rrs_upgrade_history.py analyses each commit by running upgrade_history_internal.py through utils.run_command_interruptible(), which waits for it with no time limit. When a commit makes the parse hang, the whole run stalls and keeps the layer index update lock until someone kills it. Bugs 16500 and 16504 were found this way: on a test instance, the run sat idle on one OE-Core commit for over 7 hours. The nightly RRS job on the test instance (rrs-update.service) has no timeout either. A timeout around the whole run (timeout(1) or systemd) doesn't help much: it stops the run instead of skipping the bad commit, and in the hang the stuck process was the BitBake server started by upgrade_history_internal.py, which a plain kill of the child leaves behind. Proposed fix: an optional timeout in run_command_interruptible() that kills the command's whole process group and raises subprocess.TimeoutExpired, and a --commit-timeout option to rrs_upgrade_history.py (default 1 hour). With it, a commit that times out is logged as an error and skipped.
Patch sent to yocto-patches, under review (applies on top of the RRS series for bugs 16499-16506): https://lore.kernel.org/all/20261115011321.58851-2-tim.orling@konsulko.com/
v2 sent (fixes a race in the new test; v1 is superseded): https://lore.kernel.org/all/20261115015200.64142-2-tim.orling@konsulko.com/