adding container to update models on a schedule

.github/workflows/pipeline.yaml

@@ -66,7 +66,7 @@ jobs:
           sudo apt-get update
           sudo apt-get install docker-ce
       - name: Build Docker Image
-        run: docker build --tag groundlight-edge .
+        run: docker build -f Dockerfile.app --tag groundlight-edge .
 
       - name: Start Docker Container
         id: start_container
@@ -136,7 +136,7 @@ jobs:
           sudo apt-get update
           sudo apt-get install docker-ce
       - name: Build Docker Image 
-        run: docker build --tag groundlight-edge .
+        run: docker build -f Dockerfile.app --tag groundlight-edge .
 
       - name: Start Docker Container 
         id: start_container

Dockerfile.model_updater

@@ -0,0 +1,42 @@
+# Build args 
+ARG APP_ROOT="/model_updater"
+ARG POETRY_HOME="/opt/poetry"
+ARG POETRY_VERSION=1.5.1
+
+FROM python:3.11-slim-bullseye 
+
+# Args that are needed in this stage
+ARG APP_ROOT
+ARG POETRY_HOME
+ARG POETRY_VERSION
+
+# Install required dependencies and tools
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+       bash \
+       curl \
+       libglib2.0-0 \
+       libgl1-mesa-glx \
+    && apt-get clean \
+    && rm -rf /var/lib/apt/lists/* \
+    && curl -sSL https://install.python-poetry.org | python -
+
+# Set Python and Poetry ENV vars
+ENV PYTHONUNBUFFERED=1 \
+    PYTHONDONTWRITEBYTECODE=1 \
+    POETRY_HOME=${POETRY_HOME} \
+    POETRY_VERSION=${POETRY_VERSION} \
+    PATH=${POETRY_HOME}/bin:$PATH
+
+COPY model_updater ${APP_ROOT}/model_updater
+COPY app/core ${APP_ROOT}/app/core
+
+COPY ./pyproject.toml ${APP_ROOT}/
+
+WORKDIR ${APP_ROOT}
+
+# Install production dependencies only
+RUN poetry install --only model_updater --no-interaction --no-root
+
+# Entry command 
+CMD poetry run python -m model_updater.update_models

app/core/app_state.py

@@ -14,11 +14,12 @@
 
 from app.core.utils import safe_call_api
 
-from .configs import LocalInferenceConfig, MotionDetectionConfig, RootEdgeConfig
+from .configs import LocalInferenceConfig, MotionDetectionConfig
 from .edge_inference import EdgeInferenceManager
-from .file_paths import DEFAULT_EDGE_CONFIG_PATH, INFERENCE_DEPLOYMENT_TEMPLATE_PATH
+from .file_paths import INFERENCE_DEPLOYMENT_TEMPLATE_PATH
 from .iqe_cache import IQECache
 from .motion_detection import MotionDetectionManager
+from .utils import load_edge_config
 
 logger = logging.getLogger(__name__)
 
@@ -27,29 +28,6 @@
 TTL_TIME = 3600  # 1 hour
 
 
-def load_edge_config() -> RootEdgeConfig:
-    """
-    Reads the edge config from the EDGE_CONFIG environment variable if it exists.
-    If EDGE_CONFIG is not set, reads the default edge config file.
-    """
-    yaml_config = os.environ.get("EDGE_CONFIG", "").strip()
-    if yaml_config:
-        config = yaml.safe_load(yaml_config)
-        return RootEdgeConfig(**config)
-
-    logger.warning("EDGE_CONFIG environment variable not set. Checking default locations.")
-
-    default_paths = [DEFAULT_EDGE_CONFIG_PATH, "configs/edge-config.yaml"]
-
-    for path in default_paths:
-        if os.path.exists(path):
-            logger.info(f"Loading edge config from {path}")
-            config = yaml.safe_load(open(path, "r"))
-            return RootEdgeConfig(**config)
-
-    raise FileNotFoundError(f"Could not find edge config file in default locations: {default_paths}")
-
-
 @lru_cache(maxsize=MAX_SDK_INSTANCES_CACHE_SIZE)
 def _get_groundlight_sdk_instance_internal(api_token: str):
     return Groundlight(api_token=api_token)

app/core/utils.py

@@ -1,10 +1,41 @@
+import logging
+import os
 from io import BytesIO
 from typing import Callable
 
 import ksuid
+import yaml
 from fastapi import HTTPException
 from PIL import Image
 
+from .configs import RootEdgeConfig
+from .file_paths import DEFAULT_EDGE_CONFIG_PATH
+
+logger = logging.getLogger(__name__)
+
+
+def load_edge_config() -> RootEdgeConfig:
+    """
+    Reads the edge config from the EDGE_CONFIG environment variable if it exists.
+    If EDGE_CONFIG is not set, reads the default edge config file.
+    """
+    yaml_config = os.environ.get("EDGE_CONFIG", "").strip()
+    if yaml_config:
+        config = yaml.safe_load(yaml_config)
+        return RootEdgeConfig(**config)
+
+    logger.warning("EDGE_CONFIG environment variable not set. Checking default locations.")
+
+    default_paths = [DEFAULT_EDGE_CONFIG_PATH, "configs/edge-config.yaml"]
+
+    for path in default_paths:
+        if os.path.exists(path):
+            logger.info(f"Loading edge config from {path}")
+            config = yaml.safe_load(open(path, "r"))
+            return RootEdgeConfig(**config)
+
+    raise FileNotFoundError(f"Could not find edge config file in default locations: {default_paths}")
+
 
 def safe_call_api(api_method: Callable, **kwargs):
     """

app/main.py

@@ -18,19 +18,3 @@
 app.include_router(router=ping_router)
 
 app.state.app_state = AppState()
-
-
-@app.on_event("startup")
-async def on_startup():
-    """
-    On startup, update edge inference models.
-    """
-    if not os.environ.get("DEPLOY_DETECTOR_LEVEL_INFERENCE", None):
-        return
-
-    for detector_id, inference_config in app.state.app_state.edge_inference_manager.inference_config.items():
-        if inference_config.enabled:
-            try:
-                app.state.app_state.edge_inference_manager.update_model(detector_id)
-            except Exception:
-                logging.error(f"Failed to update model for {detector_id}", exc_info=True)

deploy/bin/build-push-edge-endpoint-image.sh

@@ -38,4 +38,5 @@ docker buildx inspect tempgroundlightedgebuilder --bootstrap
 docker buildx build \
   --platform linux/arm64,linux/amd64 \
   --tag 723181461334.dkr.ecr.us-west-2.amazonaws.com/edge-endpoint:${TAG} \
+  -f ../../Dockerfile.app \
   ../.. --push

deploy/bin/build-push-model-updater.sh

@@ -0,0 +1,40 @@
+#!/bin/bash
+
+set -ex
+
+
+# Ensure that you're in the same directory as this script before running it
+cd "$(dirname "$0")"
+
+TAG=$(./git-tag-name.sh)
+
+# Authenticate docker to ECR
+aws ecr get-login-password --region us-west-2 | docker login --username AWS --password-stdin  723181461334.dkr.ecr.us-west-2.amazonaws.com
+
+# We use docker buildx to build the image for multiple platforms. buildx comes
+# installed with Docker Engine when installed via Docker Desktop. If you're
+# on a Linux machine with an old version of Docker Engine, you may need to
+# install buildx manually. Follow these instructions to install docker-buildx-plugin:
+# https://docs.docker.com/engine/install/ubuntu/
+
+# Install QEMU, a generic and open-source machine emulator and virtualizer
+docker run --rm --privileged linuxkit/binfmt:af88a591f9cc896a52ce596b9cf7ca26a061ef97
+
+# Check if tempbuilder already exists
+if ! docker buildx ls | grep -q tempmodelupdaterbuilder; then
+  # Prep for multiplatform build - the build is done INSIDE a docker container
+  docker buildx create --name tempmodelupdaterbuilder --use
+else
+  # If tempbuilder exists, set it as the current builder
+  docker buildx use tempmodelupdaterbuilder
+fi
+
+# Ensure that the tempbuilder container is running
+docker buildx inspect tempmodelupdaterbuilder --bootstrap
+
+# Build model updater image for amd64 and arm64
+docker buildx build \
+  --platform linux/arm64,linux/amd64 \
+  --tag 723181461334.dkr.ecr.us-west-2.amazonaws.com/edge-endpoint:${TAG} \
+  -f ../../Dockerfile.model_updater \
+  ../.. --push

deploy/k3s/edge_deployment/edge_deployment.yaml

@@ -40,7 +40,7 @@ spec:
       serviceAccountName: edge-endpoint-service-account
       containers:
       - name: edge-endpoint
-        image: 723181461334.dkr.ecr.us-west-2.amazonaws.com/edge-endpoint:13361241c-state-management-via-k3s
+        image: 723181461334.dkr.ecr.us-west-2.amazonaws.com/edge-endpoint:4526476fb-model-updater
         imagePullPolicy: IfNotPresent
         ports:
         - containerPort: 6717
@@ -56,12 +56,31 @@ spec:
               key: api-token
         volumeMounts:
         - name: edge-config-volume
-          # This is a path inside the container not the host
           mountPath: /etc/groundlight/edge-config
         - name: inference-deployment-template-volume
           mountPath: /etc/groundlight/inference-deployment 
         - name: model-repo
           mountPath: /mnt/models
+
+      - name: inference-model-updater
+        image: 723181461334.dkr.ecr.us-west-2.amazonaws.com/edge-endpoint:953306448-model-updater
+        imagePullPolicy: IfNotPresent
+        env:
+        - name: LOG_LEVEL
+          value: "INFO"
+        - name: DEPLOY_DETECTOR_LEVEL_INFERENCE
+          value: "True"
+        - name: GROUNDLIGHT_API_TOKEN
+          valueFrom:
+            secretKeyRef:
+              name: groundlight-secrets
+              key: api-token
+        volumeMounts:
+        - name: edge-config-volume
+          mountPath: /etc/groundlight/edge-config
+        - name: model-repo
+          mountPath: /mnt/models   
+
       imagePullSecrets:
       - name: registry-credentials
       volumes:

model_updater/update_models.py

@@ -0,0 +1,37 @@
+import os
+import logging
+import time
+from app.core.utils import load_edge_config
+from app.core.configs import RootEdgeConfig
+from app.core.edge_inference import EdgeInferenceManager
+
+log_level = os.environ.get("LOG_LEVEL", "INFO").upper()
+logging.basicConfig(level=log_level)
+
+
+def update_models():
+    if not os.environ.get("DEPLOY_DETECTOR_LEVEL_INFERENCE", None):
+        return
+
+    edge_config: RootEdgeConfig = load_edge_config()
+
+    edge_inference_templates = edge_config.local_inference_templates
+    inference_config = {
+        detector.detector_id: edge_inference_templates[detector.local_inference_template]
+        for detector in edge_config.detectors
+    }
+
+    edge_inference_manager = EdgeInferenceManager(config=edge_config.local_inference_templates, verbose=True)
+
+    for detector_id, inference_config in inference_config.items():
+        if inference_config.enabled:
+            try:
+                edge_inference_manager.update_model(detector_id=detector_id)
+            except Exception as e:
+                logging.error(f"Failed to update model for {detector_id}. {e}", exc_info=True)
+
+
+if __name__ == "__main__":
+    while True:
+        update_models()
+        time.sleep(3600)

pyproject.toml

@@ -21,6 +21,17 @@ tritonclient = {extras = ["all"], version = "2.36.0"}
 kubernetes = "^27.2.0"
 jinja2 = "^3.1.2"
 
+[tool.poetry.group.model_updater.dependencies]
+python = "^3.9"
+fastapi = "^0.88.0"
+pydantic = "^1.10.2"
+pillow = "^9.5.0"
+pyyaml = "^6.0"
+svix-ksuid = "^0.6.2"
+tritonclient = {extras = ["all"], version = "2.36.0"}
+jinja2 = "^3.1.2"
+requests = "^2.28.1"
+
 [tool.poetry.group.dev.dependencies]
 pytest = "^7.2.0"
 pytest-cov = "^4.0.0"