Scheduler Under Test

Task

A small job scheduler runs jobs on ticks, with retries, priorities and dependencies. It has a written contract and three shallow tests. The same file holds six broken builds, each of which breaks one rule of the contract. Write tests that pass on the scheduler and fail on every broken build. When a test you trust fails on the scheduler itself, fix the scheduler.

Requirements

1. handle(request) answers one request and does not change it. {op: "suite", build} runs every test_ function in the file against that build and returns {build, suite}, where suite is "pass" or "fail". {op: "run", jobs, steps} runs a small script on the scheduler and returns {states, runs, log, time}. An unknown op raises ValueError containing op, and an unknown build raises ValueError containing build.
2. Order and dependencies: tick() runs every job that can start now, once each, highest priority first, ties in the order the jobs were added. A job can start when it is waiting, every dependency is done, and the clock has reached its retry time. add() rejects a duplicate id (ValueError containing duplicate) and a dependency that was not added first (ValueError containing dependency).
3. Retries: an action that returns True succeeds. Anything else, or an exception, is a failure. A failed job is retried at most max_retries times, so it runs at most 1 + max_retries times, and then it is failed. Retry number n starts no sooner than base_delay * 2^(n - 1) seconds after the failure before it.
4. Skipping: a job with a dependency that failed or was skipped is skipped and never runs. This covers direct and indirect dependents. run_until_idle() ticks until no job is waiting, and moves the clock to the next retry time when nothing can run now.
5. Your suite must pass on the scheduler and fail on each of the six broken builds. Tests use the fake clock, not real time. Do not edit the broken builds.