Deploying Model

INFO

Files are under docker/ into quetzal-network-editor-backend

Model can be deployed on either Lambda or ECS

Configuration (Lambda)

1. Copy files from docker/template to the root of the model directory

Copy files from docker/template_lambda or docker/template_ecs

Lambda

├─ .env
├─ DockerFile.dockerignore
├─ requirements.txt
├─ modelConfig.json  
├─ step-functions.json   <-- template_lambda 
└─ Dockerfile            <-- template_lambda 

ECS

├─ .env
├─ DockerFile.dockerignore
├─ requirements.txt
├─ modelConfig.json
├─ steps.json            <-- template_ecs 
└─ Dockerfile            <-- template_ecs 

.env

2. Fill the environnement variable file .env .

AWS_ECR_REPO_NAME=<model_name>

<model_name> is defined in the infra folder /infra/models/environments/<model_name>.tfvars.

see the infra creation

DockerFile

The provided Dockerfile should work for any model without any modification. please note that the DockerFile for lambda and ECS are different

Dockerignore

3. Change <model_folder> to your model folder name. Add any other files that should be ignore to minimize the Docker image size.

**__pycache__**
**/*.pyc
**.cache**
**/*.ipynb_checkpoints*
**/*.DS_store*
<model_folder>/scenarios/*    <-- TO CHANGE
<model_folder>/.git/objects   <-- TO CHANGE

Note

Docker Build will be run from directory higher than the model directory. This is why we do write the model_folder in the paths.

Note

<model_folder>/scenarios/* should be in the dockerignore as those files are not to be dockerize. They are the scenarios files that will be on the Database (S3)

WARNING

a .git file is needed in the docker (for some obscure reason). So you should keep the <model_folder>/.git and only ignore the .git/objects subfolder.

requirements.txt

Ajust the python requirement if needed.

We recommand installing quetzal from pip. if not, modify the dockerfile to copy your local quetzal folder in the docker and provide all of its requirements.

step-functions.json (lambda)

INFO

For Lambda infra only. Skip this step if using ECS

5. Modify the step function configuration according to model steps and change <model_name> with your model name.

Each step in the step-function is a Notebook to run. It is what defines the model steps in the Run page

Lines to modify are marked with # TO EDIT. (# TO EDIT. must be removed from the filnal json file.)

The fist step is used to validate that the user has the correct access to run the model. (optional)

{
  "StartAt": "Authorization",
  "States": {
    "Authorization": {
      "Type": "Task",
      "Resource": "arn:aws:states:::lambda:invoke",
      "OutputPath": "$.Payload",
      "Parameters": {
        "Payload": {
          "authorization.$": "$.authorization",
          "model": <model_name>, # TO EDIT.
          "choice.$": "$.choice",
          "scenario_path_S3.$": "$.scenario_path_S3",
          "launcher_arg.$": "$.launcher_arg",
          "metadata.$": "$.metadata"
        },
        "FunctionName": "arn:aws:lambda:ca-central-1:142023388927:function:quetzal-api-auth:$LATEST"
      },
      ...
      "Next": "STEP 1" # TO EDIT
    },

Steps have names (STEP 1 here) and a Next argument leading to the next step. you can create as many step as you want.

The last step has "End": true instead of Next: Step_x

 "STEP 1": { # TO EDIT
      "Type": "Task",
      "Resource": "arn:aws:states:::lambda:invoke",
      "OutputPath": "$.Payload",
      "Parameters": {
        "Payload": {
          "notebook_path": "notebooks/transport/A10_STEP_1.ipynb", # TO EDIT
          "scenario_path_S3.$": "$.scenario_path_S3",
          "launcher_arg.$": "$.launcher_arg",
          "metadata.$": "$.metadata"
        },
        "FunctionName": "arn:aws:lambda:ca-central-1:142023388927:function:<model_name>"  # TO EDIT
      },
      "Retry": [
        {
          "ErrorEquals": [
            "Lambda.ServiceException",
            "Lambda.SdkClientException",
            "Lambda.TooManyRequestsException"
          ],
          "IntervalSeconds": 2,
          "MaxAttempts": 2,
          "BackoffRate": 2
        },
        {
          "ErrorEquals": [
            "Lambda.AWSLambdaException"
          ],
          "IntervalSeconds": 30,
          "MaxAttempts": 4,
          "BackoffRate": 2
        }
      ],
      "Next": "STEP 2" # TO EDIT
    },

steps.json (ECS)

INFO

For ECS infra only. Skip this step if using Lambda

5. Modify the steps configuration according to model steps.

Each step in the steps.jsons is a Notebook to run. It is what defines the model steps in the Run page

steps are executed in order.

[
    {
        "name": "default",
        "steps": [
            {
                "name": "step 1",
                "path": "notebooks/2_model/test_1.ipynb"
            },
            {
                "name": "step 2",
                "path": "notebooks/2_model/test_2.ipynb"
            }
        ]
    }
]

modelConfig.json

This file is used to set model wide configuration (values choices, units, etc) (see model config.)

Deploying

for the first deployment. you must at least deploy those 3 ressources. (see update-model)

• Docker model update
• steps steps update.
• scenario scenario update.