Module 10 Activity

Scenario

Something in your workflow produces a wrong answer without raising. This module is how you find it in minutes rather than an afternoon.

What you build

A debugging pass using state history and a trace record, ending in a logging schema you would want during an incident.

Steps

1. Plant a bug in a middle node that produces a wrong value without raising. Have a colleague run it if you can.
2. Find it using get_state_history alone, reading forwards for the first wrong value. Record how long it took.
3. Take a trace of a run and read it backwards from the symptom, marking which steps are correct given their input.
4. Reproduce a failure by replaying the recorded inputs on a fresh thread. Report whether it reproduced, and what that tells you.
5. Recover a failed run by editing state and resuming, and count how many times the expensive node ran versus a restart.
6. Write your trace schema, including state_before, state_after and route_reason.

Evidence to hand in

• The planted bug and how it was found, with the time taken.
• The backwards trace reading with each step marked.
• The replay result and what it establishes.
• The recovery, with the expensive-node call count against a restart.
• The trace schema.

Review checklist

• The bug was found from state history, not print statements.
• The backwards reading identifies the first wrong value, not the slowest step.
• The replay changed nothing except the thread id.
• Recovery ran the expensive node fewer times than a restart would have.
• The schema includes route_reason and covers successful runs, not only failures.

Mark complete