observability-baseline

Observability Baseline

The floor: every Rails 8 production app needs structured logs, request correlation, error tracking with PII scrubbing, and clear what-to-log rules. Deeper observability (APM, distributed tracing, custom metrics) is v0.3 work. This skill makes the floor real.

Why this matters

When something breaks at 2am, you have three questions: did it actually break (logs), how often (errors), and why (context). Default Rails logging answers none of these well — it's multiline, unstructured, missing user context. This skill replaces the default with the senior-Rails-dev defaults that scale.

The opinion

lograge for single-line structured logs. Tag every log with request_id + user_id. Sentry for errors (Honeybadger / Rollbar are equivalent — pick one; not all three). Use Rails.error.report (Rails 7.1+) for non-fatal errors instead of inline logging. PII scrubbing on filter_parameters (mirrored in error tracker config). Never log: passwords, card data, JWT contents, full SSN, full DOB. Log structured fields, not concatenated strings.

Counter-positions:

• No error tracker — fine for tiny pre-revenue apps. Sentry's free tier is generous; the trade-off is tiny.
• OpenTelemetry from day one — heavy investment for limited return on a small app. Stage it: lograge + Sentry first, OTel when you have multiple services.

Core patterns

Pattern 1: lograge — structured single-line logs

Before (Rails default, hard to parse):

I, [2026-05-24T10:00:00.123]  INFO -- : Started GET "/posts/42" for 192.0.2.1 at ...
I, [2026-05-24T10:00:00.124]  INFO -- : Processing by PostsController#show as HTML
I, [2026-05-24T10:00:00.125]  INFO -- :   Parameters: {"id" => "42"}
I, [2026-05-24T10:00:00.150]  INFO -- :   Rendered post/show.html.erb (Duration: 12.0ms | Allocations: 1234)
I, [2026-05-24T10:00:00.155]  INFO -- : Completed 200 OK in 30ms (Views: 12.0ms | ActiveRecord: 8.0ms)

Six lines per request. Multiline logs are painful to grep, hard to ship, and impossible to query at scale.