[william-gr]

Why

#24 made unique() wait until node B answers SSH from a new boot after its reboot. webui run 39296815434 still failed at the same step:

• The new wait saw B's new boot over SSH.
• It then waited for B's API.
• It then read B's IDs in a separate SSH call, which was refused: ssh: connect to host 10.231.0.12 port 22: Connection refused.

So sshd answering once does not mean it stays up. SSH on the nodes is enabled through middleware (tnctl enable-ssh), and middleware applying its SSH settings as it starts most likely restarts sshd. That cause is a reading, not confirmed.

Change

The loop's SSH call now returns B's boot id and its machine-id and hostid together. The call that sees the new boot also yields the IDs, so nothing depends on sshd staying up afterwards. The API wait still follows, and it does not use SSH.

Testing

• shellcheck -x appliance.sh ha.sh lib.sh (the CI command) passes.
• The loop was run locally against a fake ssh_: refused twice, then the old boot, then a new boot. It kept waiting, took the new boot id and the IDs from that one call, and the "still has A's ids" check passed.

[github-actions]

LOW — This comment states as fact that middleware "restarts sshd", but the PR description calls that cause a reading, not confirmed. The code doesn't depend on which cause it is, but these comments are where the repo records why things are the way they are. Something like "most likely restarts sshd" would keep the record accurate if the real cause turns out to be different.

[github-actions]

One finding: one LOW. No human review needed.

The change reads node B's boot id, machine-id and hostid in one SSH call, so the call that sees the new boot also returns the IDs. That removes the later ids b call, which is the one that was refused in run 39296815434. How the output is split:

• now="${out%% *}" is the boot id. It is empty when SSH fails, because || out= clears out, so the loop keeps waiting.
• b="${out#* }" is "<machine-id> <hostid>", the same format ids() returns. The [ "$a" != "$b" ] comparison with A's IDs therefore still compares like with like.
• The IDs are now read before the API wait instead of after it. They don't change within a boot, so the result is the same.

LOW — the new comment at ha.sh:245-247 states the middleware-restarts-sshd cause as fact, while the PR description says it is unconfirmed (inline).

Separately, check-ticket is failing because the PR title has no ticket reference. That comes from the title, not the code.

[github-actions]

Approved. 1 finding(s): 1 LOW.

Nothing blocks and no human review is needed.