By name: redis-debug description: Debug and troubleshoot Redis usage in ettametta. Use when investigating cache issues, pub/sub problems, connection failures, broker connectivity, memory pressure, or key pattern anomalies.
Redis Debugging
Quick Diagnostics
redis-cli -p 7204 -a "$REDIS_PASSWORD" ping
redis-cli -p 7204 -a "$REDIS_PASSWORD" info memory
redis-cli -p 7204 -a "$REDIS_PASSWORD" info clients
redis-cli -p 7204 -a "$REDIS_PASSWORD" dbsize
redis-cli -p 7204 -a "$REDIS_PASSWORD" --scan --pattern "analytics:*" | head -20
redis-cli -p 7204 -a "$REDIS_PASSWORD" llen celery
Connection Config
- Image: redis:7-alpine, port 7204 (host) -> 6379 (container)
- URL:
redis://:${REDIS_PASSWORD}@redis:6379/0 - Always DB 0 — all subsystems share one database
- Celery broker, results, cache, security, chaos, GPU semaphores, pub/sub all on DB 0
Key Patterns and TTLs
| Key Pattern | TTL | Purpose |
|---|---|---|
| fastapi-cache:* | Default | Response cache |
| analytics:report:{post_id}:{user_id} | 1h | Analytics cache |
| optimization:viral_package:{id}:{niche}:{platform} | 1h | Viral package cache |
| discovery:trends:{niche}:{horizon}:{region} | None! | Trend cache (stale risk) |
| active_batch:{strategy} | 24h | Experiment cohort |
| EM_FEATURES:{niche} | 24h | Feature store |
| sentinel:security_logs | None (trim 1000) | Security log list |
| sentinel:security_health:audit:{date} | 7d | Audit results |
| security:requests:{client_ip} | 5m | Rate tracking |
| ettametta:gpu:slots | 7d | GPU semaphore |
| token_blacklist (set) | TTL applied | JWT blacklist (fixed 2026-05-28) |
Known Issues
No connection pooling / connection explosion
12+ separate redis.from_url() calls, each creating its own pool. No centralized manager.
Affected: OptimizationService, AnalyticsService, SecuritySentinel, GpuQueueManager, ConnectionManager, DistributedEventBus, GlobalFeatureStore, hot-reload, FastAPICache.
Sync Redis clients blocking event loop
Multiple async services use synchronous redis.Redis:
src/services/optimization/service.pyline 36src/services/analytics/service.pyline 30src/services/security/service.pyline 19src/services/discovery/service.pyline 213
Token blacklist TTL
SADD with TTL applied (fixed 2026-05-28). Check: redis-cli -p 7204 scard token_blacklist
Discovery cache has no TTL
discovery:trends:* keys may never expire.
localhost-to-redis hostname hack
settings.REDIS_URL.replace("//localhost", "//redis") — fragile.
No retry/circuit breaking on Redis ops
Unlike AI providers (tenacity + CircuitBreaker), Redis operations have no retry logic.
Blocking BLPOP in async context
GpuQueueManager uses sync self.redis.blpop() inside asynccontextmanager.
Redis Wrappers
| File | Purpose |
|---|---|
| src/api/utils/redis.py | Async singleton client |
| src/services/infrastructure/event_bus.py | Redis Streams with consumer groups, DLQ |
| src/api/routes/ws.py | PubSub + Streams for WebSocket broadcasting |
| src/services/infrastructure/global_feature_store.py | HMSET/HGETALL for features |
| src/services/video_engine/synthesis_service.py | BLPOP/RPUSH GPU semaphore |
Pub/Sub Channels
| Channel | Purpose |
|---|---|
| job_updates | Video job progress |
| system_logs | Real-time log broadcast |
| system_config_reload | Settings hot-reload |
Streams
| Stream | Consumer Group | Purpose |
|---|---|---|
| VF_FLOW_STREAMS | VF_COORDINATOR_GROUP | Distributed event bus |
| VF_FLOW_DLQ | — | Dead letter queue |
Debugging Checklist
redis-cli -p 7204 ping- Memory:
redis-cli info memory | grep used_memory_human - Connections:
redis-cli info clients | grep connected_clients - Key count:
redis-cli dbsize - Slow queries:
redis-cli slowlog get 10 - Celery queue:
redis-cli llen celery - Token blacklist size:
redis-cli scard token_blacklist