Re: Kernel regression in head, now unusable for package building

On Mon, Apr 8, 2019 at 8:36 PM Charlie Li <ml+freebsd_at_vishwin.info> wrote:
> Konstantin Belousov wrote:
> > On Mon, Apr 08, 2019 at 12:10:23AM +0200, Antoine Brodin wrote:
> >> There seems to be a kernel regression in head,  that happened
> >> somewhere between r343921 and r345991.
> >> When launching "poudriere bulk -a",  the ssh session is terminated
> >> when poudriere attempts to clone/start builders (tmpfs mounts, file
> >> copying...),  the jails don't start and the consequence is that we
> >> can't build any package.
> > Are there any more details about the issue ?  It is not clear, does the
> > machine survives the event, i.e. did kernel paniced, what are the console
> > messages, any more details that you can provide.
> >
> I just ran into this both on a remote machine and the laptop I'm typing
> on right now. At least the reference jail does start and run, as any
> subsequent poudriere-bulk(8) invocations detect it. The entire login
> session is killed in the process, and the only clue of anything I can
> find (at least in the syslog) is that ntpd exits with Hangup:
>
> Apr  8 14:12:27 ardmore ntpd[74109]: ntpd exiting on signal 1 (Hangup)
>
> This seems to happen randomly, but still quite often. Restarting ntpd
> can help, though ntpd can still exit and kill the login again. Without
> ntpd running, running poudriere-bulk(8) is guaranteed to kill the login.

It should be fixed in https://svnweb.freebsd.org/changeset/base/346029

Antoine