Bug 170926 - gfs does not respect loss of quorum (split brain?)
Summary: gfs does not respect loss of quorum (split brain?)
Keywords:
Status: CLOSED DUPLICATE of bug 162725
Alias: None
Product: Red Hat Cluster Suite
Classification: Retired
Component: gfs
Version: 4
Hardware: All
OS: Linux
medium
high
Target Milestone: ---
Assignee: David Teigland
QA Contact: GFS Bugs
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2005-10-15 19:45 UTC by Axel Thimm
Modified: 2010-01-12 03:08 UTC (History)
0 users

Fixed In Version:
Clone Of:
Environment:
Last Closed: 2005-10-31 20:45:28 UTC
Embargoed:


Attachments (Terms of Use)

Description Axel Thimm 2005-10-15 19:45:32 UTC
Description of problem:
When a cluster looses quorum, cluster members can still read/write from/to a gfs
mount.

Version-Release number of selected component (if applicable):
kernel-smp-2.6.9-22.EL
GFS-kernel-smp-2.6.9-42.1
cman-kernel-smp-2.6.9-39.5
GFS-6.1.2-0
cman-1.0.2-0

How reproducible:
always

Steps to Reproduce:
1.create a three node cluster and use a shared gfs mount.
2.shutdown the cluster services on two nodes
3.continue using the gfs mount on the third node
  
Actual results:
Read/write operations are possible even though the cluster has lost quorum.

Expected results:
gfs should respect cman's "quorum lost, blocking activity" (or cman should
freeze the gfs mount until quorum is regained).

Additional info:
This bug possibly explains the split brain behaviour of bug #169693.

Comment 1 David Teigland 2005-10-31 20:45:28 UTC
This is the correct behavior as explained in bz 162725.
If a cluster loses quorum but none of the nodes mounting
gfs fs "foo" have failed, then there's no need to block
locking for foo (that's only needed to safely do journal
recovery.)  While the cluster is inquorate, no nodes are
permitted to mount or unmount any gfs file systems, though.

If there's a cluster partition, nodes on the inquorate side
will be fenced by the quorate side before any nodes on the
quorate side can mount any gfs file systems.  So, if there
are nodes using "foo" in an inquorate cluster, and a cluster
partition gains quorum, those nodes using foo will be fenced
before any nodes in the quorate cluster mount foo.

*** This bug has been marked as a duplicate of 162725 ***


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