Bug 155893

Can the daemon be changed to exit in this case? Otherwise it should not be on by
default in runlevel 5, I guess.  This is going to be annoying if it's not fixed
 for FC4. 

It must no exit, because :

o "no devmaps" is a valid temporary case
o memory allocation failures can cause the looping, better to retry later than
exit in this case too.

That said, it is obvious to rate limit the loging, which is done in this patch :
http://kernel.org/git/gitweb.cgi?p=linux/storage/multipath-tools/.git;a=commitdiff;h=457b498a392a61a0198b5dea80972930069d0537

If the daemon is going to be left consuming resources (at least CPU) every five
seconds then I'd still argue it should *not* be on by default in runlevel 5. 
Most Fedora users won't have multipath hardware, I'm guessing.

I'd recommend bringing this to fedora-devel-list for public debate.  I am not
sure if I agree with disabling the service by default.

I do agree that consuming resources every 5 seconds (even with log rate
limiting) is not good.  I just don't have a better suggestion, so this should go
to public debate.

It should be pointed out that even a running multipathd will "spam" the syslog
at irregular intervals with an empty message, but apparently so high prio that
it'll be broadcast to all consoles by default. I've not yet tracked down where
in the code this originates from.

Quite simply, the daemon should not be running unless you're using device-mapper
multipath.

We don't have support in anaconda yet, so:

    (i) The startup script needs changing to detect whether or not the daemon
should be started [not sure how to do this];  or

    (ii) The daemon should exit immediately if there are no devices - could be
controlled by a command line option, but possible problems if devices are slow
to appear on the system;

    (iii) The installer should 'chkconfig --levels 0123456 off' and leave people
to turn it on manually post-install.


I reckon (iii) is best at this stage.

Any other options?

s/installer/%post/

To achieve (iii) just change the init script chkconfig: line to read:

chkconfig: - 18 87

so that the "chkconfig --add" invocation in %post does not enable the service to
start in any runlevels to begin with.

But that would lose the default list of run levels for people who do need the
package.

To preserve that then just don't call "chkconfig --add" at all in the %post
script?  Let the users who need it do that...

I think 'added' but 'off' is better.  Otherwise you don't see it as an option in
the lists.

building device-mapper-multipath-0_4_4-2_1

That should solve the immediate problem, by not enabling it by default.

The problem of too many error messages will be fixed by Christophe upstream and
get picked up by Fedora, probably post-FC4.

> The problem of too many error messages will be fixed by Christophe upstream and
> get picked up by Fedora, probably post-FC4.

Is fixed, in
http://kernel.org/git/gitweb.cgi?p=linux/storage/multipath-tools/.git;a=commitdiff;h=457b498a392a61a0198b5dea80972930069d0537

Not fixed in 0.4.4-2.1... spot the mistake ;)

/sbin/chkconfig --add multipathd
/sbin/chkconfig --levels 0123456 off
                                ^^

Gah!

*** Bug 157461 has been marked as a duplicate of this bug. ***

Really fixed in device-mapper-multipath-0.4.4-2.3, thanks Bill.

This bug, as described in the first report,  has appeared again in FC4 for x86_64. 

Exactly which RPM is this still happening with?