Skip to main content

Configuration

A Docker Model Runner is also configured via the dais.json file.

Docker Model only​

The example below is the simplest configuration of a Docker Model Runner. It has a single Docker Model container with the attribute type: "model".

{
  "containers": [
    {
      "name": "model",
      "context": ".",
      "dockerfile": "model/Dockerfile",
      "type": "model",
      "maxModelDuration": 300
    }
  ]
}

Docker Model with a Post-Processor​

The Docker Model Runner framework also accepts a configuration which has the Docker Model paired with a Post-Processor service.

In the example configuration below, the pairing is enabled with the attribute postProcessContainerId: "post-processor" in the Docker Model container.

{
  "containers": [
    {
      "name": "model",
      "context": ".",
      "dockerfile": "model/Dockerfile",
      "type": "model",
      "maxModelDuration": 300,
      "postProcessContainerId": "post-processor"
    },
    {
      "name": "post-processor",
      "context": ".",
      "dockerfile": "post-processor/Dockerfile",
      "port": 8080
    }
  ]
}

With this configuration, Dais will notify the persistent service with name identified in the the postProcessContainerId attribute, i.e. the post-processor service, by sending it a summary of all the model runs.

Docker Model with a Pre- and a Post-Processor​

Another dais.json configuration that the Docker Model Runner framework accepts is one that has both a Pre-Processor (or Task-Generator) and a Post-Processor services.

In addition to having the postProcessContainerId attribute, the Docker Model in this configuration will also have the taskGeneratorContainerName attribute. In the example below, the pre-processor service is identified to provide the Pre-Processor service.

{
  "containers": [
    {
      "name": "model",
      "context": ".",
      "dockerfile": "model/Dockerfile",
      "type": "model",
      "maxModelDuration": 300,
      "taskGeneratorContainerName": "pre-processor",
      "postProcessContainerId": "post-processor"
    },
    {
      "name": "pre-processor",
      "context": ".",
      "dockerfile": "pre-processor/Dockerfile",
      "port": 8080
    },
    {
      "name": "post-processor",
      "context": ".",
      "dockerfile": "post-processor/Dockerfile",
      "port": 8080
    }
  ]
}

The persistent service named in the taskGeneratorContainerName attribute, will be queried to generate a list of model inputs from a single base input when a model run is triggered in the frontend.