Redis Cache Architect

Redis design for category listings: cache product data and listing IDs separately, invalidate on ERP updates, jitter TTLs and guard against stampedes.

Cacheability assessment

Invalidation

• ERP sync job deletes prod:{id}:live for every changed product after the DB commit.
• Product create, delete or recategorize deletes cat:{id}:* via a set of page keys per category (no KEYS scans).
• Static data uses TTL plus explicit delete on admin edit.

TTL jitter

const ttl = base + Math.floor(Math.random() * base * 0.1);

Spreads expiry so a whole category does not expire together.

Stampede protection

const lock = await redis.set(`lock:${key}`, 1, 'NX', 'PX', 3000);
if (!lock) return staleOrWait(key);

• One request rebuilds; others serve the stale value or wait briefly.
• Build pages with MGET on product keys to avoid N round trips.

Next steps

• Measure cache hit rate and rebuild time per key type after rollout.
• Decide the eviction policy (allkeys-lru suits this pattern).