Bug 1794260

Description of problem:
while hosted engine deployment with non 4k devices .The brick is been killed and the directory  '/gluster_bricks/engine/engine/'  is missing 

Version-Release number of selected component (if applicable):
RHGS 3.5.1 (6.0-25)
RHVH-4.3.8


Steps to Reproduce:
step1 : complete the gluster deployment with  engine brick on 4k and other brick on VDO 4k device  
step2 : Now complete the hosted engine deployment

Actual results:
directory /gluster_bricks/engine/engine/ is missing 
and the brick is not online 
Expected results:

The directory "/gluster_bricks/engine/engine/" should be present 
should not prompt ant error on the terminal

--- Additional comment from RHEL Program Management on 2019-12-30 09:15:15 UTC ---

This bug is automatically being proposed for RHHI-V 1.7 release at Red Hat Hyperconverged Infrastructure for Virtualization product, by setting the release flag 'rhiv‑1.7' to '?'.


This issue happens because of RHEL 7 systemd bug:
https://bugzilla.redhat.com/show_bug.cgi?id=1494014

So ideally, this issue happens with redeployment scenario.
Suppose, if some one umount /gluster_bricks/engine and in the attempt
to redeploy, again tries to mount engine brick at the same location - /gluster_bricks/engine,
then systemd silently mounts and umounts the brick, with rc=0

This is evident from:
<snip>
Jan 22 08:19:24 rhsqa-grafton10-nic2 systemd: Unit gluster_bricks-engine.mount is bound to inactive unit dev-disk-by\x2duuid-41e87589\x2d905e\x2d4fea\x
2daef2\x2dd4157129add3.device. Stopping, too.
Jan 22 08:19:24 rhsqa-grafton10-nic2 systemd: Stopped Migrate local SELinux policy changes from the old store structure to the new structure.
Jan 22 08:19:24 rhsqa-grafton10-nic2 systemd: Stopped target Local File Systems.
Jan 22 08:19:24 rhsqa-grafton10-nic2 systemd: Unmounting /gluster_bricks/engine...
Jan 22 08:19:24 rhsqa-grafton10-nic2 kernel: XFS (dm-13): Unmounting Filesystem
Jan 22 08:19:24 rhsqa-grafton10-nic2 systemd: Unmounted /gluster_bricks/engine.
Jan 22 08:19:30 rhsqa-grafton10-nic2 python: ansible-mount Invoked with src=UUID=7702205b-2c11-49b9-8a0e-fb3bbe6ed5db dump=None boot=True fstab=None pa
ssno=None fstype=xfs state=mounted path=/gluster_bricks/test backup=False opts=inode64,noatime,nodiratime
Jan 22 08:19:30 rhsqa-grafton10-nic2 kernel: XFS (dm-24): Mounting V5 Filesystem
Jan 22 08:19:30 rhsqa-grafton10-nic2 kernel: XFS (dm-24): Ending clean mount
Jan 22 08:19:30 rhsqa-grafton10-nic2 systemd: Unit gluster_bricks-test.mount is bound to inactive unit dev-disk-by\x2duuid-17f47f26\x2d61d8\x2d40f3\x2d
b47e\x2dd56512794a7c.device. Stopping, too.
</snip>

In this case, remember /gluster_bricks/engine is not yet mounted,
but ansible is unaware of this. When creating the gluster volume, brick is created
directly on the root filesystem /gluster_bricks/engine/engine

Now when deployment reaches HE deployment, some step here calls - 'systemctl daemon-reload'
which leads to the mount point to appear from nowhere and its mounted over the existing
contents. This is evident from:

<snip>
Jan 22 08:22:26 rhsqa-grafton10-nic2 systemd: gluster_bricks-test.mount: Directory /gluster_bricks/test to mount over is not empty, mounting anyway.
Jan 22 08:22:26 rhsqa-grafton10-nic2 systemd: Mounting /gluster_bricks/test...
Jan 22 08:22:26 rhsqa-grafton10-nic2 systemd: gluster_bricks-engine.mount: Directory /gluster_bricks/engine to mount over is not empty, mounting anyway.
Jan 22 08:22:26 rhsqa-grafton10-nic2 systemd: Mounting /gluster_bricks/engine...
Jan 22 08:22:26 rhsqa-grafton10-nic2 kernel: XFS (dm-24): Mounting V5 Filesystem
Jan 22 08:22:26 rhsqa-grafton10-nic2 kernel: XFS (dm-13): Mounting V5 Filesystem
Jan 22 08:22:26 rhsqa-grafton10-nic2 kernel: XFS (dm-24): Ending clean mount
Jan 22 08:22:26 rhsqa-grafton10-nic2 kernel: XFS (dm-13): Ending clean mount
Jan 22 08:22:26 rhsqa-grafton10-nic2 systemd: Mounted /gluster_bricks/engine.
Jan 22 08:22:26 rhsqa-grafton10-nic2 systemd: Mounted /gluster_bricks/test.
</snip>

So now /gluster_bricks/engine becomes the mount point and its empty. This is how the 'engine' dir under /gluster_bricks/engine disappears


Workaround from the bug is to do 'systemctl daemon-reload' before mounting the bricks


Treating this bug as BLOCKER as it may affect customers, while redeploying.