# Monitoring Plan — frontend-shell Decisions: [monitoring-setup-plan.md](../plans/monitoring-setup-plan.md) (Q1 = A, Q2 = D, Q3 = A, Q4 = A, Q5 = A, Q6 = A + production now, Q7 = A). ## Overview ```mermaid graph LR Visitor["Visitor browser"] -->|"JS errors"| Sentry["Sentry (EU), existing org"] Visitor -->|"page views, cookieless"| Umami["Umami on the Pi, analytics.slpsoftware.nl"] Uptime["UptimeRobot"] -->|"HTTP checks"| Site["andreaslop.nl and test.andreaslop.nl"] Sentry -->|"e-mail, new error type"| Owner["Owner"] Uptime -->|"e-mail, site down"| Owner classDef visitor fill:#f6e05e,stroke:#c05621,stroke-width:1px,color:#000; classDef service fill:#e9d5ff,stroke:#6b21a8,stroke-width:1px,color:#000; classDef site fill:#9ae6b4,stroke:#2f855a,stroke-width:1px,color:#000; classDef owner fill:#63b3ed,stroke:#2b6cb0,stroke-width:1px,color:#000; class Visitor visitor; class Sentry,Umami,Uptime service; class Site site; class Owner owner; ``` Text alternative: visitors' browsers send JavaScript errors to Sentry and cookieless page views to Umami; UptimeRobot checks both sites from outside; Sentry and UptimeRobot e-mail the owner. ## Chosen Approach (all three, reusing existing services) | Area | Service | What it covers | Code needed | |---|---|---|---| | **Logging** | **Sentry** (existing EU organisation, new project `andreaslop-fotografie`) | JavaScript errors in visitors' browsers: render errors caught by `ErrorPage` and unhandled errors, tagged `test` or `production` | Yes | | **Dashboards: visitors** | **Umami** (existing install on the Pi, two new "websites": test and production) | Page views, referrers, devices, without cookies or personal data | Yes | | **Dashboards: availability** | **UptimeRobot** (existing account, new monitors) | Whether test and production respond; response time history | No | | **Alerting** | Sentry + UptimeRobot defaults | E-mail on a **new** type of error (Sentry) and when a site is **down** (UptimeRobot) | No | ## Key Decisions | # | Decision | Why | |---|---|---| | M1 | **One build for both environments**: the browser picks the environment from the hostname (`andreaslop.nl` → production, `test.andreaslop.nl` → test). Any other host (localhost, Lighthouse CI, previews) sends **nothing**. | Keeps "deploy exactly what was tested" (D-pipeline). Local runs and CI never pollute the dashboards. | | M2 | **Sentry and Umami settings live in code** (`src/monitoring/config.ts`), not in Gitea variables. | The Sentry DSN and Umami website IDs are public by design: they end up in every visitor's browser anyway. In code they are reviewed, tested and versioned. | | M3 | **Sentry sends directly** to `*.ingest.de.sentry.io` (no tunnel). | Q4 = A: the IIS host serves only static files until the .NET API exists; some errors from visitors with ad-blockers are missed. Follow-up: tunnel via the API. | | M4 | **Sentry loads after the page**, not in the main bundle. | Keeps the performance budget: the home page on mobile already sits just above its 0.8 target. Errors are queued until Sentry is ready. | | M5 | **Sentry: errors only.** No performance tracing, no session replay, no personal data (`sendDefaultPii: false`, IP storage off in the project settings). | Smallest bundle, least data, simplest privacy story. | | M6 | **Umami: separate websites for test and production**; `data-domains` restricts counting to the right hostname, and `data-do-not-track` respects the browser's Do Not Track setting. | Test traffic never mixes with real numbers. | | M7 | **CSP exceptions as narrow as possible**, generated into `web.config` and the nginx config like the other headers.