The Observability Stack I Run in 2025

For early-stage companies, modern observability is over-priced for what most teams need. Here's a leaner stack that gets 90% of the value at 20% of the cost.

Datadog will charge you $20K/month before you blink. For a 5-person startup, that's a hire. Here's what I run instead.

The minimum viable observability stack

• OpenTelemetry SDK in all services. This is non-negotiable. The vendor-neutral instrumentation means you can swap backends.
• Grafana Cloud free tier OR self-hosted Grafana + Loki + Tempo + Mimir on a single box.
• Sentry for application errors. The free tier covers small teams.
• Better Stack for uptime + status page. Ten bucks a month.

That stack handles:

• Distributed tracing (Tempo via OTel)
• Logs (Loki via OTel)
• Metrics (Mimir/Prometheus via OTel)
• Errors (Sentry)
• Uptime (Better Stack)

Total monthly cost for a small team: under $200.

When to upgrade

Migrate to Datadog or Honeycomb when:

• Volume exceeds free-tier limits (you'll know)
• You need APM features beyond what OSS offers (rare, but real for some workloads)
• You have a dedicated platform team and the time savings justify the cost

My standard service template

Every Spring Boot service I ship now starts with:

1. OTel auto-instrumentation
2. Structured JSON logging with correlation IDs
3. RED metrics (rate, errors, duration) for every endpoint
4. Health and readiness endpoints
5. /metrics Prometheus scrape endpoint
6. Sentry for unhandled exceptions

This costs maybe 4 hours to set up the first time. After that, it's a template.

What I look at every Monday

• p95 / p99 latency by endpoint, week over week
• Error rate trend
• Active alerts that fired in the last 7 days
• Top 5 slow queries (Postgres pg_stat_statements)

15 minutes. Catches 80% of issues before customers do.

References

• OpenTelemetry
• Grafana OSS
• Sentry
• Better Stack