Poll, never sleep

scope: generic · severity: law · confidence: proven · subsystem: method

A fixed sleep 60 is wrong in both directions. It wastes 40 seconds when the device came back in 18, and it reports a failure when the device needed 65.

Both probes worth having are cheap — fastboot devices and one ssh round trip cost about 0.2 s each, and under 20 ms with a warm ssh master — so there is no excuse for waiting a fixed interval “just in case”.

The shape:

. "$PORTHOLE_ROOT/tools/ph-lib.sh"
OLD=$(tk_boot_id)
... do the thing ...
tk_wait_ssh "$OLD" "$(tk_deadline_ms 300)" && echo up

tk_wait_ssh polls, auto-recovers if the device lands in the bootloader, and re-issues a reboot request that was swallowed. A reboot that finishes in 18 s returns in 18 s.

The deadline is not the same knob as the poll interval. Set the deadline from the worst case you are willing to call a failure; set the poll interval from what a probe costs. Conflating them is how a 300 s deadline turns into a 300 s wait.