Skip to content

Postmortem: Catalog endpoint latency spike

Blameless. This document explains what happened and what we’re changing — it does not assign fault to a person.

A schema-mismatched query shipped in a routine deploy caused catalog lookups to run without an index for ~14 minutes before rollback.

Slow catalog refreshes in Tidepool and the web app; no failed requests, no data loss.

None beyond the response time of the on-call engineer.

See INC-0001 — not duplicated here.

A new query shape in the catalog handler wasn’t covered by the index the old query relied on implicitly.

The latency alert fired within 2 minutes of the regression, and the Rollback Runbook resolved it in under 15.

Nothing in CI caught a query-shape change that would degrade performance under a missing index — this class of regression isn’t covered by existing tests.

ActionOwnerDueStatus
Add the missing indexplatformexample
Add a load test for catalog query shape changesplatformexample
Add a query-plan check to CI for the catalog handlerplatformexample

Platform team owns follow-through on the corrective actions above.