Atriumstatus

master9da7fa5

← all systems

Upstash Redis

Operationalhigh

The durable store behind this dashboard — every check run, uptime figure and incident.

checked 2m ago· owner AtriumConsoleStatus page

Checks

REST API reachable

Check history, uptime and incidents stop being recorded.

100%

24h

100%

30d

GET https://api.upstash.com/v2/redis/databasesexpects 401 or 403last 401 · 107ms · 2m ago

Vendor status

Upstash is reporting a problem on their side before it reaches us.

100%

24h

100%

30d

GET https://status.upstash.com/api/v2/status.jsonexpects 200last 200 · 148ms · 2m ago

What it is

Serverless Redis over HTTP. It holds everything this app knows: the run history each uptime percentage is recomputed from, the open and closed incidents, and the short lock that stops ten simultaneous page views from firing ten sweeps.

It was the one load-bearing dependency in the estate with no monitor of its own, which is a strange blind spot for a status board: the store going away does not turn anything red, it quietly turns the history into fiction.

What breaks without it

lib/health/store.ts falls back to an in-memory store rather than failing. The app keeps sweeping and the dashboard keeps rendering — but on serverless every invocation gets its own memory, so nothing accumulates. Uptime and incidents become noise shaped like data.

That fallback is deliberate (a missing credential should not take the board down) and it is why the dashboard shows a panel whenever the store is not durable. Read that panel as an outage of this system even when every monitor here is green.

Why the checks look indirect

Neither monitor here touches our database, and that is on purpose.

The REST API is probed without credentials. A 401 proves Upstash is answering; sending a real token from a status page would put a live credential in an outbound request on every sweep for no extra information.

The status feed is Upstash's own view of Upstash, which catches the case the liveness check cannot: the API answering while the service behind it is degraded.

When it breaks

Symptom

The dashboard shows the in-memory panel

Check

UPSTASH_REDIS_REST_URL and UPSTASH_REDIS_REST_TOKEN in the Vercel project.

Fix

Set both and redeploy. History before that point is gone; it was never written.

Symptom

Liveness is down but the status feed says operational

Check

Whether the outage is ours — network egress from Vercel, or a wrong region.

Fix

Retry from another network before escalating to the vendor.

Symptom

Uptime numbers jump around

Check

Whether the store has been flipping between Upstash and memory across deploys.

Fix

Confirm the credentials are set for every environment, not just production.

Incident history

No incidents recorded for this system.