Raleigh DevDay 2017: Serverless orchestration with AWS step functions

WIFI: awsDevDay | PASS: CodeHappy

U P N E X T :

Serverless Orchestration with

AWS Step Functions

T H A N K S T O O U R F R I E N D S A T :

© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.

Angela Wang, AWS Solutions Architect

August 1, 2017

Serverless Orchestration with

AWS Step Functions

© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved

I have this idea


Photo sharing app

Metadata

DynamoDB

Extract

Recognize

Store

…

Pictures

S3 Bucket


function handle(image, done, err) {

// metadata

var filemeta = extractImageMetadata(image);

if (!isSupported(metadata)) { err() };

var metadata = transformMetadata(filemeta);

done();

}

Extract the image metadata



// metadata




// image recognition

var labels = recognizePicture(image);

storeImageMetadata([metadata, labels])

done();

}

Use deep learning voodoo for object detection



// metadata







// generate thumbnail

generateThumbnail(image);

done();

}

Generate thumbnail and store in S3



// metadata









done();

}



// metadata









done();

}Everything should really be Microservices / Lambda functions, right?



// metadata

extractImageMetadata(image)

.then((img, meta) => isSupported(metadata) … )

.then((img, meta) => transform(metadata) … )


.then((img, meta) => recognizePicture(img) … )

.then((img, meta, labels) => storeImageMetadata … )


.then((img, meta, labels) =>

generateThumbnail(image) … );

}

Make sure we have our callbacks


function handle(image) {

// metadata

return extractImageMetadata(image)




.then((img, meta) => rekognizePicture(img) … )





}

What happens if that line fails?



// metadata










}

Can these two be done in parallel?



// metadata










}

This code is “dumb” (but hard to get right)

ExtractImageMetadata

ImageTypeCheck

TransformMetadata

Rekognition

StoreMetadata

Thumbnail



// metadata










}


ImageTypeCheck

TransformMetadata

Rekognition

StoreMetadata

Thumbnail

Queues and Notification



// metadata










}


ImageTypeCheck

TransformMetadata

Rekognition

Thumbnail

StoreMetadata



ImageTypeCheck

TransformMetadata

Rekognition Thumbnail

StoreMetadata

Start

End


Metadata

DynamoDB

Extract

Recognize

Store

…

Pictures

S3 Bucket

Photo sharing app


…

Photo sharing app

Pictures

S3 BucketStart Workflow

Lambda

Processing Workflow

Step Functions

Extract

Lambda

Recognize

Lambda

Store

Lambda

Metadata

DynamoDB

Object Detection

Rekognition


Demo



ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End

“I want to sequence functions”

“I want to run functions in parallel”

“I want to select functions based on data”

“I want to retry functions”

“I want try/catch/finally”

“I have code that runs for hours”

• Scales out

• Doesn’t lose state

• Deals with errors/timeouts

• Easy to build & operate

• Auditable



ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End

{

"StartAt": "ExtractImageMetadata",

"States": {

"ExtractImageMetadata": {

"Type": "Task",

"Resource": "arn:aws:lambda:mars-ter …”,

"InputPath": "$",

"ResultPath": "$.result",

"End": true

}

}

}

Define in JSON and Then

Visualize in the Console


Execute One or One Million


ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End


ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End


ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End


ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End



ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End

Seven State Types

Task: a single unit of work

Synchronous: Lambda

{


"Type": "Task",

"Resource": "arn:aws:lambda:mars-ter …”,

…

}

}



ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End

Seven State Types


Asynchronous: “Activity Task”, Polled by workers

{


"Type": ”Task",

"Resource": "arn:aws:states:mars-ter …”,

…

}

}


Use Activity Task for Manual Approval

Step Functions

Poll

taskToken

Amazon

SES

Poller

Approver

Send success

taskTokenAPI

Gatewayhttps://aws.amazon.com/blogs/compute/

implementing-serverless-manual-approval-steps-in-aws-

step-functions-and-amazon-api-gateway/


Use Activity Task for Calling Asynchronous APIs

Step Functions

Poll

taskToken

Amazon

Elastic

Transcoder

Poller

SNS

Send success

taskToken

Lambda Amazon Confidential



ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End

Seven State Types

NotSupported

Choice: Adds branching logic

{


"Type": ”Choice",

”Choices": [{

”Variable": ”…",

”StringEquals": ”…",

”Next": ”…",

}

],

”Default": ”NotSupported",

…

}

}




ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End

Seven State Types

NotSupported



Fail: Stops an execution and mark as failure

Success: Stops an execution successfully


StoreMetadata


ImageTypeCheck

TransformMetadata


Start

End

Seven State Types


Task: A single unit of work

Parallel: fork and join data across tasks




Seven State Types

Choice: a single unit of work





Pass: passes inputs to its outputs


ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End


Seven State Types







Wait: wait until n seconds.


ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End


Use Wait + Choice to Wait for AWS Batch Job

while:

sleep (30)

if status=done:

break

https://aws.amazon.com/blogs/compute/building-high-throughput-

genomics-batch-workflows-on-aws-workflow-layer-part-4-of-4/


Seven State Types







Wait: wait until n seconds.


ImageTypeCheck

TransformMetadata


StoreMetadata

Start

End


"Access Media": {

"Type": "Task",

"Resource": "arn:aws:lambda:eu-central-1:123456789012:function:FindMedia",

"TimeoutSeconds": 2,

"Next": "Graceful Exit",

"Retry": [

{

"ErrorEquals": [ "States.Timeout" ],

"IntervalSeconds": 2, "MaxAttempts": 2, "BackoffRate": 1.5

}

],

"Catch": [

{ "ErrorEquals": [ "States.ALL" ], "Next": "Clean Up" }

]

},



With AWS Step Functions, we can easily change and iterate on the application workflow of our food delivery service in order to optimize operations and continually improve delivery times. AWS Step Functions lets us dynamically scale the steps in our food delivery algorithm so we can manage spikes in customer orders and meet demand.

Mathias Nitzsche, CTO, foodpanda

“


Order Fulfillment



With AWS Step Functions, it was easy to build a multi-step product

updating system to ensure our database and website always have the

latest price and availability information.

AWS Step Functions let us replace a manual updating process with an

automated series of steps, including built-in retry conditions and error

handling, so we can reliably scale before a big show, and keep pace

with rapidly changing fashions.

Jared Browarnik, CTO, TheTake

“


Business Processes



AWS Step Functions makes it simple to coordinate information

from many different infrastructure systems using easy to design

workflows and create a more intelligent monitoring system for our

Platform as a Service (PaaS).

With AWS Step Functions, we can reliably automate monitoring

decisions and actions in order to reduce human intervention by

over 60%, which improves infrastructure operation productivity and

customer application availability on our platform.

Pedro Pimenta, VP R&D, OutSystems

“


13 AWS Lambda Task States

6 Choice States

1 Fail State

DevOps Automation


Document Processing


Integrate with other AWS services

• Create state machines and Activities with AWS

CloudFormation

• Call AWS Step Functions with Amazon API Gateway

• Start state machines in response to events, or on a

schedule, with Amazon CloudWatch Events

• Monitor state machine executions with Amazon CloudWatch

• Log API calls with AWS CloudTrail


Input/Output Processing

State [x]"InputPath" "ResultPath" "OutputPath"

Input

Output


InputPath example

{

”s3key”:”apple.png”,

”timestamp”:12324

}

Input


InputPath example

{



}

"InputPath”:“$”

Input


InputPath example

{



}

State [x]"InputPath”:“$”

Input

Input received by state x:

{



}


InputPath example

"InputPath”:“$.s3key”

{



}


Input


{



}


InputPath example

State [y]"InputPath”:“$.s3key”

{



}

”apple.png”


Input

Input received by state y:


{



}


ResultPath example

{



}

Input

State [x]State output:

{“type”: “png"}

"InputPath”

: “$”

”ResultPath”:“$” { ”type”:”png}Output


ResultPath example

{



}

Input

State [x]State output:

{“type”: “png"}

"InputPath”

: “$.s3key”

”ResultPath”:“$” { ”type”:”png}Output

”ResultPath”:“$.metadata”

{


”timestamp”:12324,

“metadata”: {

“type”: “png"}

}

Output


Where does transient application

state live?

In the machine, in JSON texts

passing from state to state.A:

Q:

© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reservedhttps://states-language.net/spec



2.5¢

How much?

per thousand

state transitions

4,000 free

transitions/month

Free tier:




“I want to retry functions”


Is this you?

“I have code that runs for hours”

“I want to run functions in parallel”


Resources

• Github repo:

https://github.com/awslabs/lambda-refarch-imagerecognition

• Blogs, reference architectures, tutorials on AWS Step

Functions:

https://aws.amazon.com/step-functions/getting-started



Thank You!

Don’t Forget Evaluations!

Raleigh DevDay 2017: Serverless orchestration with AWS step functions

Technology