能否通过触发器触发Azure容器实例创建？求实现方案与示例

Absolutely, this scenario is fully supported on Azure—let’s break down how to build this event-driven workflow step by step, including practical code examples and key considerations:

You’ll be building an event-driven serverless container workflow with three key components:

• Event Source: Cosmos DB (or Blob Storage) to trigger workflows when new entries are added
• Event Router: Azure Event Grid to capture those change events and kick off container instances
• Execution Unit: Azure Container Instances (ACI) to run your legacy app alongside a lightweight proxy script

Cosmos DB natively pushes change events (create/update/delete) to Azure Event Grid. Here’s how to tie that to ACI:

• First, enable an Event Subscription for your Cosmos DB account. Target either a Logic App or Azure Function—Logic Apps are great for no-code/low-code ACI deployment, while Functions give you more code control.
• If using Logic Apps:
  1. Create a new Logic App and use the "When a resource event occurs" trigger, configured to listen for Microsoft.DocumentDB.DocumentCreated events (for new entries).
  2. Add the "Create container group" action (Azure Container Instances connector). Configure your container image, CPU/memory resources (match your legacy app’s needs), and pass critical config as environment variables (like Cosmos DB connection string, target document ID).

Your container needs a small proxy script to handle data fetching, app execution, and result push-back. Below is a Python example (easily adapted to Shell/Node.js):

Proxy Script (proxy.py) #

import os
import json
from azure.cosmos import CosmosClient

# Pull config from environment variables (set when creating ACI)
cosmos_conn_str = os.getenv("COSMOS_CONN_STR")
db_name = os.getenv("DATABASE_NAME")
container_name = os.getenv("CONTAINER_NAME")
target_doc_id = os.getenv("TARGET_DOC_ID")

# Initialize Cosmos client
client = CosmosClient.from_connection_string(cosmos_conn_str)
db = client.get_database_client(db_name)
cosmos_container = db.get_container_client(container_name)

# 1. Fetch data from Cosmos DB
target_doc = cosmos_container.read_item(item=target_doc_id, partition_key=target_doc_id)
with open("/app/input/data.json", "w") as f:
    json.dump(target_doc, f)

# 2. Run your legacy application (adjust command to match your app's syntax)
os.system("/app/legacy-app --input /app/input/data.json --output /app/output/result.json")

# 3. Push results back to Cosmos DB
with open("/app/output/result.json", "r") as f:
    analysis_result = json.load(f)

# Update the original document or create a new one
target_doc["analysis_result"] = analysis_result
cosmos_container.upsert_item(target_doc)

print("Analysis completed successfully!")

Dockerfile Example #

# Use a base image compatible with your legacy app (Ubuntu shown here)
FROM ubuntu:22.04

# Install dependencies (Python + Cosmos SDK, plus any legacy app requirements)
RUN apt-get update && apt-get install -y python3 python3-pip
RUN pip3 install azure-cosmos

# Copy your legacy app, proxy script, and working directories
COPY legacy-app /app/legacy-app
COPY proxy.py /app/proxy.py
RUN mkdir -p /app/input /app/output

# Set working directory and default command
WORKDIR /app
CMD ["python3", "proxy.py"]

If you prefer Blob Storage over Cosmos DB, the workflow is nearly identical:

• Listen for Microsoft.Storage.BlobCreated events (trigger when a new blob is uploaded)
• Modify the proxy script to download the blob via Azure Storage SDK, run the legacy app, then upload the result blob.

• Resource Quotas: ACI supports max runtimes of 7 days, which easily covers your 10-minute test. Configure CPU/memory to match your legacy app’s needs to avoid performance issues.
• Permissions: Ensure your Logic App/Function has permissions to create ACI instances, and the container uses either a managed identity or connection string to access your storage service.
• Cost Efficiency: ACI bills by the second, so you only pay for compute time when containers are running—perfect for on-demand, short-lived workloads like this.

内容的提问来源于stack exchange，提问作者dommer