Opinionated first-SLO pack for Phoenix SaaS teams.

Register in one line:

config :parapet, providers: [Parapet.SLO.StarterPack.WebSaaS]

Ships three slices with documented default objectives, each pinned to the real Prometheus series this codebase emits. These are opinionated, inspectable, and overridable defaults — not auto-generated targets.

Slices

HTTP Availability (web_saas_http_availability)

• Metric: parapet_http_request_count — the real HTTP counter emitted by Parapet's plug (parapet.http.request.count → Prometheus underscores).
• Objective: 99.5% — approximately 3.65 hours/month of non-2xx/3xx budget.
• Alert class: :ticket — aggregate HTTP errors are noisy on first adoption; ticket-level keeps alert fatigue low. Upgrade to :page once you understand your actual traffic patterns.
• Matches: status_class label (values "2xx", "3xx", "4xx", "5xx"). The plug emits status_class as a low-cardinality tag; use it for label selectors.

Login Journey (web_saas_login_journey)

• Metric: parapet_journey_login_count — the Sigra integration counter (parapet.journey.login.count → Prometheus underscores).
• Objective: 99.9% — approximately 43 minutes/month of user-impacting auth failures.
• Alert class: :page — auth failures are silent revenue loss; worth paging. Login failures are low-volume and directly user-impacting.
• Matches: outcome label (:success / :failure).

Oban Job Success (web_saas_oban_job_success)

• Metric: parapet_oban_jobs_total — the Oban integration counter (parapet.oban.jobs.total → Prometheus underscores).
• Objective: 99.0% — approximately 7.3 hours/month of job-level failures.
• Alert class: :ticket — jobs include retries, so transient failures are expected; 99.0% accommodates retry-normal patterns. Adopters with critical-path jobs (e.g. payment processing) should override to 99.9% + :page.
• Matches: state label (values "success", "failure", "cancelled", "discarded").

Low-Traffic Safety

All slices use the default min_total_rate: 0.01 (~0.6 requests/minute over a 1-hour window), which the Generator renders as ... and <total_rate_record> > 0.01 in every alert expression. This prevents alert flapping on low-traffic services. Zero Generator changes are required.