Skip to content

4 · Redis data & TTL

Everything the API stores lives in exactly one place: a single Redis process.

Key lifecycle

Rules that will bite you

  • Values are strings. The API does String(value): 42 → "42", null → "null", {} → "[object Object]". Send strings.
  • TTL is per-key seconds via ttlSeconds → SET … EX n.
  • Re-PUT without ttlSeconds clears the expiry. Plain SET semantics — the TTL is gone, the key lives forever.
  • ttlSeconds: 0 means no TTL (JS truthiness check, not validation). A truthy non-number like "abc" can explode in the Redis layer → 503.
  • GET /api/kv runs KEYS *. Fine here; in a real system it blocks Redis on big datasets. Production answer: SCAN.
  • Keys are URL path params — URL-encode anything special.
  • DELETE is idempotent-ish: {"deleted":0} for a key that isn't there.

Ways a key disappears

Cause Visible how
ttlSeconds expiry GET → 404 after n seconds
DELETE {"deleted":1}
Re-PUT… no, that keeps the key but kills the TTL TTL silently gone
Redis restart / container replace everything → 404

That last row is the big one: no appendonly, no RDB volume, no replica — in Compose and in the k8s manifests alike.

What this project does not configure

Known gaps

  • No maxmemory / eviction policy → Redis fills up and writes start failing
  • No password, ACL, or TLS on redis://
  • No command timeout — requests queue during Redis outages instead of failing fast
  • REDIS_URL is printed in the startup log — don't put credentials in it

Full checklist in 9 · Gotchas & limits.


Next: 5 · Health vs readiness