[OE-core] sysvinit upgrade woes

[Martin Jansa]

On Thu, Apr 16, 2015 at 06:42:58PM +0000, Bryan Evenson wrote:
> Paul,
> 
> > -----Original Message-----
> > From: Paul Eggleton [mailto:paul.eggleton at linux.intel.com]
> > Sent: Thursday, April 16, 2015 1:57 PM
> > To: Bryan Evenson
> > Cc: openembedded-core at lists.openembedded.org
> > Subject: Re: [OE-core] sysvinit upgrade woes
> > 
> > Hi Bryan,
> > 
> > On Thursday 16 April 2015 13:53:14 Bryan Evenson wrote:
> > > > -----Original Message-----
> > > > From: openembedded-core-bounces at lists.openembedded.org
> > > > [mailto:openembedded-core-bounces at lists.openembedded.org] On
> > Behalf
> > > > Of Bryan Evenson
> > > > Sent: Tuesday, April 14, 2015 11:26 AM
> > > > To: openembedded-core at lists.openembedded.org
> > > > Subject: [OE-core] sysvinit upgrade woes
> > > >
> > > > I am using opkg for package management and sysvinit for init.  I am
> > > > testing a large jump on an upgrade and I have been having a few issues.
> > > > One issue I have traced to sysvinit's upgrade, but I'm unsure how to
> > > > remedy the solution.
> > > >
> > > > At some point during the upgrade process, opkg records what the new
> > > > version is of all the packages getting installed.  However, when I
> > > > upgrade sysvinit causes sysvinit to restart before opkg has recorded
> > > > which packages were just installed.  So opkg thinks that it didn't
> > > > upgrade anything even though all packages were upgraded.  I verified
> > > > that if I call "opkg upgrade" for all packages that require upgrade
> > > > except for sysvinit that opkg correctly records which packages were
> > > > upgraded.  However, the end user doesn't have that kind of control,
> > > > so upgrading in this manner is not an option.
> > > >
> > > > Has anyone else seen a similar issue?  Any ideas on how to hold off
> > > > sysvinit from restarting until after opkg has finished recording its
> > > > information?
> > >
> > > I've found the problem but don't yet have a good solution.  At the end
> > > of the sysvinit-inittab pkg_postinst script is the line:
> > >
> > > 	kill -HUP 1
> > >
> > > This kills init before opkg has finished doing its job.  On my system,
> > > I've added a .bbappend with a pk_postinst script as follows for items
> > > that require a reboot after installation:
> > >
> > >     touch /var/run/reboot-required
> > >     echo "${PN}" >>/var/run/reboot-required.pkgs
> > >
> > > After my system performs firmware upgrade, it checks if
> > > /var/run/reboot-required exists and then reboots the system.  So on my
> > > system, I changed the pkg_postinst script for sysvinit-inittab to
> > > replace the "kill -HUP 1" with the reboot required notations.
> > >
> > > Is there some suitable solution that would be reasonable to change in
> > > the sysvinit-inittab recipe that would work for other people?
> > 
> > SIGHUP is supposed to make sysvinit re-read inittab, not kill it. Can you tell
> > what is actually happening?
> 
> Ah, I don't know for sure what is happening but now I have a different guess.  I have tried running the upgrade with "opkg -V4 upgrade" (insanely verbose debug output) and I can tell that is doing the postinst scripts when the upgrade stops and I get logged out.  I am doing my testing through the debug serial port on my system.  My best guess without further proof is that the SIGHUP restarts my terminal which in turn kills "opkg upgrade" which was commanded through my terminal.
> 
> If that is what is happening, then the upgrade should continue normally if I send it to the background ("opkg upgrade &"), correct?  Or it would run just fine if a cron job started the upgrade?

You can also run it in GNU screen session to verify this. I remember
seeing this behavior couple years ago and I prefer screen since then
(also for using the same screen after you set the network over serial
cable and then want to continue in the same shell but over network).

> > Cheers,
> > Paul
> > 
> > --
> > 
> > Paul Eggleton
> > Intel Open Source Technology Centre
> -- 
> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core at lists.openembedded.org
> http://lists.openembedded.org/mailman/listinfo/openembedded-core

-- 
Martin 'JaMa' Jansa     jabber: Martin.Jansa at gmail.com