Distributed computing
On this page 8 topics
Overview
Topics
- Nodes
- network latency
- partial failures
- clocks
- failure models
Learn first
2 h of practice (estimate)
Learning outcomes
- List the fallacies of distributed computing and give a failure example for each.
- Explain partial failure: why a timeout does not tell you whether the operation happened.
- Explain clock skew and why wall-clock timestamps cannot order events across nodes.
- Describe crash, omission, and byzantine failure models at a conceptual level.
Lessons
Lab and exercises
Lab
MID-023-LAB1 · Unreliable network
Outcomes: 1–4 · Time: ~2 h
- Add Toxiproxy between Shop and its payment provider stub (Docker Compose).
- Run scenarios: added latency, timeout shorter than latency, dropped response, connection reset.
- After each, compare Shop's order status with the provider's record.
- Write the observed inconsistencies and which fallacy or failure model each demonstrates.
Acceptance: list of every inconsistency observed.
Exercises (assessment: failure-scenario quiz; five exercises)
| ID | Outcome | Task |
|---|---|---|
| MID-023-EX01 | 1 | For each of the eight fallacies, write one failure example from Shop. |
| MID-023-EX02 | 2 | Explain, with a timeline, why a timeout on charge() does not mean the card was not charged. |
| MID-023-EX03 | 3 | Two services log OrderPlaced and OrderPaid with local timestamps; paid appears earlier. Explain two causes and a fix. |
| MID-023-EX04 | 4 | Classify: node dies, link drops 5% of packets, node returns corrupted prices. |
| MID-023-EX05 | 1 | Estimate the minimum latency of a request that makes 8 sequential same-datacentre calls of 1 ms each, and say how to reduce it. |
Show solutions
- EX01 For example: reliable network, lost confirmation; zero latency, N+1 remote calls; secure network, plaintext traffic; topology, stale IP; one administrator, firewall change; zero transport cost, large payload bills; homogeneous, mixed API versions; infinite bandwidth, bulk export slows checkout.
- EX02 Timeline: request sent, provider charges, reply lost, client times out. Shop cannot distinguish this from a request that never arrived.
- EX03 Clock skew or NTP correction; fix by ordering with a sequence, a version number, or a single log (e.g. one Kafka partition per order).
- EX04 Crash; omission; byzantine.
- EX05 At least 8 ms plus processing; reduce by batching, parallel calls, caching, or merging the calls.
Quiz
Outcomes: O1 fallacies · O2 partial failure · O3 clocks · O4 failure models.
A self-check for you: nothing is graded, saved or sent anywhere. Choose an answer, then check it. You can try again as often as you like.
Practical part
Five exercises MID-023-EX01 to EX05 plus the lab. Pass: at least four exercises correct and a lab list that names at least three distinct inconsistencies with their causes.
Common mistakes
Treating a timeout as a failure; ordering events by timestamps from different machines.
Assessment
Failure-scenario quiz.
Career relevance
Every later distributed-systems module.