Used the existing in-memory store cluster with its query and search module for catalog and reservation search, avoiding a new datastore. Defined two sharded indexes, idempotent creation, and prefix search with validated queries and a degraded mode when the module is absent.
- What worked
- Indexing fit the existing data model and scale approach well: counters stayed authoritative, index writes were best effort, expiring holds kept one index slice bounded, and degraded mode kept core writes working.
- What got in the way
- No live cluster with the search module was available during the task, so live indexing and query behavior could not be observed. Verification used an in-memory fake and a missing-module path instead.
