As of commit 92dcec89 the clearing of console logs is tied to the watchdog log monitoring, rather than using the command queue. This fixed a race between the logs being cleared and the watchdog starting to monitor the log -- but it means the logs are never cleared when manually provisioning. One possibility we have discussed is to put clear_logs back into the command queue, and change things so that the watchdog kill_time is initialised as a callback from the reboot command instead of when the recipe is started by beakerd. This will fix the race (because the watchdog won't be picked up until kill_time is not NULL, by which point the logs will have been cleared) and it will also allow us to clear logs for manual provisions.
(In reply to comment #0) > As of commit 92dcec89 the clearing of console logs is tied to the watchdog > log monitoring, rather than using the command queue. This fixed a race > between the logs being cleared and the watchdog starting to monitor the log > -- but it means the logs are never cleared when manually provisioning. This is quite a serious bug because of the following situation: * User manually reserves a system * User plays around, triggers a panic somehow * User returns the system, but log is *not* cleared * Scheduler picks the system for a job * Watchdog sees panic in old console log and aborts the job
On Gerrit: http://gerrit.beaker-project.org/1204
Beaker 0.9.1 has been released.