Preparing your design workspace

Loading the problem and your latest saved attempt…

Design a Vehicle Sensor and Obstacle Data Platform

Ingest hourly vehicle data at petabyte scale and support fresh regional searches over raw observations and year-long spatial summaries.

Problem statement

For this exercise, assume 10,000 vehicles upload hourly batches of timestamped, location-bearing sensor payloads and structured obstacle observations. The fleet produces about one petabyte per day before compression.

Authorized engineers search by time range, geographic region, and observation type, inspect recent raw records, and use coarser spatial and temporal summaries for older ranges. Accepted batches are immutable and may be retried; raw observations must become searchable within five minutes without double counting.

Retain raw data for seven days and lineage-bearing aggregates for one year. A query reads one published dataset generation so pagination cannot mix old and new aggregate versions.

Until retention expiry, publishing a newer generation may add newly accepted observations but must not reduce the count for an already published cell and interval. Vehicle control, live driving decisions, model training, and collecting data before the hourly upload are outside scope.

What to build

  1. A vehicle can idempotently register and upload an immutable hourly batch, and an operator can inspect its validation and indexing state.
  2. An authorized engineer can search a time range and geographic region by observation type at raw or aggregate resolution with stable pagination.
  3. An authorized engineer can stream the immutable payload behind a recent raw result without routing the entire object through the query metadata service.
  4. The platform publishes spatial and temporal aggregates with source lineage, freshness, and gap information while retaining raw data for seven days and aggregates for one year.

At scale

  • Assume 10,000 vehicles produce about one petabyte per day before compression and retry hourly uploads after intermittent connectivity.
  • Make an accepted batch's searchable metadata visible within five minutes and preserve it through a process or single-zone failure.
  • Retain raw observations for seven days and lineage-bearing spatial and temporal aggregates for one year.
  • Return the first metadata or aggregate page within two seconds at p95 for a bounded region and time range; raw payload delivery is streamed separately and is not expected to finish within that latency.