Handle failures and retries

Map failure classes for GET /metrics: retryable vs fatal vs needs-human. Retries need budgets, jitter, and idempotency rules aligned to alerts require multi-window evidence; missing scrape ≠ healthy silence. The chapter’s signature failure — dashboard averages hide tenant Acme 40% error rate — must take a deliberate branch, not a generic catch-all.

Implement the degrade/rollback/refuse behavior SRE watching error budget during a model bump needs when MON-TENANT-BLIND-9 repeats. Prefer scoped controls (one flag, one weight, one tenant, one secret version) over fleet-wide restarts. Preserve evidence; do not delete logs to “clean the demo.”

After containment, check alert_precision on game-day ≥ 0.9 and scrape_up == 1 moves in the safe direction and watch for retry amplification. Write the stop condition that ends the incident response for this lab.

Chaos note: inject only one fault class at a time and restore fixtures after. Watch for dual failures — dependency down and retry amplifier — which is how dashboard averages hide tenant Acme 40% error rate becomes an outage. Customer communication templates (even if only for the drill) beat silence. If you queue deferred work, define poison-message handling. Budget documents should state the maximum extra spend allowed during retries. Close the loop by linking the containment action to a dashboard panel for alert_precision on game-day ≥ 0.9 and scrape_up == 1.

Draw a state diagram for degrade modes and put it in the repo as ASCII if needed. Cap concurrent retries across the process, not only per request. Ensure cancellation propagates to downstream HTTP clients. When failing closed, choose a user-visible message that does not leak internals. Practice the single command that flips the kill switch or weight to zero. After recovery, drain or inspect deferred work before declaring green. In this chapter the product is Prometheus+Grafana board for AI answer API golden signals, the human stakeholder is SRE watching error budget during a model bump, and the incident id you design against is MON-TENANT-BLIND-9. Re-state the oracle in your notes — inject 5% 500s for 10m → Alert AnswerErrorBurn fires; recovery clears within 15m — and keep the invariant visible: alerts require multi-window evidence; missing scrape ≠ healthy silence. Track alert_precision on game-day ≥ 0.9 and scrape_up == 1 as the scoreboard. Surface under change control: GET /metrics. If you only have forty minutes, finish the fixture for dashboard averages hide tenant Acme 40% error rate before polishing UI. Promotion language stays ternary: promote, hold, or roll back based on evidence, not hope.