Media Replay Engine (MRE) is a framework for building automated video clipping and replay (highlight) generation pipelines using AWS services for live and video-on-demand (VOD) content. With MRE, you can focus on building the business logic around video clipping without worrying about pipeline orchestration, data movement and persistence.
MRE supports the following features:
- Catchup Replay generation
- After event Replay generation
- Integration with MediaLive for processing live or VOD content
- Event and Replay video export in MP4 and HLS formats
- Event and Replay data export in EDL and JSON formats
This repository contains the core MRE Framework
which is a set of secure REST APIs that you can interact with directly. It also contains the MRE Frontend application built using React if you are someone who prefers the ease-of-use of a graphical user interface (GUI) to interact with the APIs.
To get a head start in building automated video clipping pipelines using the MRE framework, check out the MRE Samples repository which contains sample Plugins, Profiles and ML Model Notebooks for performing feature detection for clip selection and prioritization.
- python == 3.11
- aws-cli
- aws-cdk >= 2.24.1
- docker
- node >= 20.10.0
- npm >= 10.2.3
- git
NOTE: Using an Amazon EC2 Instance is the SUPPORTED and RECOMMENDED method to install MRE:
- Use the
env-init.sh
script as User Data when launching your EC2 instance to automatically install all prerequisites- Configure permissions as appropriate for CDK usage (see
env-role-permissions.json
for policy statement example)- Tested configuration:
- Amazon Linux 2023 (AL2023)
- Architecture: x86_64
- Instance Type: t2.large
- EBS Storage: 16 GB gp3 (minimum requirement)
Before deployment:
- Activate the Python virtual environment created by
env-init.sh
:
source mredeploy-env/bin/activate
(located in/home/ec2-user/mredeploy/
)- Ensure your environment variables are set as needed
IMPORTANT: You are responsible for stopping or terminating your EC2 instance after deployment to avoid unnecessary charges
Greenfield deployment of MRE is suitable for customers who are looking to install from scratch.
Run the following commands to build and deploy MRE. Be sure to define values for REGION
and VERSION
first.
REGION=[specify the AWS region. For example, us-east-1]
VERSION=2.9.1
git clone https://github.com/awslabs/aws-media-replay-engine
cd aws-media-replay-engine
cd deployment
./build-and-deploy.sh --enable-ssm-high-throughput --version $VERSION --region $REGION [--profile <aws-profile> --admin-email <admin-email>]
In order to upgrade MRE Backend (StepFunctions, Lambda, API Gateway, EventBridge Rules, etc.), run the following commands. Be sure to define values for REGION
and VERSION
first.
REGION=[specify the AWS region. For example, us-east-1]
VERSION=2.9.1
git clone https://github.com/awslabs/aws-media-replay-engine
cd aws-media-replay-engine
cd deployment
./build-and-deploy.sh --disable-ui --version $VERSION --region $REGION [--profile <aws-profile>]
MRE uses AWS Amplify which offers a fully managed CI/CD and hosting service to deploy frontend apps on AWS by connecting to a Git repository such as GitHub or GitLab. As a part of MRE Greenfield deployment, AWS Amplify temporarily uses Amazon S3 to store the branch assets while deploying.
To upgrade MRE Frontend to the latest version, follow the below steps. Be sure to define a value for REGION
which is typically where MRE Frontend is already deployed.
- After cloning the repo from GitHub, navigate to aws-media-replay-engine/source/frontend/cdk folder.
- Run the following commands to update the Amplify Environment variables and/or Custom headers.
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
NOTE: The warning message about redeploying the frontend application after running the above command can be ignored as we will be doing that shortly.
- Run
python3 create-env-file.py $REGION
- Run
npm i --legacy-peer-deps and npm run build
from aws-media-replay-engine/source/frontend folder. - Zip the build assets located in aws-media-replay-engine/source/frontend/build folder
zip -r -q -X ./build.zip *
. - Run
python3 init-amplify.py $REGION Update
If you have the MRE Framework deployed to AWS you can configure and run the MRE frontend without having to redeploy the frontend stack. This can be useful if you want to quickly experiment and tinker with the user interface.
- Firstly you want to configure the frontend to use the correct AWS resources, you can create a .env file in the aws-media-replay-engine/source/frontend folder by copying the env.template file (executing
python3 create-env-file.py $REGION
will also create a .env file for you) - Fill in this env file with the correct information, this can all be found in the outputs of the MRE cloudformation stacks.
- Once the .env file is filled you can now install and run the MRE frontend from your local machine with the below commands:
npm i --legacy-peer-deps
npm run start
- This will install the packages needed by the frontend and then run it on localhost:3000. Be sure that you have Node 20.10.0 installed for this to work as intended.
If you choose to interact with the MRE framework using the REST APIs directly, you will need the below information from the Outputs tab of the ControlPlane CloudFormation stack:
- EndpointURL is the endpoint for accessing the APIs to create, read, update, delete (CRUD) Plugins, Models, Profiles, and schedule Events, Replays for processing.
The control plane is an API Gateway endpoint that includes APIs to create and manage different components of the video clipping and highlights generation process. These include:
Segmentation and Detection Pipeline
- An AWS Step Functions state machine generated dynamically with one or more user-defined Plugins (Lambda functions) to identify the mark-in (start) and mark-out (end) timestamp of the segments (clips) as well as detect interesting actions happening within those segments. To help decide the outcome of its analysis, a Plugin can optionally depend on a user-defined AI/ML model hosted either in AWS using services such as Rekognition, SageMaker, etc. or outside AWS. The pipeline is also configured to automatically publish different segmentation event notifications to an EventBridge event bus monitored by MRE for event-based downstream processing.Clip Generation
- A pre-defined AWS Step Functions state machine that is invoked as a part of the Segmentation and Detection pipeline to generate MP4 preview clips (for the GUI) and HLS manifest using the mark-in and mark-out timestamps of the identified segments. Clip generation, like the Segmentation and Detection pipeline, sends clip related event notifications to the MRE EventBridge event bus.Replay Generation
- Another pre-defined AWS Step Functions state machine which automatically selects segments containing key events to create Replay (Highlight) in various resolutions in popular video formats such as HLS and MP4. Replays can be chosen to be created in Catchup mode or after an Event is fully streamed and is completely event driven via Amazon EventBridge rules.Data Export
- Exports Event and Replay data into popular formats such as EDL and JSON via clip generation events triggered through Amazon EventBridge. This data can be optionally enriched (via a custom process) and ingested into video editing systems to create engaging fan user experience by overlaying video and other key event data on a timeline.
The data plane is an API Gateway endpoint that includes APIs using which the Plugins within the Segmentation and Detection pipeline can store and retrieve media assets as well as the processing metadata. There are also quite a few helper APIs available in the data plane that the Plugins can use for performing complex data queries and manipulations.
Path | Description |
---|---|
deployment/ | shell scripts and Dockerfile |
deployment/build-and-deploy.sh | shell script to build and deploy the solution using AWS CDK |
deployment/lambda_layer_factory/Dockerfile | install dependencies and create a container image |
deployment/lambda_layer_factory/docker-entrypoint.sh | shell script to build and package the Lambda layers as zip files within the container |
deployment/lambda_layer_factory/build-lambda-layer.sh | shell script to run docker for building and packaging the Lambda layers |
docs/ | shell scripts and code to build and deploy the API docs from source |
samples/ | folder containing sample plugins, profiles, and ML model notebooks |
source/ | source code folder |
source/frontend/ | source code folder for the Frontend application |
source/api/controlplane/ | source code folder for the control plane |
source/api/controlplane/*/infrastructure/ | control plane CDK application |
source/api/controlplane/*/runtime/ | control plane Chalice application |
source/api/dataplane/ | source code folder for the data plane |
source/api/dataplane/infrastructure/ | data plane CDK application |
source/api/dataplane/runtime/ | data plane Chalice application |
source/lib/ | source code folder for the custom Lambda layers |
source/lib/MediaReplayEnginePluginHelper/ | source code for the MediaReplayEnginePluginHelper library |
source/lib/MediaReplayEngineWorkflowHelper/ | source code for the MediaReplayEngineWorkflowHelper library |
source/gateway/infrastructure/ | API Gateway CDK application |
source/gateway/runtime/ | API Gateway Chalice application |
source/shared/infrastructure/ | Shared resources CDK application |
source/backend/caching/infrastructure/ | Segment Caching CDK application |
source/backend/caching/runtime/ | Segment Caching related Lambda functions |
source/backend/clipgeneration/infrastructure/ | ClipGeneration CDK application |
source/backend/clipgeneration/runtime/ | ClipGeneration related Lambda functions |
source/backend/data_export/infrastructure/ | Data Export CDK application |
source/backend/data_export/runtime/ | Data Export related Lambda functions |
source/backend/event-life-cycle/infrastructure/ | Event life cycle CDK application |
source/backend/event-life-cycle/runtime/ | Event life cycle related Lambda functions |
source/backend/replay/infrastructure/ | Replay CDK application |
source/backend/replay/runtime/ | Replay related Lambda functions |
source/backend/workflow_trigger/infrastructure/ | Workflow Trigger CDK application |
source/backend/workflow_trigger/runtime/ | Workflow Trigger related Lambda functions |
Check out Create automated intelligent highlights and replays
in AWS M&E Demo Landing Page to watch a demo of an automated Football (Soccer) video clipping pipeline built using MRE.
To know more about how MRE works and for instructions on how to build a video clipping application with MRE, refer to the Developer Guide.
MRE uses AWS_IAM to authorize REST API requests for both the Control and Data planes. The following screenshot shows how to test authentication to the MRE Control plane API using Postman. Be sure to specify the AccessKey
and SecretKey
for your own AWS environment.
The CloudFront distribution deployed by MRE allows for SSLv3 or TLSv1 for HTTPS viewer connections. However, it is recommended that customers help protect viewer connections by specifying a viewer certificate that enforces a minimum of TLSv1.2 in the security policy. In order to enforce TLSv1.2 in the security policy, update the Cloudfront Distribution by specifying a CNAME and the location of the SSL/TLS certificate. For more information on this, please refer to Cloudfront Distribution Viewer Certificate
After a CNAME has been configured for the MRE Cloudfront distribution, be sure to execute the following steps to reflect this change within MRE deployment.
- Navigate to AWS Systems Manager in the AWS Management console. Choose Parameter store and replace the value of the parameter /MRE/ControlPlane/MediaOutputDistribution with the CNAME of the MRE Cloudfront distribution.
- Navigate to AWS Amplify in the AWS Management console. Under All apps, choose mre_frontend. In the left pane, under app settings, choose Environment Variables. Change the value of the environment variable REACT_APP_CLOUDFRONT_DOMAIN_NAME to the CNAME of the MRE Cloudfront distribution.
- In the left pane, under app settings, choose Custom Headers. Edit the Header information by replacing the the default MRE Cloudfront distribution url with the CNAME of the MRE Cloudfront distribution.
- Within the AWS Amplify console, under All apps, choose MRE Frontend. Click on Build and then click on Redeploy this version. This will redeploy the MRE frontend taking into account the new CNAME for the Cloudfront distribution.
You are responsible for the cost of the AWS services used while running this solution.
AWS Service | Quantity | Cost |
---|---|---|
Amazon API Gateway | 150000 requests | $0.16 |
Amazon DynamoDB | 750000 writes, 146250 reads, 0.30 GB storage | $1.18 |
AWS Lambda | 12000 invocations, 2-minute avg. duration, 256 MB memory | $6 |
AWS Step Functions | 92400 state transitions | $2.21 |
Amazon S3 | 10 GB storage, 4000 PUT requests, 4000 GET requests | $0.26 |
AWS Elemental MediaConvert | 240 minutes | $4.08 |
Amazon Rekognition | 9000 Image analysis, 3 Custom Label inference units | $22.32 |
Amazon SageMaker | 2 inference endpoints | $5.13 |
These cost estimates are for a video clipping and replay (highlight) generation pipeline built using MRE to segment a Tennis game with a duration of 3 hours. This specific pipeline had a total of 4 plugins included in the profile (with 2 of those plugins using Machine Learning models hosted in Rekognition and SageMaker). At the end of the game, the pipeline outputted a total of 282 Tennis clips.
NOTE: For tips on how to reduce the processing cost of a pipeline built using MRE, please refer to the Operations Developer Guide.
-
While MRE deploys all the relevant AWS resources to facilitate automated video clipping and replay generation pipelines, you are still responsible for managing the service limits of those AWS resources via either AWS Service Quotas or AWS Support Center.
-
MRE currently uses a custom resource provider (based on Lambda) to work around a known AWS CDK limitation (aws/aws-cdk#12246) by creating the DynamoDB GSIs programmatically and polling them periodically to know when they become active. As of today, the custom resource provider in CDK has a maximum timeout limit of 2 hours and in some cases where the PluginResult table has more data (>20 GB), it could cause the Dataplane stack deployment to fail (especially when upgrading MRE to v2.5.1 or later). If this happens to you, please try one of the following workarounds:
- Delete old and unwanted Events via the MRE Frontend or
DELETE /event/{name}/program/{program}
Controlplane API. - Backup existing PluginResult table, purge it, rerun the MRE installer, and then restore the data from backup.
- Delete old and unwanted Events via the MRE Frontend or
# Delete the Frontend stack
cd aws-media-replay-engine/source/frontend/cdk
cdk destroy [--profile <aws-profile>]
# Delete the Gateway API stack
cd aws-media-replay-engine/source/gateway/cdk
cdk destroy [--profile <aws-profile>]
# Delete the Dataplane stack
cd aws-media-replay-engine/source/api/dataplane/infrastructure
cdk destroy [--profile <aws-profile>]
# Delete the Controlplane stacks
cd aws-media-replay-engine/source/api/controlplane/replay/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/api/controlplane/event/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/api/controlplane/workflow/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/api/controlplane/profile/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/api/controlplane/plugin/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/api/controlplane/model/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/api/controlplane/system/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/api/controlplane/contentgroup/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/api/controlplane/program/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/api/custompriorities/program/infrastructure
cdk destroy [--profile <aws-profile>]
# Delete the remaining stacks
cd aws-media-replay-engine/source/backend/workflow_trigger/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/backend/replay/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/backend/event-life-cycle/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/backend/data_export/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/backend/clipgeneration/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/backend/caching/infrastructure
cdk destroy [--profile <aws-profile>]
cd aws-media-replay-engine/source/shared/infrastructure
cdk destroy [--profile <aws-profile>]
- Execute
uninstall.sh
- Sign-in to the AWS CloudFormation console.
- Select the MRE Frontend stack.
- Choose Delete.
- Select the aws-mre-gateway stack.
- Choose Delete.
- Select the MRE Dataplane stack (aws-mre-dataplane).
- Choose Delete.
- Select all Controlplane stacks with the prefix "aws-mre-controlplane-" and delete them in the the order as outlined in the section Option 1 (Uninstall using AWS CDK).
- Select and delete the following stacks
- aws-mre-workflow-trigger
- aws-mre-replay-handler
- aws-mre-event-scheduler
- aws-mre-event-life-cycle
- aws-mre-data-exporter
- aws-mre-clip-generation
- aws-mre-segment-caching
- aws-mre-shared-resources
aws cloudformation delete-stack --stack-name <frontend-stack-name> --region <aws-region>
aws cloudformation delete-stack --stack-name <gateway-stack-name> --region <aws-region>
aws cloudformation delete-stack --stack-name <dataplane-stack-name> --region <aws-region>
Repeat this command for all stacks with the prefix aws-mre-controlplane- Delete the controlplane stacks in the the order as outlined in the section Option 1 (Uninstall using AWS CDK).
aws cloudformation delete-stack --stack-name <aws-mre-controlplane-*> --region <aws-region>
Delete the remaining Stacks
aws cloudformation delete-stack --stack-name aws-mre-workflow-trigger --region <aws-region>
aws cloudformation delete-stack --stack-name aws-mre-replay-handler --region <aws-region>
aws cloudformation delete-stack --stack-name aws-mre-event-scheduler --region <aws-region>
aws cloudformation delete-stack --stack-name aws-mre-event-life-cycle --region <aws-region>
aws cloudformation delete-stack --stack-name aws-mre-data-exporter --region <aws-region>
aws cloudformation delete-stack --stack-name aws-mre-clip-generation --region <aws-region>
aws cloudformation delete-stack --stack-name aws-mre-segment-caching --region <aws-region>
aws cloudformation delete-stack --stack-name aws-mre-shared-resources --region <aws-region>
MRE creates 5 S3 buckets that are not automatically deleted. To delete these buckets, follow the steps below:
- Sign in to the Amazon S3 console.
- Select the
LambdaLayerBucket
bucket. - Choose Empty.
- Choose Delete.
- Select the
MreMediaSourceBucket
bucket. - Choose Empty.
- Choose Delete.
- Select the
MreMediaOutputBucket
bucket. - Choose Empty.
- Choose Delete.
- Select the
MreSegmentCacheBucket
bucket. - Choose Empty.
- Choose Delete.
- Select the
MreDataExportBucket
bucket. - Choose Empty.
- Choose Delete.
- Select the
MreAccessLogsBucket
bucket. - Choose Empty.
- Choose Delete.
To delete the S3 bucket using AWS CLI, run the following command:
aws s3 rb s3://<bucket-name> --force
When utilizing a bucket where you are not the owner, only the bucket owner can configure notifications on a bucket. However, bucket owners can use a bucket policy to grant permission to other users to set this configuration with s3:PutBucketNotification permission.
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "TriggerPermission",
"Effect": "Allow",
"Principal": {
"Service": "lambda.amazonaws.com"
},
"Action": "s3:PutBucketNotification",
"Resource": "arn:aws:s3:::{BUCKET_NAME}",
"Condition": {
"StringEquals": {
"aws:SourceAccount": "{ACCOUNT_NUMBER}"
}
}
}
]
}
If you have a restrictive bucket policy that would explicitly prevent bucket access when providing your own bucket to MRE workflow- add the following policy to your bucket to allow access
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "MediaConvertRoleGet",
"Principal": {
"AWS": "arn:aws:iam::{ACCOUNT_NUMBER}:role/{MREMediaConvertIamRole from aws-mre-shared-resources stack}"
},
"Effect": "Allow",
"Action": [
"s3:GetObject",
"s3:ListBucket"
],
"Resource": [
"arn:aws:s3:::{BUCKET_NAME}",
"arn:aws:s3:::{BUCKET_NAME}/*"
]
},
{
"Sid": "ProbeVideoRoleGet",
"Principal": {
"AWS": "arn:aws:iam::{ACCOUNT_NUMBER}:role/{ProbeVideoLambdaRole from aws-mre-controlplane-profile stack}"
},
"Effect": "Allow",
"Action": [
"s3:GetObject",
"s3:ListBucket"
],
"Resource": [
"arn:aws:s3:::{BUCKET_NAME}",
"arn:aws:s3:::{BUCKET_NAME}/*"
]
},
{
"Sid": "AllowS3Trigger",
"Principal": {
"AWS": "arn:aws:iam::{ACCOUNT_NUMBER}:role/{ChaliceRole from aws-mre-controlplane-event stack}"
},
"Effect": "Allow",
"Action": [
"s3:*BucketNotification*",
],
"Resource": [
"arn:aws:s3:::{BUCKET_NAME}"
]
},
{
"Sid": "AllowS3Selection",
"Principal": {
"AWS": "arn:aws:iam::{ACCOUNT_NUMBER}:role/{ChaliceRole from aws-mre-controlplane-system stack}"
},
"Effect": "Allow",
"Action": [
"s3:List*Bucket*"
],
"Resource": [
"arn:aws:s3:::{BUCKET_NAME}"
]
}
]
}
See the CONTRIBUTING file for how to contribute.
The MRE Samples is licensed under MIT-0 license while the rest of the project is licensed under Apache-2.0 license.
Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.