Bug 1173290 - Add "find_multipaths yes" to the default multipath.conf
Summary: Add "find_multipaths yes" to the default multipath.conf
Keywords:
Status: CLOSED WONTFIX
Alias: None
Product: Red Hat Enterprise Virtualization Manager
Classification: Red Hat
Component: vdsm
Version: 3.5.0
Hardware: Unspecified
OS: Unspecified
medium
high
Target Milestone: ---
: 3.5.4
Assignee: Nir Soffer
QA Contact: lkuchlan
URL:
Whiteboard: node
: 1185759 1207543 (view as bug list)
Depends On:
Blocks: 1184995 1207543
TreeView+ depends on / blocked
 
Reported: 2014-12-11 20:19 UTC by Fabian Deutsch
Modified: 2016-02-10 20:06 UTC (History)
16 users (show)

Fixed In Version:
Doc Type: Bug Fix
Doc Text:
Clone Of:
: 1294910 (view as bug list)
Environment:
Last Closed: 2015-05-27 12:44:53 UTC
oVirt Team: Node
Target Upstream Version:
Embargoed:


Attachments (Terms of Use)


Links
System ID Private Priority Status Summary Last Updated
Red Hat Bugzilla 1075166 0 medium CLOSED [RFE] Provide drivers for Dell Shared PERC8 RAID Controller with RHEVH 2021-02-22 00:41:40 UTC

Internal Links: 1075166

Description Fabian Deutsch 2014-12-11 20:19:32 UTC
Description of problem:
To be on the safe side, we should set "find_multipaths yes" in multipath.conf on 7.0 and 6.6.

Otherwise the multipath behavior on RHEL 6 and RHEL 7 will differ.

On RHEL 6 mutlipath will grab all devices  except when they are blacklisted.
On RHEL 7 multipath will grab no  devices  until 1) the same serial appears twice 2) explicitly named 3) (forgot)

Please also refer to bug 1152948

Comment 1 Yaniv Lavi 2015-01-20 18:59:46 UTC
Why was this changed in RHEL 7?
What are we considering as the correct behavior?

Comment 2 Fabian Deutsch 2015-01-21 10:10:15 UTC
Thsi change was needed to fix the usage of multipath. In summary we are now much closer to what RHEL has, and multipath now only claims devices if it sees that there is more than one path to a single device.

The long answer is: Before 6.6 and 7.0 our multipath config tried to claim (to make a multipath device) all devices, independently if there was one or several paths to the device.

This lead to all kinds of problems, one: When the system came up, multipath tried to claim the cd-drive (i.e. sr0) and make a mpath device (i.e. dm-0), but dracut also wanted to boot of it.
Sometimes dracut won, and the boot was successful, and sometimes multipath was faster, and the boot failed, because sr0 which was used for booting, was claimed by multipath to provide dm-0.

Another case which lead to problems was the installation: During the installation partitions and lvms get added to the host. For each partition and LV creation we had a race between the installer discovering finding the device node for the new partition/lv and multipath creating a mpath device for the new partition/LV. Which is basically the same problem as above: If the installer figured out to use the raw device (hda), but multipath claimed it (dm-0) before the installer would actually operate on it, the installer would fail, because when the installer started to operate, the raw device hda was claimed by multipath. So again a race. (Some calls related to partitioning werw also causing this problem, so the fix included fixing multipath and removing the calls).

After all the situation is now that multipath devices will only be claimed if:
1. The wwid of an mpathd evice is given in the kernel commandline (this is used for all boots)
2. Or multipath sees that a device has more than one path (which is effectively: Many devices popup in /dev which have the same serial).
3. The wwid of the device is listed in the wwids file

Comment 3 Fabian Deutsch 2015-01-21 10:14:23 UTC
I might have gotten you wrong in comment 2, as you were asking specifically about RHEL, so here my RHEL answers:

(In reply to Yaniv Dary from comment #1)
> Why was this changed in RHEL 7?

This change was already done in RHEL 6.6, maybe even 6.5.
To my understanding it was needed to stabilize multipath.

> What are we considering as the correct behavior?

The recommended (and to my understanding thus correct) behavior is to use "find_multipaths yes".

Ben, maybe you got more comments to the questions above.

Comment 4 Ben Marzinski 2015-01-21 17:37:34 UTC
We enabled find_multipaths in the default /etc/multipath.conf file in RHEL7 because without it, setting up multipath correctly pretty much required manually editting the blacklists section in /etc/multipath.conf to make sure that multipath was only trying to access the correct devices.  Turning on find_multipaths makes it easier for users to setup multipath on their system.  Assuming that they are using a device that auto-configures, they just enable multipathing, and it will only grab devices with multiple paths, and it will automatically configure them correctly. This makes it work "out of the box" on most setups.

Comment 5 Fabian Deutsch 2015-01-26 09:01:07 UTC
*** Bug 1185759 has been marked as a duplicate of this bug. ***

Comment 6 Eyal Edri 2015-02-25 08:39:36 UTC
3.5.1 is already full with bugs (over 80), and since none of these bugs were added as urgent for 3.5.1 release in the tracker bug, moving to 3.5.2

Comment 8 Eyal Edri 2015-04-28 11:23:23 UTC
moving to 3.5.4 due to capacity planning for 3.5.3.
if you believe this should remain in 3.5.3, please sync with pm/dev/qe and a full triple ack for it. also - ensure priority is set accordingly to the bug status.

Comment 9 Fabian Deutsch 2015-05-05 13:39:32 UTC
*** Bug 1207543 has been marked as a duplicate of this bug. ***

Comment 10 Fabian Deutsch 2015-05-27 12:44:53 UTC
According to the email discussion I'm now closing this bug, because vdsm assumes that all devices are handled like multipath devices.


Note You need to log in before you can comment on or make changes to this bug.