Compare Scenarios

Migration Step 1

Both scenarios define a migration step at this stage. ML Feature Serving Platform: triggered by 'Feature computation logic duplicated across 3+ model serving'. Social Feed Platform: triggered by 'PostgreSQL timeline read query p99 > 500ms; query plan for "'.

Migration Step 2

Both scenarios define a migration step at this stage. ML Feature Serving Platform: triggered by 'Regulatory or reproducibility requirement to re-run model pr'. Social Feed Platform: triggered by 'Fan-out worker queue lag during celebrity post events > 5 mi'.

Migration Step 3

Both scenarios define a migration step at this stage. ML Feature Serving Platform: triggered by 'Semantic similarity search recall insufficient from BM25 ful'. Social Feed Platform: triggered by 'Redis memory utilization approaching 80% of a single node; R'.

Feature store p99 rising from < 5ms to > 50ms immediately following a model deployment or pod scaling event; Cassandra or PostgreSQL feature store read QPS spiking 10–100x simultaneously with serving fleet restart; feature store connection pool exhaustion visible in application logs during the cold start window; Redis cache hit rate dropping to < 10% for the first 60 seconds after deployment

Tier 1: Cache Cold Start Latency Spike: Simultaneous cache cold start across all serving pods after deployment, converting sequential feature requests into a thundering herd against the backing feature store. Recommended evolution: Implement pre-warming: before a new model version receives traffic, a warm-up job runs inference requests for a representative sample of entity IDs, populating the Redis cache before the pod enters the serving fleet. Use probabilistic early cache refresh (fetch from backing store before TTL expiry when remaining TTL < 20% under high request rate) to prevent simultaneous TTL expiry on hot features. Stagger deployment rollout: deploy 10% of pods, wait for cache warm, then proceed to the next 10%. .

Model performance metrics (precision, recall, AUC) degrading without a corresponding input distribution shift; ClickHouse drift dashboard showing feature value distributions served online diverging from training population distributions; explicit skew audit (comparing offline training feature values against replayed online feature values for the same entity at the same timestamp) showing systematic differences in specific features

Tier 2: Training-Serving Skew Detection: Feature computation logic divergence between the offline training pipeline and the online serving pipeline: a feature transformation applied in training is not applied identically in serving. Recommended evolution: Enforce a single feature computation function registered per feature in a shared feature registry, executed by both the online pipeline and the offline training pipeline. The two paths must use the same code, not independently maintained implementations. Implement a skew monitoring job that computes a random sample of features using both paths and alerts on distribution divergence above a threshold. .

Kafka consumer group lag for fan-out worker group growing steadily; feed propagation latency (time from post write to follower feed update) exceeding 30s p95; Redis write rate on feed keys elevated but not saturated; post activity rate normal

Tier 1: Fan-Out Worker Queue Backlog: Fan-out worker pool undersized for burst post activity or a celebrity post creating a sustained high-fan-out event. Recommended evolution: Add fan-out worker replicas; implement fan-out cost routing: route high-follower-count fan-out events to a dedicated high-cost worker pool with separate Kafka consumer group and Redis write quota; use follower count threshold (e.g., >50k followers) as the routing decision. Monitor fan-out cost per post as a first-class metric. .

Redis memory utilization > 75%; eviction rate rising; cache miss rate on feed reads increasing; cold feed read fallback queries appearing in PostgreSQL slow query log; feed read p99 > 100ms despite Redis being online

Tier 2: Redis Feed Memory Ceiling: Redis feed list storage approaching memory limit; feed items being evicted before TTL; or feed list cap set too high for available memory. Recommended evolution: Reduce feed list cap from current value toward 100–150 items; increase Redis cluster capacity or shard feed keys by user_id range across multiple Redis primaries; implement tiered feed storage: hot recent items in Redis, older items fetched from PostgreSQL on demand with explicit product UX affordance .