D&C GLug - Home Page

[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]

Re: [LUG] OS Architecture

 

On Wed, 22 Nov 2006 13:43:01 +0000
"Ben Goodger" <goodgerster@xxxxxxxxx> wrote:
> > On Wednesday 22 November 2006 12:47, Simon Waters wrote:
> >
> > > My suspicion is that many GNU/Linux boxes aren't rebooted often enough,
> > ....
> >
>
> Er, no... the norm has always been to reload the service that was updated
> (usually done by APT) and then continue.

This "hook" is only used by libraries directly related to the service
being restarted. It is not the "norm" for more distant dependencies to
cause a reload of the service - despite the fact that the previous
version *is* still in memory after the new version completes
installation.

In many cases, new connections to that service will cause new processes
to be forked and those new processes can use the new code but that
leads to problems if the new process ever tries to do something that
affects an old process.

e.g. if service A depends on libB which in turn depends on libC,
upgrading libC may or may not cause the restart of service A - unless
service A depends on both libB and libC. This is the "indirect
dependency" problem that is currently a problematic issue in GNU/Linux:
http://lists.debian.org/debian-devel-announce/2005/11/msg00016.html
http://rerun.lefant.net/checklib/faq.html

> In the world of Joe User, this
> should be constrained to logging in and out again if the update is
> major-ish, or just carrying on as normal otherwise.

No - these are generally system services, like apache or bind, which
are not restarted when logging out and back in. That process only
restarts X related services - DCOP in KDE and GConf in Gnome - although
these can (and are) restarted directly where the need arises, during
the update process.

The question is more that there is a suspicion that not every
dependency of every service causes a reload of every affected service
at every update. It doesn't appear to cause any identifiable issues
but it is a theoretical problem.

--


Neil Williams
=============
http://www.data-freedom.org/
http://www.nosoftwarepatents.com/
http://www.linux.codehelp.co.uk/

Attachment: pgpznO4ijqLfc.pgp
Description: PGP signature

-- 
The Mailing List for the Devon & Cornwall LUG
http://mailman.dclug.org.uk/listinfo/list
FAQ: http://www.dcglug.org.uk/linux_adm/list-faq.html