RED HAT Cluster Suite RHEL3 U5 Release Notes

Copyright(c) 2005 Red Hat, Inc.
	-------------------------------------------------------

Introduction:

   The following topics are covered in this document:

     o New Red Hat Cluster Suite Features for RHEL3 U5
     o Defects Fixed in the Release
     o Related Documentation


New Red Hat Cluster Suite Features for RHEL3 U5

o STONITH support for various Bull systems

  A STONITH module for the Bull Novascale FAME architecture has been added.
  Support is made possible by accessing the Platform Administration
  Processor (PAP) management console. The configurable parameters in the 
  Cluster Configuration Tool (redhat-config-cluster) GUI is as follows:

pap:
 IP Address:   <-- IP address of PAP management console
 Port:         <-- Domain (the Bull virtual host)
 Username:     <-- Administrative user capable of issuing power on/off
                   commands for the given Domain
 Password:     <-- Password to authenticate the administrative username

  Additionally, STONITH module support has been added for any BULL machine 
  that uses Intelligent Platform Management Interface over local area 
  networks (IPMI-over-LAN).  The configurable parameters in the Cluster 
  Configuration Tool (redhat-config-cluster) GUI is as follows:

ipmilan:
 IP Address:   <-- IP address of the node's IPMI port
 Port:         <-- Unused
 Username:     <-- Administrative user capable of issuing power on/off
                   commands to the given IPMI port
 Password:     <-- Password to authenticate the administrative username


o STONITH support for IBM BladeCenter

  A STONITH module for the IBM BladeCenter has been added. Support is made 
  possible by accessing the BladeCenter console via telnet. The configurable 
  parameters in the Cluster Configuration Tool (redhat-config-cluster) GUI 
  is as follows:

bladecenter:
 IP Address:   <-- IP address of management blade
 Port:         <-- Blade number
 Username:     <-- Administrative user capable of issuing power on/off
                   commands on the Blade Center
 Password:     <-- Password to authenticate the administrative username


o Per-IP address link monitoring for clusters where heartbeating is done
  over a private LAN

  A checkbox that enables link monitoring has been added to the IP address 
  dialog box of the Cluster Configuration Tool (redhat-config-cluster)


Defects Fixed in the Release

o Service number mismatches in pathological cases are retried
  
  Users periodically received a "service number mismatch 4,6" at the
  <emerg> log level.  The issue has been isolated to the Red Hat Enterprise
  Linux kernel, and a retry was added so that clumanager only prints the 
  message if it receives bad data multiple instances in a row. Cluster 
  development is working with Kernel engineers to find the root cause of the
  issue.


o STONITH subsystem left temporary files unaccounted for

  The STONITH subsystem would pass configuration information in files
  in the /tmp directory.  When a new STONITH device (like a fence agent)
  was configured, it would write all the information necessary to use that 
  device into a file in /tmp.  The temporary file was mode 0600, but still 
  contained passwords and login names for the power switch.  The temporary 
  files are now removed after being added to the cluster configuration.


o The clumanager init script missed a default case

  The clumanager service init script did not cause an error message in default
  cases (such as entering an invalid option or not entering an option at all)
  as it should have.


o Routed IP tiebreaker addresses were not properly handled in some cases

  The length of the ICMP packet (stored in the ICMP packet) was incorrect,
  which caused pings to not get responses from various network router models.


o Random restarts of services with bogus status returns

  Clumanager would restart a service after receiving an invalid status return.
  Cluster developers could not reproduce the issue, but isolated it to a wait()
  call that was interrupted by a signal.


o Upgrade postuninstall script fails if /etc/cluster.xml did not exist

  An error message was reported if the /etc/cluster.xml configuration file was 
  not found  during an upgrade of Red Hat Cluster Suite packages.


o Signals blocked when starting user services

  Cluster services that used signals for communication/wakeups were broken
  because of blocked signals during service initialization.


o Force unmount killed too many processes

  Cluster Manager kills all processes using a specified mount, regardless 
  of whether or not Cluster Manager started the processes. Cluster Manager
  must kill all processes using a file system in order to unmount it. 
  However, an issue was found in which a force unmount killed too many 
  processes.

  Suppose there are two different mounts:

/var/mail         <-- managed by cluster
/var/mail-backup  <-- not managed by cluster; local copy

  When Cluster Manager killed the processes using /var/mail while trying to
  force-unmount it, it also killed all processes using /var/mail-
  backup.