Merge pull request #126 from stackhpc/ci-stability-improvements

CI stability improvements

Author: sd109

.github/workflows/test-pr.yml

@@ -69,6 +69,17 @@ jobs:
 
       - name: Run Gradio app tests
         run: |
+          echo "Installing ollama..."
+          curl -fsSL https://ollama.com/install.sh | sh
+
+          # Give ollama systemd unit time to start
+          sleep 5
+
+          MODEL=smollm2:135m
+          echo "Pulling model $MODEL..."
+          ollama pull $MODEL
+
+          echo "Running Gradio test script"
           ./test-images.sh $(git rev-parse --short ${{ github.sha }})
         working-directory: web-apps
 

charts/azimuth-llm/ci/default-values.yaml

@@ -10,6 +10,8 @@ api:
     enabled: false
   # No GPUs in CI runners
   gpus: 0
+  # Limit context length for efficiency
+  modelMaxContextLength: 100
 ui:
   service:
     zenith:

web-apps/chat/Dockerfile

@@ -1,10 +1,11 @@
-FROM python:3.11-slim
+FROM ghcr.io/astral-sh/uv:python3.11-bookworm-slim
 
 ARG DIR=chat
 
 COPY $DIR/requirements.txt requirements.txt
+RUN sed -i s$../utils$./utils$ requirements.txt
 COPY utils utils
-RUN pip install --no-cache-dir -r requirements.txt
+RUN uv pip install --system --no-cache-dir -r requirements.txt
 
 COPY purge-google-fonts.sh purge-google-fonts.sh
 RUN bash purge-google-fonts.sh

web-apps/chat/defaults.yml

@@ -1,8 +1,7 @@
 
 # Default target is a local ollama instance
-# running inside the same docker network
 model_name: smollm2:135m
-backend_url: http://ollama:11434
+backend_url: http://localhost:11434
 
 host_address: 0.0.0.0
 

web-apps/chat/requirements.txt

@@ -1,7 +1,7 @@
 gradio<6
 gradio_client
 openai
-langchain
+langchain<1.0
 langchain_openai
 pydantic
 structlog

web-apps/flux-image-gen/Dockerfile

@@ -1,4 +1,4 @@
-FROM python:3.11
+FROM ghcr.io/astral-sh/uv:python3.11-bookworm
 
 # https://stackoverflow.com/questions/55313610/importerror-libgl-so-1-cannot-open-shared-object-file-no-such-file-or-directo
 RUN apt-get update && apt-get install ffmpeg libsm6 libxext6  -y
@@ -7,7 +7,7 @@ RUN apt-get update && apt-get install ffmpeg libsm6 libxext6  -y
 ARG DIR=flux-image-gen
 
 COPY $DIR/requirements.txt requirements.txt
-RUN pip install --no-cache-dir -r requirements.txt
+RUN uv pip install --system --no-cache-dir -r requirements.txt
 
 COPY purge-google-fonts.sh .
 RUN bash purge-google-fonts.sh

web-apps/image-analysis/Dockerfile

@@ -1,10 +1,11 @@
-FROM python:3.11-slim
+FROM ghcr.io/astral-sh/uv:python3.11-bookworm-slim
 
 ARG DIR=image-analysis
 
 COPY $DIR/requirements.txt requirements.txt
+RUN sed -i s$../utils$./utils$ requirements.txt
 COPY utils utils
-RUN pip install --no-cache-dir -r requirements.txt
+RUN uv pip install --system --no-cache-dir -r requirements.txt
 
 COPY purge-google-fonts.sh purge-google-fonts.sh
 RUN bash purge-google-fonts.sh

web-apps/test-images.sh

@@ -1,6 +1,12 @@
 #!/bin/bash
 set -e
 
+# NOTE(sd109): This script relies on docker host networking in order
+# to communicate from Gradio to ollama running on localhost. This means
+# it does not work on MacOS since docker host networking does not work.
+# It is intended to be run on a Linux machine and is primarily used in
+# GitHub runners as part of CI testing.
+
 IMAGE_TAG="${1:-latest}"
 echo Testing image tag $IMAGE_TAG
 
@@ -41,10 +47,12 @@ test() {
 
     if [[ -f $1/test.py ]]; then
 
-        DOCKER_NET_NAME=azimuth-llm-shared
-        if [[ ! $(docker network ls | grep $DOCKER_NET_NAME) ]]; then
-            docker network create $DOCKER_NET_NAME
-        fi
+        DOCKER_NET_NAME=host
+
+        # DOCKER_NET_NAME=azimuth-llm-shared
+        # if [[ ! $(docker network ls | grep $DOCKER_NET_NAME) ]]; then
+        #     docker network create $DOCKER_NET_NAME
+        # fi
 
         # Ensure app image is available
         IMAGE=$(image_name $1):$IMAGE_TAG
@@ -56,33 +64,41 @@ test() {
         fi
 
         # Ensure Ollama instance is available
-        if [[ $(curl -s localhost:11434) == "Ollama is running" ]]; then
-            log "Using existing ollama process running on localhost:11434"
-        else
-            log "Ollama process not running - starting containerised server"
-            docker run --rm --network $DOCKER_NET_NAME -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama
-            sleep 3
-            docker exec ollama ollama pull smollm2:135m
+        if [[ $(curl -s localhost:11434) != "Ollama is running" ]]; then
+            log "Ollama not running on localhost:11434 - aborting test"
+            exit 1
+            # log "Using existing ollama process running on localhost:11434"
+        # else
+        #     log "Ollama process not running - starting containerised server"
+        #     docker run --rm --network $DOCKER_NET_NAME -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama
+        #     sleep 3
+        #     docker exec ollama ollama pull smollm2:135m
         fi
 
         log "Starting Gradio app container"
         docker run --network $DOCKER_NET_NAME -d --name $1-app $IMAGE
 
         # Give the app time to start
-        sleep 3
+        sleep 10
 
         log "Running tests"
+        # docker run --network $DOCKER_NET_NAME --rm \
+        #     --name $1-test-suite \
+        #     -e GRADIO_URL=http://$1-app:7860 --entrypoint python \
+        #     $IMAGE \
+        #     test.py
         docker run --network $DOCKER_NET_NAME --rm \
             --name $1-test-suite \
-            -e GRADIO_URL=http://$1-app:7860 --entrypoint python \
+            --entrypoint python \
             $IMAGE \
             test.py
 
         log "Removing containers:"
-        docker rm -f ollama $1-app
+        # docker rm -f ollama $1-app
+        docker rm -f $1-app
 
-        log "Removing docker network:"
-        docker network rm $DOCKER_NET_NAME
+        # log "Removing docker network:"
+        # docker network rm $DOCKER_NET_NAME
 
         echo
         echo "----- Tests succeed -----"