Observability for a 5-person team — Sentry, Grafana, plain logs

A 5-engineer team running production needs observability. Datadog or New Relic would solve it — for $30K-100K/year and ongoing tuning. There's a lighter stack that covers 90% of needs for under $200/month and requires no dedicated ownership.

The three pillars, applied small

1. Errors — Sentry

Sentry catches application errors, groups by signature, alerts. For a 5-person team:

• Free tier covers ~5K errors/month — enough for small SaaS.
• Team plan ($26/user/month) scales without surprises.
• Integrations with everything (Slack, Linear, GitHub).
• Release tracking ties errors to deployments.
• Source maps for client-side errors.

Setup: one SDK install, one DSN per environment. Done in 30 minutes.

2. Metrics and dashboards — Grafana Cloud

Free tier includes 10K metrics, 50GB logs, 50GB traces. Generous for small SaaS.

Use:

• Prometheus exporters for app and infrastructure metrics.
• Loki for log aggregation.
• Tempo for traces (when needed).
• Built-in dashboards for common services.

Setup: 1-2 days for instrumentation + dashboards.

3. Logs — structured JSON to Loki or BetterStack

Every log line as JSON with consistent fields:

{
  "level": "error",
  "time": "2026-06-12T15:23:01Z",
  "service": "api",
  "event": "payment_failed",
  "user_id": "u-42",
  "order_id": "o-9876",
  "error": "insufficient_funds",
  "trace_id": "abc123"
}

Searchable, parseable, joins with metrics via trace_id.

The minimum useful set of metrics

• Request rate per endpoint.
• Error rate per endpoint (4xx separated from 5xx).
• Latency p50, p95, p99 per endpoint.
• Database query time p95.
• External API call latency + error rate per provider.
• Background job queue depth and processing time.
• Infrastructure — CPU, memory, disk I/O on each instance.

Resist the urge to instrument everything. Add metrics when they answer a specific question.

The minimum useful set of dashboards

• Overall health. Request rate, error rate, p95 latency across services.
• Per-service deep dive. Same metrics broken down by endpoint.
• External dependencies. Latency and error rates for third-party APIs.
• Infrastructure. CPU/memory/disk on each instance.
• Business metrics. Signups, conversions, revenue (per hour/day).

Five dashboards. Each one fits on a screen. Each has one purpose.

Alerts that don't burn out on-call

Alert only on issues that require immediate action:

• Error rate >5% sustained for 5 minutes.
• p95 latency >2x baseline for 5 minutes.
• Disk space >90%.
• Database connection pool exhaustion.
• Background queue not draining.
• Payment processor down.

Page-worthy alerts: 5-15. Anything else is a dashboard signal, not a page.

Tracing — when you need it

Tracing is powerful but expensive in time. Add it when:

• You can't explain latency from metrics alone.
• Multi-service requests are common.
• Customer-reported issues need fine-grained debugging.

Skip it when:

• You have one service.
• Logs with trace_id correlation are enough.
• Cost vs benefit doesn't pencil.

On-call rotation for small teams

• 2-week rotation per engineer.
• Weekly handoff Mondays.
• Runbook for top 10 alert types — what to check, who to escalate to.
• Postmortem after every page (lightweight, 1-page max).

Tools: PagerDuty free tier (3 users), incident.io free tier.

What to skip

• Full OpenTelemetry pipeline. Overkill for small team.
• Dedicated monitoring engineer. Distributed responsibility works at this scale.
• Enterprise APM ($30K+/year). Sentry + Grafana cover 90% for 1% the cost.
• Synthetic monitoring at scale. 1-2 endpoint checks from Pingdom or BetterStack is enough.
• SLI/SLO frameworks. Useful at scale. Premature for small team.

Cost

For a 5-engineer team running typical SaaS:

• Sentry Team plan: $130/month (5 users).
• Grafana Cloud Free or Pro: $0-150/month.
• PagerDuty free.
• BetterStack uptime: $20-40/month.

Total: $150-320/month. Versus $2,500-8,000/month for Datadog at equivalent coverage.

Verdict

Small teams don't need enterprise observability. Sentry for errors, Grafana for metrics/dashboards, structured JSON logs to Loki, lean alerting. Setup in a week, $150-320/month, covers 90% of needs. Add tracing and SLOs when you've outgrown this — usually at 30+ engineers, not 5.