Demo runbook

In one sentence

Run each anti-pattern once in the order below, with the services already up, and keep the captured numbers ready as a fallback for every live step.

This runbook is for presenting or rehearsing the talk. The commands are repeated with expected output in Running the demos. If a live step misbehaves, see Troubleshooting.

Before you start

Rehearsal checklist

• `PAYMENT_GATEWAY_URL` resolves locally for the mode you are running. For SQLite run, use `http://localhost:8080`. See the [gotchas](/guide/setup#gotchas).
• Both services are up: the Orders API on port 8000 and the gateway on port 8080. `curl localhost:8000/health` returns `{"status":"ok"}`.
• `curl localhost:8000/demo/ap5/info` shows the pool mode you expect.
• The database is seeded. `curl "localhost:8000/orders?limit=1"` returns rows.
• Each anti-pattern has been run once beforehand, so imports are warm and timings are clean.
• The captured fallbacks are open: `benchmarks/*` and `talk/assets/*.svg`.

Run of show: 20 minutes

The DEMO_GUIDE gives a 20-minute run of show. The talk itself is 30 minutes, with the rest of the time spent on framing, the checklist, and Q&A.

Minutes	Segment	Action
0 to 2	Baseline	Show the baseline endpoints, and /docs
2 to 6	AP1	Three concurrent bad and good calls. Run the health-check trick. Show the asyncio debug output.
6 to 9	AP2	Run bench_ap2_di.py. Say Postgres is about 10×.
9 to 13	AP3	Show the 500 from bad-crash, then the query-count script
13 to 16	AP4	curl both endpoints, then the cProfile call counts
16 to 19	AP5	Restart in bad, run locust, restart in good, run locust. Show the timeout traceback.
19 to 20	Wrap	The performance checklist

Talk timing: 30 minutes

The presenter guide paces the 30-minute slot in segments. The live demos sit inside these windows.

Time	Segment	Live demo
0:00 to 2:00	Opening hook	None
2:00 to 5:00	Framing: what async buys you	None
5:00 to 9:00	AP1	Bad and good, side by side
9:00 to 13:00	AP2	Optional: bench_ap2_di.py (about 2 s)
13:00 to 17:00	AP3	bad-crash first, then bad-n1 and good
17:00 to 21:00	AP4	Not live. Show the captured cProfile.
21:00 to 25:00	AP5	Captured slide by default. Live variant only if ahead on time.
25:00 to 27:00	Checklist and toolkit	None
27:00 to 28:00	Repo call to action	None
28:00 to 30:00	Q&A	None. See the FAQ.

Cut order when running over time

Cut AP4's live curl first, since it is optional. If time is still tight, compress AP2 to "here are the numbers" without narrating the benchmark script.

The five demos at a glance

Pattern	Live step	Fallback if the live step fails
AP1	Three concurrent bad calls, then good	Captured 3.046 s and 1.040 s
AP2	bench_ap2_di.py	Captured 17.48 ms and 1.73 ms table
AP3	bad-crash 500 response	missing-greenlet-response.json, the safest demo
AP4	Optional curl pair	Captured cProfile counts
AP5	Restart and locust, optional	Captured locust tables and the traceback

Related

• Running the demos: commands and expected output for every step
• Troubleshooting: gotchas and what to do on stage
• Setup