feat(devtools): add devenv/direnv, default linting and formatting & scripts for build/scan vuln

.envrc

@@ -0,0 +1,7 @@
+#!/usr/bin/env bash
+
+eval "$(devenv direnvrc)"
+
+# You can pass flags to the devenv command
+# For example: use devenv --impure --option services.postgres.enable:bool true
+use devenv

.gitignore

@@ -21,3 +21,19 @@ charts/*/charts
 **/flagged/
 web-apps/**/overrides.yml
 **/.env
+
+# devenv
+.devenv
+.devenv.flake.nix
+devenv.lock
+
+# Devenv
+.devenv*
+devenv.local.nix
+devenv.local.yaml
+
+# direnv
+.direnv
+
+# pre-commit
+.pre-commit-config.yaml

devenv.nix

@@ -0,0 +1,133 @@
+{ pkgs, ... }:
+let
+  allComponents = "chat image-analysis flux-image-gen";
+  imagePrefix = "ghcr.io/stackhpc/azimuth-llm";
+  webAppsDir = "./web-apps";
+
+  # Resolves "all" or empty arg to the full list, validates otherwise.
+  resolveComponents = ''
+    ALL_COMPONENTS="${allComponents}"
+
+    resolve_components() {
+      local input="$1"
+      if [ -z "$input" ] || [ "$input" = "all" ]; then
+        echo "$ALL_COMPONENTS"
+      else
+        for c in $input; do
+          if ! echo " $ALL_COMPONENTS " | grep -q " $c "; then
+            echo "Unknown component: $c" >&2
+            echo "Available: $ALL_COMPONENTS" >&2
+            return 1
+          fi
+        done
+        echo "$input"
+      fi
+    }
+
+    image_name() {
+      echo "${imagePrefix}-''${1}-ui"
+    }
+  '';
+in
+{
+  env.GREET = "azimuth-llm dev environment";
+
+  packages = with pkgs; [
+    git
+    # Container
+    grype
+    # Helm / K8s
+    kubernetes-helm
+    chart-testing
+    kind
+    kubectl
+    # CI tooling
+    jq
+    yq-go
+    # Python
+    python311
+    ruff
+    black
+  ];
+
+  treefmt = {
+    enable = true;
+    config.programs = {
+      nixfmt.enable = true;
+      black.enable = true;
+    };
+  };
+
+  git-hooks.hooks = {
+    treefmt = {
+      enable = true;
+      settings.fail-on-change = false;
+    };
+  };
+
+  difftastic.enable = true;
+
+  scripts = {
+    build.exec = ''
+      ${resolveComponents}
+      TAG="latest"
+      COMPONENT=""
+      while [ $# -gt 0 ]; do
+        case "$1" in
+          --tag) TAG="$2"; shift 2 ;;
+          *)     COMPONENT="$COMPONENT $1"; shift ;;
+        esac
+      done
+      COMPONENT="''${COMPONENT## }"
+
+      TARGETS=$(resolve_components "$COMPONENT") || exit 1
+      for c in $TARGETS; do
+        echo "==> Building $c (tag: $TAG)"
+        docker build \
+          -t "$(image_name "$c"):$TAG" \
+          -f ${webAppsDir}/"$c"/Dockerfile \
+          ${webAppsDir}/
+      done
+    '';
+
+    scan.exec = ''
+      ${resolveComponents}
+      TAG="latest"
+      FAIL_ON="critical"
+      COMPONENT=""
+      while [ $# -gt 0 ]; do
+        case "$1" in
+          --tag)     TAG="$2"; shift 2 ;;
+          --fail-on) FAIL_ON="$2"; shift 2 ;;
+          *)         COMPONENT="$COMPONENT $1"; shift ;;
+        esac
+      done
+      COMPONENT="''${COMPONENT## }"
+
+      TARGETS=$(resolve_components "$COMPONENT") || exit 1
+      EXIT=0
+      for c in $TARGETS; do
+        build "$c" --tag "$TAG"
+
+        IMAGE="$(image_name "$c"):$TAG"
+        echo ""
+        echo "==> Scanning $IMAGE (fail-on: $FAIL_ON)"
+        if ! grype "$IMAGE" --fail-on "$FAIL_ON" --only-fixed; then
+          EXIT=1
+        fi
+      done
+      exit $EXIT
+    '';
+  };
+
+  enterShell = ''
+    echo "$GREET"
+    echo ""
+    echo "Commands (component = chat | image-analysis | flux-image-gen | omit for all):"
+    echo ""
+    echo "  prek  -a                                       Format/lint all files"
+    echo "  build [component] [--tag TAG]                  Build container image(s)"
+    echo "  scan  [component] [--tag TAG] [--fail-on SEV]  Build if needed + Grype scan"
+    echo ""
+  '';
+}

devenv.yaml

@@ -0,0 +1,13 @@
+inputs:
+  git-hooks:
+    url: github:cachix/git-hooks.nix
+    inputs:
+      nixpkgs:
+        follows: nixpkgs
+  nixpkgs:
+    url: github:NixOS/nixpkgs/nixpkgs-unstable
+  treefmt-nix:
+    url: github:numtide/treefmt-nix
+    inputs:
+      nixpkgs:
+        follows: nixpkgs

scripts/perf-test/stress.py

@@ -16,11 +16,12 @@
 prompts = [
     "Hi, how are you?",
     "What's the weather like with you?",
-    "Who's the best footballer of all time?"
+    "Who's the best footballer of all time?",
 ]
 
 client_count = 3
-request_count = 5 # Requests per client
+request_count = 5  # Requests per client
+
 
 def make_requests(client_id: int):
     client = Client(url)
@@ -32,7 +33,12 @@ def make_requests(client_id: int):
         timings.append(time.time() - start_time)
     return timings
 
-results = list(Parallel(n_jobs=client_count)(delayed(make_requests)(i) for i in range(1, client_count+1)))
+
+results = list(
+    Parallel(n_jobs=client_count)(
+        delayed(make_requests)(i) for i in range(1, client_count + 1)
+    )
+)
 all_timings = []
 for client_timings in results:
     all_timings += client_timings

web-apps/chat/app.py

@@ -62,23 +62,26 @@ class PossibleSystemPromptException(Exception):
     streaming=True,
 )
 
+
 def inference(latest_message, history):
     # Allow mutating global variable
     global BACKEND_INITIALISED
     log.debug("Inference request received with history: %s", history)
 
     try:
         context = []
-        model_instruction = settings.model_instruction.replace("{date}", f"{date.today()}")
+        model_instruction = settings.model_instruction.replace(
+            "{date}", f"{date.today()}"
+        )
         if INCLUDE_SYSTEM_PROMPT:
             context.append(SystemMessage(content=model_instruction))
         elif history and len(history) > 0:
             # Mimic system prompt by prepending it to first human message
-            history[0]['content'] = f"{model_instruction}\n\n{history[0]['content']}"
+            history[0]["content"] = f"{model_instruction}\n\n{history[0]['content']}"
 
         for message in history:
-            role = message['role']
-            content = message['content']
+            role = message["role"]
+            content = message["content"]
             if role == "user":
                 context.append(HumanMessage(content=content))
             else:
@@ -102,10 +105,10 @@ def inference(latest_message, history):
             # The "think" tags mark the chatbot's reasoning. Remove the content
             # and replace with "Thinking..." until the closing tag is found.
             content = chunk.content
-            if '<think>' in content or thinking:
+            if "<think>" in content or thinking:
                 thinking = True
                 response = "Thinking..."
-                if '</think>' in content:
+                if "</think>" in content:
                     thinking = False
                     response = ""
             else:
@@ -175,7 +178,7 @@ def inference_wrapper(*args):
     js=settings.custom_javascript,
     title=settings.page_title,
 ) as demo:
-    gr.Markdown('# ' + settings.page_title)
+    gr.Markdown("# " + settings.page_title)
     gr.ChatInterface(
         inference_wrapper,
         type="messages",
@@ -187,10 +190,10 @@ def inference_wrapper(*args):
             sanitize_html=True,
             autoscroll=False,
             latex_delimiters=[
-                {"left": "$$", "right": "$$", "display": True },
-                {"left": "$", "right": "$", "display": False }
-                ],
-            ),
+                {"left": "$$", "right": "$$", "display": True},
+                {"left": "$", "right": "$", "display": False},
+            ],
+        ),
     )
 
 

web-apps/chat/gradio-client-test.py

@@ -6,7 +6,7 @@
 gradio_host = sys.argv[1]
 
 retries = 60
-for n in range(1, retries+1):
+for n in range(1, retries + 1):
     try:
         client = Client(gradio_host)
         result = client.predict("Hi", api_name="/chat")

web-apps/chat/test.py

@@ -7,6 +7,7 @@
 url = os.environ.get("GRADIO_URL", "http://localhost:7860")
 client = Client(url)
 
+
 class TestSuite(unittest.TestCase):
 
     def test_gradio_api(self):
@@ -19,5 +20,6 @@ def test_gradio_api(self):
     #         # mock_response.assert_called_once_with("Hi", [])
     #         self.assertEqual(result, "Mocked")
 
+
 if __name__ == "__main__":
     unittest.main()

web-apps/flux-image-gen/api_server.py

@@ -31,10 +31,12 @@ class ImageGenInput(BaseModel):
     prompt: str
     add_sampling_metadata: bool
 
+
 @app.get("/")
 def health_check():
     return "Server is running"
 
+
 @app.get("/model")
 async def get_model():
     return {"model": model}
@@ -61,7 +63,9 @@ async def generate_image(input: ImageGenInput):
             add_sampling_metadata=input.add_sampling_metadata,
         )
         if not image:
-            return JSONResponse({"error": {"message": msg, "seed": seed}}, status_code=400)
+            return JSONResponse(
+                {"error": {"message": msg, "seed": seed}}, status_code=400
+            )
     # Convert image to bytes response
     buffer = io.BytesIO()
     image.save(buffer, format="jpeg")