Senior
Distributed Systems
What can go wrong will, in an order you did not expect.
Time, clocks and ordering, failure detection, consensus and leader election, replication and consistency in depth, idempotency and exactly-once, and the failure modes — partitions, split brain, cascading failure — with the designs that survive them.
2 lessons written3 outlined1 modules~1h reading
After this course you can
- Reason about ordering without a global clock
- Explain leader election and what a lease guarantees
- Design for partial failure and prove the design survives it
Curriculum
5 lessons · outlined lessons show their plan01
Time, clocks and orderingWall clocks lie, monotonic clocks, Lamport and vector clocks, and why "at the same time" is undefined.outlinedIdempotency and deduplicationAt-least-once is the only delivery you get: check-then-process caught 3 duplicate charges of 300, an atomic claim caught all, and a claim never released lost 40 payments.13 minBackpressureThe queue that grows forever, measured: five seconds of wait in eight, an OutOfMemoryError at 15,758 bodies, and what blocking, rejecting and shedding each buy.13 minFailure detection and leader electionHeartbeats, timeouts, phi accrual, leases, fencing, and the two leaders that both believed they were alone.outlinedFailure modes and resilient designPartitions, cascading failure, metastable states, retries as amplifiers, and the levers: timeouts, budgets, shedding, isolation.outlined