Date: Sat, 26 Nov 2011 14:00:30 -0500 From: Jason Hellenthal <jhell@DataIX.net> To: Michael Ross <gmx@ross.cx> Cc: Christian Kastner <debian@kvr.at>, freebsd-hackers@freebsd.org Subject: Re: cron(8) mis-feature with @reboot long after system startup Message-ID: <20111126190030.GA58253@DataIX.net> In-Reply-To: <op.v5jgu0rmg7njmm@michael-think> References: <20111125070241.GA7915@DataIX.net> <4ED00A68.4040606@kvr.at> <20111126051121.GA4628@DataIX.net> <op.v5jgu0rmg7njmm@michael-think>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, Nov 26, 2011 at 06:43:38AM +0100, Michael Ross wrote: > Am 26.11.2011, 06:11 Uhr, schrieb Jason Hellenthal <jhell@dataix.net>: > > > > > > > On Fri, Nov 25, 2011 at 10:36:40PM +0100, Christian Kastner wrote: > >> Hi, > >> > >> On 2011-11-25 08:02, Jason Hellenthal wrote: > >> > So with that said... is there a way we could actually make this run > >> @reboot only ? > >> > >> Debian's cron[0] and Fedora's cronie[1] have solved this by touching a > >> file on first startup and running @reboot only when this file does not > >> yet exist. > >> > >> Note that while [0] may point to other patches that might be of interest > >> to FreeBSD, they are still WIP (as evident from the linked patch) as we > >> are still in the process of quiltifying our current code base. > >> > > > > While this sounds like a perfectly sane way to handle the problem at > > hand this also introduces a need to write some cleanup code to take care > > of the file semantics. I think comparing daemon start times to the time > > a system was booted or similiar would alleviate the need for all that. > > Maybe a flag for @reboot "-s <seconds>" seconds after boottime to allow > > running @reboot jobs. And set the default to 3600 seconds. At least this > > would allow adjustment for those startup processes that may take some > > considerable time before multiuser mode is entered. > > > > Just some thought. > > > > I really don't think we need to go the route of using files to store > > information when there is enough information available already via > > syscall's. > > If system startup were to be unusually delayed (dhcp or nfs trouble eg), > $time_period might have passed when cron starts, so there would have to be > some notifying mechanism for @reboot jobs not being run, and operator > action would be required. > I agree but also disagree. 1 hour or 3600 seconds is plenty of time to wait for the "@reboot" extension scripts to fire. > One could "cron restart" within $time_period and run @reboot twice. > > I don't like this idea. > > In my bikeshed, @reboot should be called @cronstart, but it shouldn't > because of uniformity across OSs. > The cron manpage should reflect the actual behaviour (already patched in > svn), > and all "definitely only on boot"-scripts should go in local/etc/rc.d. > This has nothing to do with rc.d. This is a cron(8) problem only.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20111126190030.GA58253>