I am trying to deploy a sample web-classification flow to azure app service but instructions are either incomplete or incorrect.

[niveditau]

Once you have created a connection with azure open ai and generated a 'dist' folder and trying to build a docker image it gets stuck at
COPY ./connections/* /connections/ because there was no connections directory generated inside dist directory. I tried creating it manually and it moves forward with the docker image creation but then fails to create a connection at run time. Is here anyone able to successfully deploy the sample using docker even if just locally? Please share your work arounds. Thanks!

[wangchao1230]

Thanks for reaching out to us.
I reproed the issue and we will take a look.

[iljoong]

Since the documentation is missing connection setup, I had the same problem. Here is the workaround.

First move to /runit/promptflow-serve folder in the dist folder and open the run file.
You'll find how promptflow ininitalizes the app and you'll also see one of more pf connection create --file ... lines, just like below.

pf connection create --file /connections/azure_openai_connection.yaml

The name of connection file is the name of the connection used in the promptflow so it may differ in your file.

You also need copy the connection template file and update the value of api_key/api_base to environment variable placeholder.

For azure_openai_connection.yaml file example,

$schema: https://azuremlschemas.azureedge.net/promptflow/latest/AzureOpenAIConnection.schema.json
name: azure_open_ai_connection
type: azure_open_ai
api_key: "${env:AZURE_OPENAI_KEY}"
api_base: "${env:AZURE_OPENAI_BASE}"
api_type: "azure"

In case you have more than one connections, you need to do the same step.

Last, you need to set the right value of environment variable in the Web App Configuration or provide environment variable parameters for local docker run.

Hope this helpful!

[YingChen1996]

Hi, @niveditau. The fix is merged #1242 and will be released in the next stable version.