prometheus（Prometheus 监控）

prometheus

Purpose

Prometheus is used for monitoring and alerting on metrics from various targets. It collects time-series data via HTTP pulls, stores it, and allows querying to trigger alerts.

When to Use

Use this skill when monitoring infrastructure, applications, or services in a DevOps/SRE environment. Apply it for real-time metrics collection, anomaly detection, or scaling decisions, such as tracking server health in Kubernetes clusters or alerting on high error rates in microservices.

Key Capabilities

• Metrics Collection: Scrapes HTTP endpoints using configurable jobs; specify targets in YAML config, e.g., scrape_configs: - job_name: 'node' static_configs: - targets: ['localhost:9100'].
• Querying: Use PromQL for data retrieval; example: query CPU usage with rate(node_cpu_seconds_total{mode="idle"}[5m]).
• Alerting: Define rules in YAML files to fire alerts; e.g., groups: - name: example rules: - alert: HighCPU usage: (avg by(instance) (rate(node_cpu_seconds_total{mode="system"}[5m])) > 0.8) for: 1m.
• Storage and Retention: Handles time-series data with configurable retention; set via --storage.tsdb.retention.time=15d flag.
• Federation: Aggregate metrics from multiple Prometheus instances for larger setups.

Usage Patterns

To monitor a target, start by creating a YAML config file (e.g., prometheus.yml) with scrape jobs. Run the Prometheus server with that config. For querying, use the built-in API or integrate with tools like Grafana. Always set up alerting rules early. If using in a container, mount the config volume and expose the web port (default 9090). For production, enable authentication by setting --web.external-url and using basic auth with env vars like $PROMETHEUS_AUTH_USER and $PROMETHEUS_AUTH_PASS.

Common Commands/API

• CLI Commands: Start server with prometheus --config.file=prometheus.yml --web.listen-address=":9090" --storage.tsdb.path="/prometheus". Reload config dynamically with curl -X POST http://localhost:9090/-/reload. Use promtool for testing rules: promtool check rules prometheus.rules.yml.
• API Endpoints: Query metrics via GET /api/v1/query with query params, e.g., curl "http://localhost:9090/api/v1/query?query=up". For range queries, use GET /api/v1/query_range?query=up&start=1630000000&end=1630003600&step=15s. If auth is required, include headers like Authorization: Bearer $PROMETHEUS_API_KEY.
• Code Snippets: // Simple Go client to query Prometheus client, err:= prometheus.NewClient(http.Client{}, "http://localhost:9090") result, err:= client.Query(context.Background(), "up", time.Now()) # Python example using prometheus-api-client from prometheus_api_client import MetricsList metrics = MetricsList('http://localhost:9090/api/v1/query?query=up') print(metrics)
• Config Formats: Use YAML for main config; example snippet: global: scrape_interval: 15s. For alert rules, use YAML arrays as shown in Key Capabilities.

Integration Notes

Integrate Prometheus with exporters (e.g., Node Exporter for system metrics) by adding scrape jobs in the config. For visualization, connect to Grafana by adding a Prometheus data source with URL like http://prometheus:9090 and auth via env var $GRAFANA_PROM_DS_KEY. To federate, configure remote write in YAML: remote_write: - url: 'http://federate-prometheus:9090/api/v1/write'. If using with Kubernetes, deploy via Helm charts and set up ServiceMonitors with labels. Always handle API keys via env vars, e.g., export them as $PROMETHEUS_API_KEY for secure access.

Error Handling

Common errors include config syntax issues (check with promtool check config prometheus.yml), scrape failures (verify targets and timeouts in config), or query errors (use API response codes like 422 for bad queries). To debug, enable logging with --log.level=debug and check logs for messages like "error scraping target". For API calls, handle HTTP errors: if status is 500, retry with exponential backoff; use code like:

if err != nil && strings.Contains(err.Error(), "context deadline exceeded") { log.Fatal("Scrape timeout; increase timeout in config") }

Validate PromQL queries with the /api/v1/query endpoint first. If authentication fails (e.g., 401), ensure env var $PROMETHEUS_AUTH_TOKEN is set and passed correctly.

Concrete Usage Examples

1. Monitor a Local Web Server: Create a prometheus.yml with: scrape_configs: - job_name: 'web' metrics_path: '/metrics' static_configs: - targets: ['localhost:8080']. Start Prometheus: prometheus --config.file=prometheus.yml. Query uptime: curl "http://localhost:9090/api/v1/query?query=up{job='web'}". This collects metrics every 15 seconds and allows alerting if the server goes down.
2. Set Up CPU Alert: Define a rule file: groups: - name: instance rules: - alert: HighCPU expr: avg by(instance) (rate(node_cpu_seconds_total{mode="system"}[5m])) > 0.9 for: 2m. Load it via config: add rule_files: ["alert.rules.yml"] to prometheus.yml. Run prometheus --config.file=prometheus.yml. Integrate with Alertmanager by adding: alerting: alertmanagers: - static_configs: - targets: ['localhost:9093']. This triggers alerts for high CPU and sends notifications.

Graph Relationships

• Belongs to cluster: devops-sre
• Tagged with: monitoring, metrics, alerting
• Related skills: (e.g., via tags) grafana for visualization, alertmanager for notifications