chore(dev-stack): drop the two kill calls that never matched anything

[gabitoesmiapodo]

Summary

Closes #188

The issue reports that down leaves the dApp dev server holding 3012. Running the full stack showed the port is freed. What it did find is two kill calls that reach no process at all.

Changes

• down no longer runs a pkill pattern that matched nothing
• stop_pidfile no longer runs a pkill -P that never reached vite
• The comment on the surviving kill says what actually stops the dev server

Deviations

• The reported symptom does not reproduce. Both observations in the issue are right, but the plain kill already stops the whole chain, so nothing leaked. Worth editing dev-stack.sh down leaves the dApp dev server holding port 3012 #188 to say the code was dead, not broken.

Acceptance criteria

• down frees 3010-3012 and its port report prints (all free)
• No kill call in the script targets a process that cannot exist
• Behaviour is identical before and after

Test plan

Automated tests

None. No test runner covers scripts/dev-stack.sh.

Manual verification

1. Check out this branch and run ./scripts/dev-stack.sh docker-up
2. Run ./scripts/dev-stack.sh up and wait for "Stack is up"
3. Open http://localhost:3012 and confirm the dApp loads
4. Run ./scripts/dev-stack.sh down
5. Expect its port report to print (all free)
6. Run lsof -nP -iTCP:3010-3012 -sTCP:LISTEN and expect no output

Breaking changes

None.

Checklist

• Self-reviewed my own diff
• Tests added or updated
• Docs updated (if applicable)
• No unrelated changes bundled in

Screenshots

None.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
demo.canton-dappbooster	Ready	Preview	Sep 14, 2026 6:15pm UTC
docs.canton-dappbooster	Ready	Preview	Sep 14, 2026 6:15pm UTC

[fernandomg]

I've been facing the same problem during local runs, and reached a different solution.

CC Report

The premise doesn't hold: killing the pid on file doesn't reach vite. The pidfile holds the outer pnpm run app:dev; vite is two levels down (pnpm run app:dev → pnpm -C dapp/frontend run dev → vite.js). Same shape, no ports involved:

{ "scripts": { "outer": "pnpm run inner", "inner": "node -e \"setInterval(()=>{},1000)\"" } }

nohup pnpm run outer & PID=$!; sleep 3; kill $PID; sleep 2; pgrep -f setInterval   # still running

The grandchild survives under pnpm 11.24.0 (our pin) and 12.3.4. So down reaches less after this PR than before — the removed pkill -P at least took the direct child.

Agreed on the pkill -f "vite --host …" line: vite's argv is …/vite/bin/vite.js --host …, so it never matched. Removing it is right.

A down that works regardless of the process tree kills by port instead: lsof -t -iTCP:3012 -sTCP:LISTEN, filtered to listeners whose cwd is this checkout so another checkout's stack is left alone.

[gabitoesmiapodo]

@fernandomg

My CC's answer to this...

Ran the full loop through scripts/dev-stack.sh on this branch before changing anything.

The kill already reaches Vite

down does free 3012, and the single kill does reach vite.

With the stack up, the pidfile held the outer pnpm run app:dev, reparented to init:

78283     1 node .../pnpm run app:dev
78306 78283 node .../pnpm -C dapp/frontend run dev
78332 78306 node .../vite/bin/vite.js

vite.js held 3012 and curl http://localhost:3012/ returned 200. After ./scripts/dev-stack.sh down:

• port report printed (all free)
• lsof -nP -iTCP:3010-3012 -sTCP:LISTEN returned nothing
• no app:dev, no frontend run dev, no vite.js, no canton-wallet-service left

Also ran your minimal outer / inner / setInterval case on pnpm 11.24.0 and node 24.20.0. The grandchild died on kill $PID. So pnpm forwards the signal down both levels here.

The pkill -P "$pid" this PR removes would not have helped either way: it reaches the direct child, which is the middle pnpm, and vite is one level below that.

What I took from your suggestion

Killing by port is worth having, just not for the reported symptom: it covers the case no pidfile can: a crashed up, or a dev server someone started by hand. Before this, down reported the port busy and left you to find the pid.

Added in 4d17826

down calls it for 3010 and 3012 after the two stop_pidfile calls.

One detail worth knowing: the cwd match has to be a prefix. wallet-service runs from the repo root, vite from dapp/frontend. A plain = against $ROOT_DIR would skip vite and the filter would never fire.

Three cases checked on a full up / down:

1. Normal run: prints nothing, because the pidfile kills already freed both ports.
2. Untracked dev server on 3012, no pidfile: Freeing port 3012 from a stray dApp dev server (pid 78625), then (all free).
3. Listener on 3012 whose cwd is outside the checkout: left alone, still serving, and the port report shows 3012 in use.

shellcheck is clean. Teardown verified: no containers, no survivors, 3010-3012 free.