Search the practice set
275 questions, 991 terms and 600 topics in 30 areas.
60 results for “Real-Time Serving Layer”
Serving Latency Budget
The end-to-end time from an event occurring to its effect being queryable, allocated across ingest, processing and serving.
Clock Drift
Device clocks diverging from real time, which corrupts ordering, expires certificates early or late, and produces data that appears to arrive from the past.
Latency Budget Decomposition
Allocating a total response-time target across the components of a request path, so each layer has an explicit share and overruns are attributable.
Load Balancing Algorithm
The rule deciding which backend receives a request — round robin, least connections, least response time, or hash-based.
Open Table Format
A metadata layer over files in object storage that supplies ACID transactions, schema evolution and time travel — the thing that turns a data lake into a lakehouse.
Stale While Revalidate
Serving cached data immediately while fetching a fresh copy in the background, which makes an interface feel instant at the cost of a brief inconsistency.
Sub-Second Aggregation Store
A database built to ingest continuously and answer aggregate queries over recent data in milliseconds, occupying the gap between OLTP and the warehouse.
Unbounded Dataset
Input with no known end, which removes the option of waiting for completeness and forces every aggregate to be provisional.
Anti-Corruption Layer
A translation layer that converts a legacy or external system's model into your own, so its concepts do not leak into your domain.
Bake Time
The deliberate wait between rollout steps, sized by how long the fault you are watching for takes to appear.
Bronze Layer Contract
What the raw landing zone promises and refuses to promise — a faithful replayable copy of the source, with no correction applied.
Dwell Time
The period between an attacker gaining access and being detected — the metric that determines how much damage an intrusion can do.
Event Time Versus Processing Time
The distinction between when something happened and when the system saw it, which determines whether results are reproducible.
Google Maps and Planetary-Scale Spatial Serving
Map serving is fast because almost nothing is computed on request — the world is precomputed into a pyramid of tiles, and space is indexed onto a one-dimensional curve.
Layer 4 vs Layer 7 Load Balancing
Balancing on connection metadata (IP and port) versus on the content of the request (path, host, headers).
Lead Time to Value
The elapsed time from identifying an opportunity to delivering measurable benefit, which is usually dominated by waiting rather than by building.
Metric Definition Layer
Business metrics defined once in a versioned, tested place and served to every consuming tool, so the number cannot differ by which tool asked.
Output Validation Layer
A deterministic check applied to model output before it is used, treating the model as an untrusted component.
Point-in-Time Recovery
Restoring a database to any moment within a retention window by replaying transaction logs onto a base backup, rather than only to a snapshot boundary.
Queue Time Attribution
Splitting elapsed lead time into work and waiting, and naming which queue each wait sat in.
Recovery Time Objective
The maximum acceptable duration between a failure and restored service, agreed with the business rather than chosen by engineering.
Round-Trip Time
The time for a packet to travel to a destination and back — a physical floor that no application optimisation can reduce.
Time Travel
Querying a table as it existed at a previous version or timestamp, made possible by keeping the metadata and files of prior commits.
Time to Interactive Gap
The interval between a page appearing complete and actually responding to input, during which the user's taps do nothing.
A core mainframe system with no API supports nightly batch file exchange only. The business needs near-real-time order status. Design the integration.
Establish the real constraint "No API" usually means no API the mainframe team will build on your timeline . Find out what exists: message queue interfaces, dat
A business sponsor asks for a real-time data platform because "the competition has one". Reporting is currently a nightly batch that lands at 06:00 and nobody has complained. How do you handle this?
Do not answer the technology question "Real time platform" is a solution, and it has arrived without a problem attached. Answering it directly leads either to a
Quarterly access reviews take two weeks of manager time and everyone approves everything. How do you make this a real control?
Recognise what it currently is A review where everything is approved is not producing a decision; it is producing a record of a decision shaped activity. Two we
An end-to-end suite of 340 tests takes four hours and fails spuriously about half the time. The team wants to parallelise it. Is that the right move?
Parallelising treats the symptom It might halve the runtime. It will not touch the flakiness — in fact parallelisation often worsens it, by exposing shared stat
An executive asks why the new system "feels slower" when your dashboards show average response time improved. How do you explain it?
The average is the problem, and that is the explanation Averages hide the tail. A system where most requests got faster and the slowest ten percent got much slo
Delivery leadership says releases are too slow and wants the engineering team to "move faster". Lead time from commit to production is 21 days. How do you investigate, and what do you expect to find?
Do not accept the framing "Move faster" assumes the constraint is how fast code is written. That is almost never where the twenty one days went, and starting an
Design a URL shortener handling 100 million new links per month and 10 billion redirects. Where is the real difficulty?
What the interviewer is testing The classic warm up. What is being assessed is not whether you can shorten a URL — it is whether you do capacity arithmetic befo
Netflix personalises an entire home page in the time a TV takes to draw a screen. How, given that scoring every title for every member on request is impossible?
What the interviewer is testing Whether you reach for the offline/online split — the single most reusable idea in latency critical personalisation — rather than
Partner integrations take four months from contract signature to production traffic. Engineering says the work is three days. Where does the time go and how do you cut it?
Map the elapsed time honestly The engineering estimate is correct and irrelevant. The four months are: Legal and contractual review. Security questionnaires and
Precompute every user's timeline at write time, or assemble it at read time? Explain why the answer for a social feed is neither.
Why each pure strategy fails Fan out on read. Store each post once; on timeline load, query the posts of everyone the user follows and merge. Writes are trivial
Real-Time Serving Layer
Where a low-latency read of a streaming aggregate actually lands.
Real-Time Analytical Stores
Druid, Pinot and ClickHouse — ingest-and-query engines for sub-second aggregation.
Streaming & Real-Time Data
General material on continuous processing of unbounded data.
Backfill & Reprocessing
Replaying history through changed logic without double-counting the live output.
CDC to Stream
Turning database changes into an event log, and how that differs from a domain event.
Dead Letter Handling
The poison message that blocks a partition, and the queue nobody reads.
Exactly-Once Semantics
What the phrase really means, where it holds, and the idempotent sink underneath it.
Feature Freshness
How stale a feature can be before the model degrades, and the pipeline that follows.
Kappa vs Lambda
One pipeline replayed versus two pipelines reconciled, and the maintenance each carries.
Partition Keys & Ordering
Ordering guaranteed only within a partition, and choosing the key that makes that enough.
Stateful Stream Processing
Keyed state, state backends, checkpoint size, and the restore time that follows.
Stream Processing Frameworks
Flink, Kafka Streams, Spark Structured Streaming — state, checkpointing and recovery.
Stream-Table Duality
A changelog and a table as two views of the same thing, and materialising between them.
Streaming Cost
Always-on compute, retention and cross-zone traffic as the three bills that surprise.
Streaming Joins
Joining two unbounded streams, the buffering it needs, and the enrichment alternative.
Streaming SLOs
End-to-end latency, consumer lag and completeness as commitments rather than dashboards.
Streaming Schema Evolution
Changing an event's shape while a retained log still holds every older version of it.
Streaming vs Batch
The freshness requirement that actually justifies streaming, and the cost of assuming one.
Watermarks & Late Data
Deciding a window is complete when events can still arrive, and what to do when they do.
Windowing
Tumbling, sliding and session windows, and the aggregation each one answers.
Anti-Corruption Layer
Translating a foreign model at the boundary so it does not leak in.
Client Caching & Data Layer
Stale-while-revalidate, invalidation and optimistic updates on the client.
Layer 4 vs Layer 7
Connection-level versus request-level balancing, and what each unlocks.
Real User Monitoring
Field data from real devices and networks, against the synthetic run that looked fine.
Semantic Layer
Metric definitions held once and served to every tool that asks.