如何直接向Cortex推送指标？是否有Python客户端支持该操作？

Absolutely! You can skip Prometheus entirely and send metrics straight to Cortex—here’s how to do it, including Python tools to streamline the process:

1. Use Cortex's Built-in Remote Write API #

Cortex is designed to work with Prometheus’s Remote Write API out of the box, which is exactly what you need for direct metric ingestion. The core endpoint to target is:
http://<your-cortex-instance>/api/v1/push

Quick Test with curl #

To verify this works, you can send a sample metric via curl (note: the payload needs to be in Prometheus’s Protobuf format):

# First, use promtool to convert a text metric to Protobuf (you'll need promtool installed)
echo "test_counter_total{env=\"dev\"} 123" | promtool remote-write encode > metric.pb

# Then send it to Cortex
curl -X POST http://your-cortex-url/api/v1/push \
  -H "Content-Type: application/x-protobuf" \
  -H "X-Prometheus-Remote-Write-Version: 0.1.0" \
  --data-binary @metric.pb

2. Python Implementation Options #

You’ve got a couple solid choices for Python clients—here are the most straightforward ones:

Option 1: Official prometheus-client + requests #

The official Prometheus Python library lets you create metrics easily, and you can pair it with requests to send them to Cortex. Here’s a working example:

import requests
from prometheus_client import Counter, generate_latest
from prometheus_client.parser import text_string_to_metric_families
import prometheus_client.remote_write as remote_write

# Create a sample counter metric
test_counter = Counter('api_requests_total', 'Total API requests', ['endpoint'])
test_counter.labels(endpoint='/checkout').inc()

# Convert the metric to Protobuf (required for Remote Write)
metric_families = list(text_string_to_metric_families(generate_latest().decode('utf-8')))
write_request = remote_write.build_write_request(metric_families)

# Send to Cortex
cortex_url = "http://your-cortex-url/api/v1/push"
headers = {
    "Content-Type": "application/x-protobuf",
    "X-Prometheus-Remote-Write-Version": "0.1.0"
    # Add auth headers here if needed, e.g., "Authorization": "Bearer YOUR_TOKEN"
}

response = requests.post(cortex_url, data=write_request.SerializeToString(), headers=headers)
response.raise_for_status()  # Raises an error if the push fails
print("Metrics pushed successfully to Cortex!")

Option 2: Third-Party prometheus-remote-write Library #

If you want a more streamlined approach, the prometheus-remote-write library handles the Protobuf encoding and API calls for you. First install it:

pip install prometheus-remote-write

Then use it in your code:

from prometheus_remote_write import Client, Metric

# Initialize the Cortex client
client = Client("http://your-cortex-url/api/v1/push")

# Create and send a gauge metric
gauge_metric = Metric(
    name="system_cpu_usage",
    labels={"host": "web-server-01"},
    value=78.2,
    type="gauge"
)
client.send(gauge_metric)

# You can also send multiple metrics at once
another_metric = Metric(
    name="disk_free_gb",
    labels={"host": "web-server-01"},
    value=45.6,
    type="gauge"
)
client.send([gauge_metric, another_metric])

3. Key Things to Keep in Mind #

• Authentication: If your Cortex deployment uses API keys, OAuth, or other auth methods, add the corresponding headers to your requests (e.g., Authorization: Bearer <your-token>).
• Metric Format: Stick to Prometheus naming rules (alphanumeric characters and underscores only) to avoid ingestion errors.
• Rate Limits: Check your Cortex config for rate limits—don’t flood the endpoint with too many metrics at once.

内容的提问来源于stack exchange，提问作者neastlia