监控、告警与日志记录

Ever wonder how big systems stay up 24/7? It's thanks to good observability! This lesson covers three pillars: Monitoring, Alerting, and Logging.

These practices let us understand a system's internal state from its external outputs. They're crucial for high availability and reliability.

Monitoring is the continuous process of collecting and analyzing data about your system's performance and health.

• It provides real-time insights into how your applications and infrastructure are behaving.
• Think of it as the system's vital signs, showing you if everything is running smoothly or if there's a problem brewing.

What should you monitor? Key metrics fall into a few categories:

• System Metrics: CPU usage, memory, disk I/O, network traffic.
• Application Metrics: Request rates, error rates, latency, active users, queue sizes.

Focus on metrics that indicate performance bottlenecks or potential failures.

Many tools help collect and visualize metrics. Some popular ones include:

• Prometheus: An open-source system that collects metrics from configured targets.
• Grafana: Often paired with Prometheus, it's a powerful tool for creating dashboards and visualizing data.
• Datadog/New Relic: Commercial solutions offering end-to-end monitoring for infrastructure, applications, and logs.

While monitoring shows you what's happening, alerting tells you when something is wrong, and often, who needs to know.

• An alert is a notification triggered when a monitored metric crosses a predefined threshold or behaves abnormally.
• It helps engineering teams react quickly to issues, minimizing downtime and impact.

Designing good alerts prevents "alert fatigue" (too many non-critical alerts).

• Thresholds: "CPU usage > 90% for 5 minutes."
• Baselines: Alert if traffic drops significantly below the usual pattern.
• Severity Levels: Critical, Warning, Info. Different issues need different urgency.

Aim for actionable alerts that provide enough context to start troubleshooting.

Once an alert is triggered, it needs to reach the right people. Common channels include:

• Email/SMS: For less critical alerts or initial notifications.
• Slack/Teams: For team-wide awareness and discussion.
• PagerDuty/Opsgenie: For critical, on-call alerts that require immediate attention, often with escalation policies.

Escalation ensures that if the first responder doesn't acknowledge an alert, it's sent to the next person in line.

Logging is the recording of events that happen within your system. These events can be anything from user actions to system errors.

• Logs provide detailed context for debugging and understanding what went wrong (or right!).
• They're like a historical record of your system's behavior.

Example: INFO: User 'john.doe' logged in from IP 192.168.1.100

As systems grow, collecting logs from many servers becomes essential. Centralized logging aggregates logs from all parts of your system into one place.

• This makes searching, analyzing, and correlating events across services much easier.
• Popular stacks include the ELK stack (Elasticsearch, Logstash, Kibana) or commercial solutions like Splunk.

Ensuring high availability means proactively identifying and addressing issues. Which of the following would be the MOST effective proactive measure to detect a potential system performance issue?

You've learned that Monitoring, Alerting, and Logging (MAL) are fundamental for building and maintaining highly available and reliable systems.

• Monitoring provides visibility into system health.
• Alerting notifies teams of critical issues.
• Logging offers detailed historical context for debugging.

Mastering these ensures you can quickly detect, diagnose, and resolve problems, keeping your systems running smoothly!