fix for EKS deployment

[danielrosendo]

No description provided.

[renan-souza]

@danielrosendo Read this warning :) Please don't manually change version.py

[danielrosendo]

I did not change version is 'resources/sample_settings.yaml'. Made only this change:

• allowed_hosts:
• • "0.1.2.3:*"

• mcp_allowed_hosts: ["localhost:", "127.0.0.1:"] # Default is ["localhost:", "127.0.0.1:"]
• mcp_allowed_origins: ["http://localhost:", "http://127.0.0.1:"] # Default is ["http://localhost:", "http://127.0.0.1:"]

[renan-souza]

Did you change the file version.py? It looks changed in this PR.

[danielrosendo]

No, I did not touch version.py

[renan-souza]

We already have a Dockerfile. Do we need two?

[danielrosendo]

The new file 'deployment/Dockerfile-aws' is a version that works for AWS EKS.

• It install only what we need (instead of .[all])
• It starts agent directly
• It exposes the MCP port

[renan-souza]

This can be made default in the existing Dockerfile

• It starts agent directly, if llm_agent is given in the EXTRAS deps (see below).
• It exposes the MCP port

So the only conflict that would require further analysis and/or changes would be checking if we really need [all] -- we'd need to check the CI that uses the container and if, we really need [all], parametrizing it with Docker build arguments (ARG)

Something like:

 ARG EXTRAS="all"

 # Use the ARG value in the pip install command. If no build arg EXTRAS is passed, it's going to install optional deps.
 RUN conda run -n flowcept pip install -e .[${EXTRAS}]

Then you can pass specific deps in your build command like this:

    docker build \
      --build-arg EXTRAS="lmdb,telemetry,dev,llm_agent,extras" \
      -t flowcept:latest \
      -f deployment/Dockerfile .

Flowcept has several dependable services and can be started in multiple ways. I think having one Dockerfile per way makes it harder to maintain. Also, that Dockerfile can be reused in multiple other envs, including ORNL's OpenShift cluster not only EKS.

[danielrosendo]

Sounds good!