March 31, 2025

Observability and monitoring in distributed systems

This week, I tackled the essentials of observability and monitoring in distributed systems to ensure you can find issues when they occur.

March Theme: Distributed Systems

This week in our distributed systems series, we're exploring observability and monitoring – the practices that let us see inside our complex distributed systems and understand what's really happening.

So far this month, we've covered scalability, concurrency, fault tolerance, and geographic distribution. Now, let's tackle the crucial question: How do you know if your distributed system is healthy?

What is Observability?

Observability is your system's ability to expose its internal state in a way that helps you answer questions about what's happening inside. It's about making the inner workings visible.

While monitoring tells you when something is wrong, observability helps you figure out why it's wrong – even if you've never seen this particular failure before.

As Charity Majors (co-founder of Honeycomb, an observability startup) puts it: "Monitoring is for known-unknowns; observability is for unknown-unknowns."

The Three Pillars of Observability

1. Metrics

   • Numerical representations of data measured over time

   • Examples: CPU usage, request counts, error rates, latency percentiles

   • Strengths: Low overhead, great for dashboards and alerts

   • Limitations: Lose context and detail, pre-aggregated

2. Logs

   • Time-stamped records of discrete events

   • Examples: Application logs, system logs, audit logs

   • Strengths: Rich context, detailed information about specific events

   • Limitations: High volume, expensive to store, hard to correlate across services

3. Traces

   • Records of a request's journey through your distributed system

   • Shows the path and timing as a request travels across services

   • Strengths: Perfect for understanding cross-service interactions

   • Limitations: Requires instrumentation across all services

Unique Challenges in Distributed Systems

Monitoring distributed systems is fundamentally different from monitoring monoliths:

1. No Single Source of Truth

   • Each service has its own logs, metrics, and view of the world

   • Need to correlate events across many services

2. Complex Failure Modes

   • Problems often manifest in unexpected places

   • The service that's throwing errors isn't necessarily the one with the problem

3. Data Volume Explosion

   • More services = exponentially more logs and metrics

   • Need sampling, filtering, and aggregation strategies

4. Causality is Hard

   • Clock drift between machines makes ordering events difficult

   • "Did A cause B, or did they both happen because of C?"

Building an Effective Observability Strategy

1. Instrument Everything That Matters

   • The Four Golden Signals (from the Google SRE book):

     • Latency: How long it takes to serve a request

     • Traffic: How much demand is placed on your system

     • Errors: Rate of failed requests

     • Saturation: How "full" your system is

2. Propagate Context

   • Use correlation IDs across service boundaries

   • Include user IDs, session IDs, and request IDs in logs

3. Standardize Your Approach

   • Consistent logging formats across all services

   • Common metric naming conventions

   • Centralized collection and storage

Modern Observability Tools

The observability landscape has exploded in recent years:

1. Metrics Systems

   • Prometheus, Datadog, New Relic

   • Real-time dashboards, alerting, anomaly detection

2. Log Management

   • ELK Stack (Elasticsearch, Logstash, Kibana)

   • Splunk, Graylog, Loki

3. Distributed Tracing

   • Jaeger, Zipkin, AWS X-Ray

   • OpenTelemetry (open standard for telemetry data)

4. Full-Stack Observability

   • Honeycomb, Lightstep

Practical Observability Tips

1. Start Simple

   • Begin with basic metrics and expand

   • Get your logging basics right before pursuing complex solutions

2. Instrument at the Right Level

   • Too much data is as bad as too little

   • Focus on high-value services first

3. Design for Debugging

   • Add context to logs and traces that will help future-you

   • Log at boundaries (service entry/exit points)

4. Watch for Alert Fatigue

   • Too many alerts = ignored alerts

   • Alert on symptoms, not causes

   • Use severity levels appropriately

The Deep Dive

This week's deep dive recommendation is the free e-book "Distributed Systems Observability" by Cindy Sridharan. It provides excellent practical guidance on building observable distributed systems.

Final Thoughts

Observability isn't just a technical practice, it's a mindset. The best distributed systems are designed with observability in mind from day one, not bolted on afterward.

Remember: You can't fix what you can't see, and you can't improve what you can't measure. Invest in observability, and your future self will thank you when that 2 AM production incident hits.

What observability tools do you use and have you found most valuable in your distributed systems? Reply to this email, I'd love to hear your experiences!

Thank you!

If you made it this far, then thank you! Next week we will start on our April topic, Mentoring!

Here’s a silly web comic I made this week:

Have comments or questions about this newsletter? Or just want to be internet friends? Send me an email at brittany@balancedengineer.com or reach out on Bluesky or LinkedIn.

Thank you for subscribing. It would be incredibly helpful if you tell your friends about this newsletter if you like it! :)