On Thu, 25 Apr 2013, Andreas Friedrich wrote:
> On Mi, Apr 24, 2013 at 11:04:56 -0700, Sage Weil wrote:
>
> Hello Sage,
>
> ...
> > > If the start script exits with or without error, the remote config
> > > file will be removed from <remote-host>:/tmp - but only for the last
> > > host (the former trap calls will be overwritten by the last trap
> > > call)! This makes no sense to me.
> >
> > Bah.. I was thinking they would stack. How about
> >
> > diff --git a/src/init-ceph.in b/src/init-ceph.in
> > index 61c10e1..195bf76 100644
> > --- a/src/init-ceph.in
> > +++ b/src/init-ceph.in
> > @@ -218,8 +218,11 @@ for name in $what; do
> > else
> > unique=`dd if=/dev/urandom bs=16 count=1 2>/dev/null | md5sum | awk '{print $1}'`
> > scp -q $conf $host:/tmp/ceph.conf.$unique
> > - trap "ssh $host rm /tmp/ceph.conf.$unique" EXIT
> > cur_conf="/tmp/ceph.conf.$unique"
> > +
> > + # clean up on exit
> > + remote_configs="$host:/tmp/ceph.conf.$unique $remote_configs"
> > + trap 'for f in '$remote_configs' ; do ssh ${f%:*} rm ${f#*:} ; done' EXIT
> > fi
> > cmd="$cmd -c $cur_conf"
>
> yes, this looks fine but I have a question:
>
> I don't know the background for copying the config files to
> <remote-host>:/tmp. Why don't use all the daemons their local config
> /etc/ceph/ceph.conf?
>
> May be you want to ensure that the remote daemons will use the same
> config as exists on the host from which the cluster was started. But
> if the configs on the remote hosts differ from the config of the
> "cluster-start-host", then we still might have a problem when a remote
> host is booting: In this case '/etc/init.d/ceph start' is called and
> the remote host will use its local (different) config.
>
> I think a clear strategy would be:
> - All daemons always use their local config (no copying to /tmp).
> - If a cluster-start is initiated (/etc/init.d/ceph -a start) the
> script first checks if the configs on all cluster hosts are equal
> (maybe we need a content check here - not a literal check).
> - If the configs are not equal, the cluster-start command aborts with
> an error message.
>
> What do you think about that?
That makes a lot more sense, although I would suggest a warning instead of
an error. For cuttlefish, I'm hesitant to change behavior in the 11th
hour, but we should make this change going forward.
Care to send a patch? :)
sage
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
