Warehouse / Serving
Use a workload-to-store matrix for BI, Pinot, Trino, raw truth, and operational query expectations.
On this page
Workload matrix
Match dashboards, ad hoc SQL, and ops consumers to the right serving layers.
Different consumers need different serving layers, so I match BI, ad hoc SQL, real-time OLAP, and raw forensic queries to the right engines instead of forcing one store to do everything.
Quick jumps
Click to jumpMonthly users
200M-250M
Docs use both 200M+ and 250M assumptions
Daily active users
80M
Used for heartbeat and Kafka math
Peak concurrency
15M
Useful for QoE and live ops workloads
Match each consumer to the right serving store
One store should not pretend to solve BI, ad hoc, live ops, and raw forensic truth equally well.
Workload-to-store matrix
Selected workload
BI dashboards
Recommended
Why
Not ideal
Tie serving decisions back to freshness
Make dashboard latency and warehouse expectations explicit.
Live concurrent viewer dashboard
5 seconds to 1 minute
Expected availability for this consumer style
Trending content detection
1 to 5 minutes
Expected availability for this consumer style
Fraud or anomaly alerting
Seconds to 1 minute
Expected availability for this consumer style
Resume watching position
Sub-second to a few seconds
Expected availability for this consumer style
Real-time recommendation features
Seconds to minutes
Expected availability for this consumer style
Daily content performance report
Available by morning
Expected availability for this consumer style
Monthly revenue report
Hours acceptable
Expected availability for this consumer style
ML training dataset
Daily or every few hours
Expected availability for this consumer style
Use a serving-layer answer that sounds intentional
The main point is matching workload shape to store shape.
Say This In Interview
Different consumers need different serving layers, so I match BI, ad hoc SQL, real-time OLAP, and raw forensic queries to the right engines instead of forcing one store to do everything.
30-second answer
FastDifferent consumers need different serving layers, so I match BI, ad hoc SQL, real-time OLAP, and raw forensic queries to the right engines instead of forcing one store to do everything.
2-minute answer
BalancedUse a workload-to-store matrix for BI, Pinot, Trino, raw truth, and operational query expectations. Focus on MODELING, 7 min, Match dashboards, ad hoc SQL, and ops consumers to the right serving layers..
Deep answer
DeepDifferent consumers need different serving layers, so I match BI, ad hoc SQL, real-time OLAP, and raw forensic queries to the right engines instead of forcing one store to do everything. Then go layer by layer through MODELING, 7 min, Match dashboards, ad hoc SQL, and ops consumers to the right serving layers., including trade-offs, failure handling, and why this shape is right for the consumer SLAs.
Last reviewed June 2026 · By Prasoon Parashar
Numbers are interview assumptions, not real Netflix internal figures.