Read the handshake, not the docs.
A packet-level network simulator in your browser: links with latency and loss, DNS, TCP, TLS and HTTP as they actually happen. Configure a client, run the traffic, and be graded on what the packets did.
Why this one
Every incident is a timeout. The layer below the application is the layer nobody on the team learned, and it is where p99 lives.
The real argument: a model writes the network client without a model of the network. A generated fetch with a retry loop, no jitter, no idempotency key and a fresh connection per request looks correct, passes the unit test, and at load becomes a retry storm, a TIME_WAIT exhaustion, or a DNS TTL stampede. None of those are visible in the code. They are visible in a capture, and reading one requires knowing what a handshake costs.
How it is graded
Not against a quiz key. The simulator produced the situation, so it already knows the answer: round trips per request, retransmissions, whether a non-idempotent request reached the origin twice, what state every socket ended in, and whether the request came back inside its budget. Every one of those is counted by the scheduler as it happened.
The budget a task holds you to is measured the same way — it is what a configuration that solves the task actually cost, not a number somebody picked.
- lessons
- 9
- tasks
- 7
- backend
- none
- account
- none