Red Hat Cluster Suite 4 RHEL4 U2 Release Notes
Copyright(c) 2005 Red Hat, Inc.
-------------------------------------------------------
Introduction
The following topics are covered in this document:
o Changes to Red Hat Cluster Suite 4
o Important Notes
o Bugs Fixed in the Release
o Related Documentation
Changes to Red Hat Cluster Suite 4
Red Hat Cluster Suite 4 contains no new features for Red Hat Enterprise
Linux 4 Update 2.
Important Notes
Cluster Upgrade Consistency Requirement
All nodes in a cluster must be upgraded to the same update version. Nodes
running different versions of cluster packages on the same cluster are not
supported.
Cluster Name Length Restriction
If the cluster name exceeds 15 characters, the cluster could hang. A bug is
open on this problem (Bugzilla Bug 169139) but a fix is not available for
this release.
Workaround: When naming a cluster, restrict the cluster name length to
no more than 15 characters.
Adjusting Heartbeat Rate
To tune CMAN heartbeat rates for when nodes join a cluster as well as the
time elapsed before a node is determined dead, perform the following steps:
Run the following commands as root:
echo 10 > /proc/cluster/config/cman/hello_timer
echo 100 > /proc/cluster/config/cman/deadnode_timeout
The first command above configures the rate at which a cluster node thread
sends the "HELLO" message to join cluster to 10 seconds (the default value is
5). The second command configures the timeout period (in seconds) to
determine when a node is considered dead to 100 seconds (from the default
21 seconds).
Apply these values to each node prior to joining them to a cluster with
the "cman_tool join" command.
Bugs Fixed in the Release
o 164577 -- Piranha does not detect link-down events which would prevent
proper operation
o 159965 -- Cluster Management window does not refresh after configuration
change and traceback
o 160127 -- Clicking "Send to Cluster" does not clear "modified" indicator
from "Current File Name" box
o 160240 -- removing /etc/cluster out from under the GUI will cause a
traceback when saving
o 160264 -- garbage cluster.conf files cause the GUI to die at startup
o 161344 -- GUI fails to start with valid conf file: error in in
resolve_fence_instance_types
o 161974 -- creates borken fence_manual configuration
o 163237 -- GUI allows cluster node and fence device to have the same name
o 163673 -- Icon path in system-config-cluster dekstop file is wrong
o 164407 -- system-config-cluster generates poorly formed XML in cluster.conf
o 164857 -- redhat-config-cluster does not show nodes after transfering services
between nodes
o 164936 -- traceback when attempting to edit a gnbd fence device
o 165130 -- Fence levels are created under <multicast> tag
o 165140 -- Traceback when fencing cluster node and no fence device exists
o 165169 -- GNBD fence configuration is broken
o 165184 -- Removal of Failover Domain doesn't update Services
o 166140 -- system-config-cluster doesn't allow force_unmount to be set for FS
o 166663 -- global name 'gfs_name' is not defined when editing a filesystem
resource
o 167429 -- Autostart of a service gets automatically set
o 159637 -- local variable used before set
o 157327 -- ip.sh ignores SLAVE flag when looking for interfaces
o 159767 -- service operations on non-existent services are processed anyway
o 162824 -- Shared nfsclient references by multiple filesystems don't get
monitored
o 162936 -- rgmanager segfaults with certain configurations using inheritance
o 163651 -- ip.sh's monitor_link is improperly handled
o 155478 -- unnecessary/ugly output from fence and clvm init script in syslog
o 159685 -- fenced init script should not faiil if it's a gulm cluster
o 162805 -- fence_wti does not fence properly
o 164627 -- fence_ipmilan does not have a default operation; should be "reboot"
o 158592 -- cman master confusion due to recovery
o 159615 -- cman schedules within spinlock
o 159628 -- sm join/leave bug
o 162014 -- gfs mounts allowed before fencing completes
o 163587 -- kernel: CMANsendmsg failed: -101
o 154990 -- dlm returns incorrect result to userspace on lock error
o 157295 -- Connecting to an existing lockspace breaks reference count
o 155478 -- unnecessary/ugly output from fence and clvm init script in syslog
o 155478 -- unnecessary/ugly output from fence and clvm init script in syslog
o 157094 -- No indication for result of config file propagation with "Send to
Cluster" button
o 162422 -- Recovery problem when the gulm master node is fenced
For more information about bugs fixed in this release and other information
about the release, refer to the following errata on RHN (https://rhn.redhat.com/):
o RHBA-2005:798 -- iddev bug fix update
o RHBA-2005:778 -- dlm bug fix update
o RHBA-2005:746 -- piranha bug fix update
o RHBA-2005:753 -- system-config-cluster bug fix update
o RHBA-2005:777 -- magma-plugins bug fix update
o RHBA-2005:775 -- magma bug fix update
o RHBA-2005:738 -- rgmanager bug fix update
o RHBA-2005:737 -- fence bug fix update
o RHBA-2005:734 -- cman-kernel bug fix update
o RHBA-2005:735 -- dlm-kernel bug fix update
o RHBA-2005:739 -- cman bug fix update
o RHBA-2005:736 -- ccs bug fix update
o RHBA-2005:733 -- gulm bug fix update
Related Documentation
The following related documentation is available at
http://www.redhat.com/docs/manuals/csgfs/:
o Red Hat Cluster Suite Configuring and Managing a Cluster
(for Red Hat Cluster Suite 4)
o Release notes for Red Hat Cluster Suite 4
o Red Hat GFS 6.1 Administrator's Guide
o Release notes for Red Hat GFS 6.1