One-command startup: `rfdf.sh`¶

Lives on: the follower Pi 5, at ~/capstone/rfdf/rfdf.sh. Source in this repo under agentcore/rfdf/.

What it does¶

Brings up the whole follower drone stack with one command, or stops it, or just checks it:

~/capstone/rfdf/rfdf.sh up             # start whatever isn't running, then check each piece
~/capstone/rfdf/rfdf.sh status         # check only; starts/stops nothing
~/capstone/rfdf/rfdf.sh down           # stop AgentCore and the DOA app
~/capstone/rfdf/rfdf.sh down --kraken  # also stop the Kraken software on the Pi 4

Before this script, starting a bench test meant three terminals and five manual steps in the right order. Now it's one line, and logs go to files instead of terminal windows.

Walkthrough: what `up` does, in order¶

#	Step	How it knows the step worked
1	Pi 4 reachable over the onboard cable	SSH answers within 5 s
2	Pi 4 clock synced to the Pi 5	`timedatectl` reports synchronized (waits up to 30 s)
3	Kraken software running on the Pi 4	A fresh sample (< 3 s old) appears (waits up to 2 min)
4	DOA app running	Something is listening on local port 8770
5	AgentCore running	Its web UI answers on port 8085

The order matters. The DOA app starts before AgentCore so that when AgentCore's graph starts, the app it calls is already there.

Walkthrough: safe to run any time¶

Every piece is checked before it's started. If a check already passes, that step just reports "already running":

if check_kraken >/dev/null 2>&1; then
  ok "Kraken already running"
else
  info "starting Kraken on the Pi 4 (takes ~1 min)..."
  $SSH "cd $KRAKEN_DIR && nohup ./kraken_doa_start.sh > /tmp/kraken_start.log 2>&1 < /dev/null &"
  wait_for 120 check_kraken
fi

So if one piece falls over mid-test (the Kraken's USB link dropping, say), running rfdf.sh up again restarts only that piece. The check for the Kraken is the same idea as the DOA app's health check: it looks at how old the newest sample is, not just whether the page answers.

kraken_age_ms() {
  local t
  t=$(curl -s -m 1 "$DOA_URL" | cut -d, -f1 | tr -dc '0-9')
  [ -n "$t" ] && echo $(( $(date +%s%3N) - t ))
}

wait_for <seconds> <check> retries a check once a second until it passes or time runs out, then prints the final [ OK ] or [FAIL] line.

Things learned building it¶

AgentCore held its printed output back when writing to a file, so the graph's OK lines didn't show up in the log until about 4 KB had piled up. Starting it through stdbuf -oL -eL fixed that.
Everything is started with nohup, so closing the SSH session (or the test screen) does not stop the stack.

Still to do¶

Start automatically at power-on for flight, with no operator (systemd services).
Log rotation for ~/capstone/rfdf/logs/.
The same setup on the leader drone.

One-command startup: rfdf.sh¶