Fix glustershd detection on volume restart

On NetBSD and FreeBSD, doing a 'gluster volume start $volume force' causes
NFS server, quotad, snapd and glustershd to be undetected by glusterd once the
volume has restarted. 'gluster volume status' shows the three processes
as 'N' in the online column, while they have been launched successfully.

This happens because glusterd attempts to connect to its child processes
just between the child does a unlink() on the socket in
__socket_server_bind() and the time it calls bind() and listen().
Different scheduling policy may explain why the problem does not happen
on Linux, but it may pop up some day since we make no guaranteed
assumptions here.

This patchet works this around by introducing a boolean
transport.socket.ignore-enoent option, set by nfs and glustershd,
which prevents ENOENT to be fatal and cause glusterd to retry and
suceed later. Behavior of other clients is unaffected.

BUG: 1129939
Change-Id: Ifdc4d45b2513743ed42ee235a5c61a086321644c
Signed-off-by: Emmanuel Dreyfus <manu@netbsd.org>
Reviewed-on: http://review.gluster.org/8403
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Niels de Vos <ndevos@redhat.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>

0 files changed, 0 insertions, 0 deletions