Bug 1283542

+++ This bug was initially created as a clone of Bug #1282915 +++
+++                                                           +++
+++ Use this bug to send a fix for the release-3.5 branch.    +++

Description of problem: Gluster does not reload gracefully upon server restart


Version-Release number of selected component (if applicable):
3.2+

How reproducible:
     

Steps to Reproduce:
1. Install GlusterFS as per docs on Centos7.
2. Create and start volume.
3. Restart server.

Actual results:
rpcbind service will not be active.

Expected results:
gluster will have registered with rpcbind

Additional info:

/usr/lib/systemd/system/glusterd.service has the following:
"[Unit]
Description=GlusterFS, a clustered file-system server
After=network.target rpcbind.service
Before=network-online.target"

Changing to:
[Unit]
Description=GlusterFS, a clustered file-system server
Requires=rpcbind.service
After=network.target 
Before=network-online.target

Corrects the issue.

--- Additional comment from Red Hat Bugzilla Rules Engine on 2015-11-17 20:29:40 CET ---

This bug is automatically being proposed for the current z-stream release of Red Hat Gluster Storage 3 by setting the release flag 'rhgs‑3.1.z' to '?'. 

If this bug should be proposed for a different release, please manually change the proposed release flag.

--- Additional comment from Kaushal on 2015-11-18 06:55:39 CET ---

Thanks for reporting this. Turns out we've misunderstood the meaning of the `After` option.

According to the systemd.unit(5) man page, the `After` option only sets up ordering. If unit foo.service has the option `After=bar.service`, the option only ensures that if both foo and bar are started together, foo is started after bar. It doesn't ensure that starting only foo starts bar.

The `Requires` option only sets up the requirement, but doesn't set up ordering. If foo.service `Requires` bar.service, it is only guaranteed that starting foo will also start bar, but not necessarily in order.

The man page suggests that if a unit requires that another unit be started before itself, then both `After` and `Requires` must be used.

I've submitted a change for review that fixes this at http://review.gluster.org/12605

--- Additional comment from Vijay Bellur on 2015-11-18 11:21:10 CET ---

COMMIT: http://review.gluster.org/12605 committed in master by Atin Mukherjee (amukherj) 
------
commit 23440a73bc348bbc3bb43ec397f0639ee45865fc
Author: Kaushal M <kaushal>
Date:   Wed Nov 18 11:21:13 2015 +0530

    glusterd.service: Ensure rpcbind is started before glusterd
    
    Having the rpcbind.service under the `After` option only ensures that
    glusterd.service is started after rpcbind.service if both are
    enabled/started at the same time. It doesn't ensure that starting
    glusterd.service will start rpcbind.service.
    
    The systemd.unit(5) man page suggests to use both the `Requires` and
    `After` options to ensure that rpcbind is started before glusterd,
    whenever glusterd is started.
    
    BUG: 1282915
    Change-Id: Iee69965486be08711299aba235f7b00c3e2fe7e9
    Signed-off-by: Kaushal M <kaushal>
    Reviewed-on: http://review.gluster.org/12605
    Reviewed-by: Atin Mukherjee <amukherj>
    Reviewed-by: Anand Nekkunti <anekkunt>
    Tested-by: Gluster Build System <jenkins.com>
    Tested-by: NetBSD Build System <jenkins.org>