Metrics Exporter for Windows CRI Pods

Intended to be run in a container, the exporter takes a list servers (Windows Kubernetes nodes) from each it extracts the following for each pod and exports them (to be scraped by Prometheus)

• Pod name
• CPU
• Memory
• Disk usage
• Tx
• Rx

Environment Variables

---

The following env vars are required

• USER="administrator" (a user with crictl permissions)

• PASSWORD="XXXXXXX"

• LOG_LEVEL="DEBUG" (DEBUG, INFO, WARN, ERROR)

• SERVER_FILE="servers.txt" (can be called anything, one server/windows kubernetes node per line)

Build

---

docker build -t windows-cri-metrics -f Dockerfile .

Run

---

 docker run -d -p 8000:8000 --name windows-cri-metrics \
        -v $(pwd)/servers.txt:/home/metrics/servers.txt \
        -e LOG_LEVEL=DEBUG \
        -e USER="administrator" \
        -e PASSWORD="XXXXX" \
        -e SERVERS_FILE="/home/metrics/servers.txt" \
        windows-cri-metrics

Metrics

---

In Prometheus the metrics are called

• pod_cpu_usage_mcores
• pod_memory_usage_mb
• pod_disk_usage_mb
• pod_tx_mb
• pod_rx_mb

Grafana

---

If using Grafana query for the metrics are called

• pod_cpu_usage_mcores
• pod_memory_usage_mb
• pod_disk_usage_mb
• pod_tx_mb
• pod_rx_mb

Use a custom legend e.g. {{pod}} - CPU (for pod_cpu_usage_mcores) to label the query.

This will update dynamically as pods are created and destroy.

Limitations

---

All the nodes must have the same credentials, PSRemoting must be enabled and working for WSMan, though the exporter communicates with Windows nodes, it run on Linux

Files

---

windows_metrics.py - Contacts to every Windows node, extracts pod stats, parses the metrics and exposes them for Prometheus.

config.py - reads the environment variables.

metricslogger.py - stdout logging.

requirements.txt - PIP requirements.

servers.txt - sample server list.

Dockerfile - build file.