview, alpha release.

B

A

Yeah,, so how about we talk a little bit about use, cases,.

A

Why people would use this. and also since it's pretty new,, where you are seeing kind.

A

Of the first early adopters with durable, functions.

B

Yeah, so there's several use cases that I like to talk. About.

B

In fact,, I even have a couple of drawings that I may share. If.

B

That's okay, as well.

A

Yeah,, let's do some drawings.

B

In fact,, or even just kind of powerpoint slides., I'm gonna go ahead and skip over to a previous one. Here.

B

But one of them is,, like I described before,, let's say that you like to use functions.

B

and you've got a bunch of several functions which.

B

Implement some step in a sequential process, and.

B

You wanna chain those together,, like I mentioned.

B

You get a document,, maybe do some processing on it,.

B

Upload something somewhere., so that's one example of how you might use durable, functions,.

B

Is to kind of implement that., so here in this picture, I've got f1 calling f2 calling f3, so on, and so forth.

A

Right, in this particular picture though,, if this.

A

Was all fire and forget style, computation,.

C

Then f1 wouldn't need to be durable, right?.

A

Isn't it only the case if f1 was kind of, if each one of these was waiting for the caller return, and then it kind of came back, all the way back to f1?.

A

That's the case where durable functions would be important,.

A

Is that right?.

B

Well,, even if you're just going kind of in one direction- fire and forget, yeah., this is still useful for the durability, for example,.

B

If you're on f3 and then the node just dies.,.

B

I see. you wanna, be able to come. Back, continue where you left off with f3 and then get to step. F4 there.

B

I see, so durable functions. Is both this kinda idea. That.

A

You can kind of be idle,, that's probably the slightly the wrong term,, while you're waiting for your functions that you're calling.

A

But it also enables you to have some reliability characteristics.

A

That would have been more difficult. Before, is that fair?

A

sure, yeah, absolutely.

B

It might even be helpful if we jump in and just look at some code instead of pictures., so that picture that you saw, basically,.

B

I have implemented here as a simple function., so durable functions, again,. So this is just azure functions,. So.

B

If you're already familiar with that,, this won't be too surprising..

B

This is our visual studio experience..

B

So a durable function has what's called an orchestration trigger.

B

Basically, we have a nuget packet, which you can get.

B

To get started with durable functions., if we were to look at my csproj here, you can see microsoft.azure.webjobs.extensi-.

B

Ons.Durabletask,, that's what we call it. named after the durable task framework, which we depend. On.

B

And basically, there are a few methods that you can call on. That.

B

callactivityasync basically is going to take a look for.

B

Some other function within your project.

B

and it's gonna call that function and pass in whatever input that you tell it. To.

B

So in this case, sayhello, as an example, and.

B

Here's the sayhello function,, which is conveniently here., I see,, so there's kind of a registration story.

A

Of sorts at play here so callactivityasync,, I guess,.

A

By default, looks for functions in your same project.

A

Is it possible to find functions that are not in your project?.

B

Not today, right now,, the experience we're focusing on is the ones within your project.

B

Because then we can easily do error, handling and say, hey,.

C

You have a typo or something that's not in there.

A

Okay, makes sense.

B

So if we were wanting to share function, logic, at the moment,.

A

We'd have to be sharing source, as opposed to kind of sharing.

A

Live functions? yeah, source or you know, dll's, since they just get.

A

Compiled under dlls, but yeah, effectively, yeah.

B

So and just to be clear,, we talked about durability.

B

Somebody who's just looking at this code might be asking,.

B

Why am I calling this sayhello by name using this api here?.

B

Why don't I just make a regular method call to it??

B

I mean, it's right, here,, it's in my project.. What are you giving me with this thing?.

B

And to answer that is we're giving you that durability, right?, when you call callactivityasync, under the hood,.

B

We're going into azure storage and making a record that, okay, somebody scheduled this sayhello call.

B

And we invoke it,, we get a response,, we save that response.

B

And so now, when this orchestrated function, resumes,.

B

We've recorded that, all right,, we've called sayhello and.

B

Passed in the first parameter, tokyo,, or something like that.

B

And so if this process were to go down and come back, up,.

B

We can replay this orchestrator function from the beginning.

B

We callactivityasync again as part of this replay.

B

We know that,, we've already said hello to tokyo,. Let's skip that.

B

In fact, we know what the return value was.- let's just grab that and then move on to the next step, immediately..

B

So that's kind of where that durability comes in.

B

And one other advantage you get over this model is that.

B

Scale out. so let's say you've deployed this to azure..

B

We support durable functions, natively in our.

B

Azure consumption model., when you call sayhello or any function that you wanna call,.

B

That could be run on the current machine. That's running our.

B

Orchestrator. or it could be called on some other machine elsewhere in the data center., and so, if maybe you're, doing a whole bunch of calls in parallel,. Those could be fanned out on a whole bunch of different machines., and so that's something you wouldn't get if you were just making a regular method call into sayhello.

A

Right, so effectively, what I'm hearing is,, you can kind of load balance, and it sounds like the system. Does this for.

A

You on a function by function; basis., you don't kind of have to load balance across your whole.

A

Project, right?.

B

A

That makes sense.

A

and we also get reliability across these two different.

A

Things, across these two here. yep, makes sense.

A

So I guess,, we talked a little bit before about orchestration.

A

Are there different orchestration models that this.

A

System allows?.

B

Absolutely,, so this is the basic one, which is. I just want.

B

To do something in sequence.

C

Very easy,, maybe not even that interesting.

B

Where it gets more, interesting is in this other sample that I have,, which is about fan out and fan back in.

B

So I can show a picture real, quick before I jump into the code..

B

Let's imagine: we've got this picture, you've got some function which basically needs to fan out and.

B

Do a whole bunch of work in parallel and we'll call that f2.

B

and then all those things are happening in parallel.

B

And then, maybe at the very end of all this parallel activity,.

B

I need to go in and do some aggregation things,.

B

Let's call that f3. and these blue cylinders here just represent.

B

Typically, some sort of a state that you'd need to coordinate.

B

All of these different things.

B

So if we jump into the code, here's kind of an example of that., which is basically it's a,, I need to back up a bunch of files.

B

so for example, in this case, first thing that this function.

B

Does is it says, hey, give me a list of files, right?.

B

So we have some function that goes and collects the lists of all the files that I wanna back up..

B

Now I'm doing a for loop and enumerating all of those files.

B

And I'm calling some other function which goes and copies them into blob. Storage.

B

now you'll notice that callactivityasync returns.

B

C

I'm not awaiting that task,. So.

B

I'm creating a list of tasks right, now.

B

and then, at the very end, I'm doing a task.whenall, waiting on that.

A

Yes, I have written this exact code, before.

B

Okay, great. and that's the hope, that this will be really familiar to.

B

People who've used the test, library, before.

B

and so what's going to happen, here, during the wait,.

B

Is the function is going to go to sleep.

B

So if you're familiar with our consumption based billing model,.

B

In azure,, you pay only when your function is running.

B

And so we've scheduled potentially hundreds of function.

B

Calls in parallel, and then this one is gonna, go to sleep and.

B

Wait for all of those other tasks to complete.

A

Okay,, so it's almost like line 36 here,, the task.winall,.

A

It's almost like you run that line for free.

B

Yeah,- and this is the really interesting orchestration pattern.

B

The fact that you can do a whenall on a bunch of distributed tasks that are running on who knows what number.

B

Of vms out in the cloud., and as soon as those are done,.

B

This task that you're waiting is going to resume.

B

You have the list of all those tasks, and so, if you want to do some aggregation,, like the sum, in this case, we return.

B

The number of bytes that we uploaded on blob storage.- I can get that number, and I can go and return it to the caller.

A

Awesome,, I mean, that sounds like a ton of value for customers.

C

I'm actually pretty impressed.

A

So basically, right.

A

So you'll be, say you had a hundred tasks, yes,.

B

The files array was a hundred elements, long.

A

And so,, this service is now put to sleep and then those hundred.

A

Separate service calls,, if that's what it is, are running.

A

And then the dobo framework or some part of azure understands.

A

The last one completed, and then it wakes back up?.

A

Yes. awesome and so you're not incurring any.

A

Pricing costs during that time, for this service?.

B

That's correct,, yeah,, no double billing.

A

So what has the reaction been to that particular part??.

A

Cuz, this is obviously new.

B

Yeah, so the reaction's been super positive, because previously it was really hard to do things like this,.

B

Especially in a distributed, environments.

B

you'd have to provision a bunch of tables and write to those and keep track. okay,. What are the tasks that I started??

B

What's the state of everything? okay,? Is it maybe I have to do some polling to figure out, okay,? Are they all done so I can start the next step?.

B

In this model, you don't have to do any of that,.

B

It's kind the,, all that state management that you had.

B

To do its kind of handled internally by the framework.

A

Makes sense,, can you tell us a little bit more bout, the actual api?.

A

So you've shown us really just one api, which was the-.

A

The call activity.

B

Yeah, the call activity, async api and obviously all the other.

A

Apis like test that when all, they're just framework apis.

C

Are there any other apis beyond the attributes that you,, like.

A

I guess durable orchestration context is one,, but.

A

That's kind of just the context that you need to work. With.

B

Yeah,, so you have the orchestration context and.

B

Let's see,, what's another good one?, so one thing I haven't actually shown yet is how do you actually start these orchestrations.

C

That's an excellent question., that's probably a good api for me to go to.

B

I would agree with that. I'll jump to this httpstart., so basically, what this function is is it's just an http trigger.

B

That I wrote,, which allows you to start some orchestration function by name.

B

So you just send in,. This is just regular http trigger.

B

I declare the authorization mechanism that I have for it,.

B

It takes in a post, request., I've got some route if you're familiar web api routes and.

B

You know this is very familiar to you.

B

and I've got this durableorchestrationclient, so.

B

This is an important api as well., and the main thing that you'll typically do with that is start.

B

A new orchestration., and so then you just pass in what is the name of.

B

The orchestration you wanna run,, what is some inputs?.

B

So that back site content one would be an example. Hey.

B

I wanna run that one and you know you're gonna, get back some sort of an instanceid. now this instanceid is important. If you want to kind of monitor.

B

What is the state of this abstraction that I just ran?.

B

Because if I'm backing up hundreds or thousands of files,.

B

Maybe it's gonna take some time,. Maybe you wanna be able to check on the status of that.

B

Maybe you wanna terminate that,, so we have a bunch of apis that'll help you with that.

B

And so I'll, just f12 on this client and we what be get back here.

C

Love f12. yes,.

B

It's one of my favorites., so we have apis like, hey,, I wanna get the status of my orchestration,. Is it still running,? Did it fail, give me some more information. or maybe I want to terminate it because hey,.

B

There's a bug in this: I need to stop it.

B

That's a little bit like cancellation.

B

yes,, a little bit like cancellation., so you get a few things like that, and some more advanced ones like raising events..

B

These functions have the ability to kind of stop and wait for.

B

Some event to be sent to it., so how do exceptions work?.

A

Great question., so if you're t1, sorry f1 is what I meant to say.

C

And you call in to f2, there is an unhandle exception in f2.

C

What does that look like from the perspective of f1?.

A

Is it just like tasks?.

B

Just like tasks. so again, this is just regular c# code.

B

So if some function throws an exception,, basically, this callactivityasync is going to turn around and.

B

Throw an exception that you can catch and.

B

Have whatever logic you want to do, compensation.

A

So I guess, if you have a badly written set of functions,.

A

Like in that picture, you went all the way to, I think f4, f5.

A

If, in that one, you have an unhandled exception and.

A

You didn't handle it,, it could come all the way back and.

A

Your whole set of functions could fail.

B

Yeah, potentially, you could determine how you wanna handle that.. If you wanna handle it,, we also have some,.

B

Going back to apis a little bit, sorry.

B

We have another version of callactivityasync that take in.

B

Some retry- I like that.,.

A

Parameters as well.

B

So if you want to, and sorry, I don't have a demo ready right.

B

Away, but we can just kinda wing it.

B

Let's see, we have a, oops, looks like it's the second parameter here.

B

we have a retry options, parameter that you can specify.

B

And that's gonna allow you to say,. If I want to automatically.

B

Retry,, if you have badly written functions, or.

B

You're calling some external service and it's flaky,.

B

You can specify hey,, I wanna retry this up to ten times.

B

And I wanna wait five seconds in between retries.

B

or maybe I wanna do kind of an exponential back-off.

B

We have apis that allow you to kinda do that,, so you.

B

Can explore the docs and see all the different confignovs there.

A

Right,, I guess you can also have code that basically,.

A

You could have a catch lane handler, and for some types of exceptions, you would manually retry.

A

Mm-Hm, absolutely. and for.

B

Other ones, you would just give up and say,. I know this.

A

Is an unrecoverable state,, so I have to do something different.

B

Mm-Hm, definitely.

A

Okay, so you talked about,, there are some nuget packages for this service.

A

Could you actually show us those again,, so people who.

A

Are watching from home kind of know what to go? Look for?.

B

Absolutely,, so I'm just gonna go back into my project file..

B

So if you're familiar with azure functions,, this is an azure functions. Project for visual studio.

B

So normally, when you create that,, you're gonna get this microsoft.net.sdk.functions.

B

And that basically tells visual studio, that, hey,.

B

This is a function app in f5, that I'm running the function's host locally.

A

And so this is a stable package?. It looks like it's not a preview one.

B

That's correct, okay, good.. This is stable. and then, for durable functions,. We have another nugget package.

B

That you can reference which is gonna, give you-.

A

In preview.

B

Yes, and that's gonna, give you all the functionality that we talked about.. We have both support for the full .net framework.

B

We also support .net standard 2.0.

B

So the attentive viewer will notice that I've got some conditional checks. Here,, like am I targeting .net standard 2.0,.

B

Or am I targeting full framework? right,, so would the guidance be that if you have fairly.

A

Simple needs-, mm-hm.

A

simple, is not quite the right word. basically,, if you don't need any apis that are in .net framework,.

A

And you don't need any specific apis that are in .net core,.

A

Then .net standard functions would be the way to go?.

B

Yes, and I should clarify too,, that as of right, now,,.

B

.Net standard support for functions is in preview,.

B

Whereas the full framework is completely supported.

B

So generally, you're gonna get,, as of right, now, you're gonna, get the best experience if you target the full framework..

B

But if you wanna do things like,, I wanna do development on a mac or linux, local development,.

B

Then .net standard is obviously a choice. You wanna go with.

B

And that moving forward,, that's what we're gonna be primarily supporting.

B

A

Is .net standard. so from an operating system? Standpoint, you just mentioned.

A

Operating systems, do I need to think about,? Am I going to.

A

Deploy this into a windows function, versus like a linux,.

A

Function,, like you kind of have to with docker containers?.

B

Or is that agnostic?.

B

To some extent you do have to know, right now., so for the local development experience,.

B

We wanted to make sure that we don't exclude anybody.

B

So with our .net standard support,, you can do it on a mac,.

B

On linux, or on windows. in azure,, if you're deploying this up to azure,.

B

Today, we only support running functions on windows.

B

So you just have to be careful if you're,, especially if you're targeting,, that you're not trying to.

B

Do anything that's platform, specific.

B

And if you're doing .net, you should probably be fine., I think most people who target other languages like node might,.

B

Some things you have to be aware: of.

B

and azure would just support windows, to-, right, and.

A

One place that people sometimes get tripped up, is the different slashes in the file system.

B

But with functions, that's.

A

Probably less of a concern, cuz, I assume for the most part,.

A

You're not hitting the disk in the traditional sense.

A

Is that right? for the most part, yeah.

B

I mean, you can, if you want., we have disks there, but usually don't need to use it.

A

Makes sense. okay, well,? I think we should kind of close up here.

A

I think we've had a great deep dive into durable functions.

A

Is there anything you kinda wanna close with? any call to action, or go over a couple key points?.

B

Let's see,, I guess,. Definitely it's in public preview,, try it out.

B

This is all open source as well,, so it's on github.

B

So I believe the repo name, it's under the azure organization.

B

Let's see, azure-functions-durable-extens-.

B

Ion. so go check it out, open up issues..

B

If you have anything, or even have just questions that you.

B

Wanna ask about this,, we actively monitor it.

B

and we're very engaged there., and we just love it,. If people could use it.

B

try it out, see if it works for your use, case.

B

and check out the documentation as well.

B

just search for durable functions. Online, and we've got.

B

Lots of docs to help make sure everyone can understand. It,, yes.

A

Okay, great, well, thanks for coming, chris.. This is certainly enlightening for me.

C

A

Yeah,, so this has been another episode of the on .net show, and today we were talking about azure, durable, functions.,.

A

Thanks for watching.