# Task Docker image

WARNING

App Engine and its Task system are still in BETA. Therefore, all information presented in this documentation is subject to potentially breaking changes.

In order to package the execution environement and implementation of a Task, App Engine relies on Docker images. This section provides guidelines on how to create a Docker image for a Task.

# Task implementation

The implementation of the Task is at the discretion of the developer, who may choose any programming language, technology, or framework, provided it can be encapsulated within a Docker container. The App Engine excutes the Task by launching a container from the specified Docker image and initiating the image's default entrypoint and/or command.

The requirements for implementing the Task are as follows:

The Docker image's entrypoint or command should initiate the Task's script or program.
The Task should read input data from the /inputs directory and output data to the /outputs directory, adhering to the guidelines outlined in the Task I/O documentation.
The main process of the Task must exit with a non-zero code to indicate failure and 0 to indicate successful execution.

# How to bundle the Task image

Once the Dockerfile is ready, the image tar archive can be created to be included in the Task bundle (which also contains the Task descriptor file).

Creating the tar archive for a Task with namespace com.my.task and version 1.0.0 involves two steps:

build the Docker image and name it using the namespace as a tag:
```
docker build -t com/my/task:1.0.0 -f Dockerfile .
```
1

save the Docker image as a tar archive:

docker save com/my/task:1.0.0 -o image.tar

WARNING

IMPORTANT: the image name must be the modified namespace and version, where the modified namespace is simply the Task namespace with . replaced by /. For example, the namespace com.my.task becomes com/my/task. The task version is then appended as a regular Docker image tag: com/my/task:1.0.0.

# Testing your Task image

Because the Task implementation is independent from Cytomine and the App Engine, it is possible to test the Task image locally before uploading it to Cytomine:

Prepare the input and output directories:
```
mkdir -p ./inputs ./outputs
```
1
Create the input files required by the Task in the ./inputs directory following the guidelines in the Task I/O documentation.
Load the Docker image from the tar archive:
```
docker load -i image.tar
```
1

Execute the Task image as a container, mounting the input and output directories:

docker run -v $(pwd)/inputs:/inputs -v $(pwd)/outputs:/outputs com/my/task:1.0.0

At the end of the execution, the outputs should have been generated in the outputs directory.

← Concepts Task I/O →