Initial commit of home-orchestrator

Project created by Hugo Ferreira

Signed-off-by: Alexandre Peixoto Ferreira <[email protected]>

.github/workflows/smarter-org-docker-buildx.yml

@@ -0,0 +1,21 @@
+name: Docker Image BuildX CI and Publish
+
+# This workflow uses actions that are not certified by GitHub.
+# They are provided by a third-party and are governed by
+# separate terms of service, privacy policy, and support
+# documentation.
+
+on:
+  schedule:
+    - cron: "19 16 * * *"
+  push:
+    branches: ["main"]
+    # Publish semver tags as releases.
+    tags: ["v*.*.*"]
+  pull_request:
+    branches: ["main"]
+  workflow_dispatch:
+
+jobs:
+  build:
+    uses: smarter-project/reusable-workflows/.github/workflows/smarter-org-docker-buildx.yml@main

.github/workflows/smarter-org-helm.yml

@@ -0,0 +1,14 @@
+# release.yaml
+name: Release Charts
+
+on:
+  push:
+    branches:
+      - main
+
+jobs:
+  release:
+    uses: smarter-project/reusable-workflows/.github/workflows/smarter-org-helm.yml@main
+    secrets:
+      GPG_KEYRING_BASE64: ${{ secrets.GPG_KEYRING_BASE64 }}
+      GPG_PASSPHRASE: ${{ secrets.GPG_PASSPHRASE }}

Dockerfile

@@ -0,0 +1,19 @@
+FROM alpine:3.19
+
+# Install required tools
+RUN apk add --no-cache \
+    curl \
+    jq \
+    bash
+
+# Set working directory
+WORKDIR /app
+
+# Copy the setup script
+COPY initial_setup.sh .
+
+# Make the script executable
+RUN chmod +x initial_setup.sh
+
+# Set the entrypoint
+ENTRYPOINT ["./initial_setup.sh"] 

README.md

@@ -0,0 +1,7 @@
+# Part of Smart-home security demo using AI - LLMs
+
+## Motivation
+
+## Usage
+
+Container that configures ML infrastructure

initial_setup.sh

@@ -0,0 +1,103 @@
+#!/usr/bin/env bash
+
+# For future readers: So here's the thing... right now, both the host and the guest load immediately,
+# which means that we had two instances of the model loaded into memory by using this script.
+#
+# There is a way to unload the model from the host as seen in the Ollama docs:
+# https://github.com/ollama/ollama/blob/main/docs/api.md#unload-a-model-1
+#
+# unload_model_from_memory() {
+#    echo "Unloading model from memory..."
+#    curl -X POST "${OLLAMA_HOST_URL}/api/chat" -d "{
+#        \"model\": \"$OLLAMA_MODEL\",
+#        \"messages\": [],
+#        \"keep_alive\": 0
+#    }"
+# }
+#
+# But that wouldn't be a robust solution, and it doesn't necessarily avoid OOM, without
+# signalling to the guest that the setup is complete, and then have the guest load the model on its own.
+#
+# So what we do instead is receive two URL's: one for the host, one for the guest.
+# The host URL is the one we use to download the model, and the guest URL is the one
+# we use to test the model by sending a simple prompt.
+
+# Environment variables with defaults
+export OLLAMA_MODEL=${OLLAMA_MODEL:-"llama3.2-vision:latest"}
+export OLLAMA_HOST_URL=${OLLAMA_HOST_URL:-"http://localhost:11435"}
+export OLLAMA_GUEST_URL=${OLLAMA_GUEST_URL:-"http://localhost:11434"}
+export OLLAMA_MAX_RETRIES=${OLLAMA_MAX_RETRIES:-30}
+export OLLAMA_RETRY_DELAY=${OLLAMA_RETRY_DELAY:-2}
+export OLLAMA_WAIT_FOREVER=${OLLAMA_WAIT_FOREVER:0}
+
+# Helper functions
+wait_for_ollama() {
+    local service_type=$1
+    local url="OLLAMA_${service_type}_URL"
+    url="${!url}" # Get either OLLAMA_HOST_URL or OLLAMA_GUEST_URL
+
+    echo "Waiting for Ollama $service_type service to be ready..."
+    for i in $(seq 1 $OLLAMA_MAX_RETRIES); do
+        if curl -s "${url}/api/tags" > /dev/null; then
+            echo "Ollama $service_type service is ready!"
+            return 0
+        fi
+
+        if [ $i -eq $OLLAMA_MAX_RETRIES ]; then
+            echo "Timeout waiting for Ollama $service_type service"
+            return 1
+        fi
+
+        echo "Attempt $i/$OLLAMA_MAX_RETRIES: $service_type service not ready yet, waiting ${OLLAMA_RETRY_DELAY}s..."
+        sleep $OLLAMA_RETRY_DELAY
+    done
+}
+
+pull_model() {
+    echo "Pulling model $OLLAMA_MODEL..."
+    if ! curl -f -X POST "${OLLAMA_HOST_URL}/api/pull" -d "{\"name\": \"$OLLAMA_MODEL\"}"; then
+        echo "Failed to pull model $OLLAMA_MODEL"
+        return 1
+    fi
+}
+
+test_model() {
+    echo "Testing model with a simple prompt..."
+    local RESPONSE
+    RESPONSE=$(curl -s "${OLLAMA_GUEST_URL}/api/generate" -d "{
+        \"model\": \"$OLLAMA_MODEL\",
+        \"prompt\": \"Why is the sky blue? Answer in less than 10 words.\",
+        \"stream\": false,
+        \"options\": { \"seed\": 123 }
+    }")
+
+    if [ $? -ne 0 ]; then
+        echo "Failed to connect to ${OLLAMA_GUEST_URL}"
+        return 1
+    fi
+
+    if ! echo "$RESPONSE" | jq -e '.response' >/dev/null 2>&1; then
+        echo "Invalid response format from model. Got:"
+        echo "$RESPONSE" | jq '.'
+        return 1
+    fi
+
+    echo "Model test successful. Response:"
+    echo "$RESPONSE" | jq -c '.response'
+}
+
+# Main execution
+main() {
+    wait_for_ollama HOST || exit 1
+    pull_model || exit 1
+    wait_for_ollama GUEST || exit 1
+    test_model || exit 1
+    echo "Setup complete!"
+    if [ ${OLLAMA_WAIT_FOREVER} -gt 0 ]; then
+	while true; do
+	    sleep 300
+	done
+    fi
+}
+
+main