このコンテンツは選択した言語では利用できません。
1.19. cman
1.19.1. RHBA-2009:1435: bug fix update
Note
This update has already been released (prior to the GA of this release) as errata RHBA-2009:1435
The Cluster Manager (cman) utility provides user-level services for managing a Linux cluster.
This update applies the following bug fix:
* in several places internally, cman assumed a transition message meant the node in question (or the sending node) was joining the cluster rather than just sending its current post-transition state. In some circumstances, this could lead to cman killing the wrong nodes. With this update, cman now checks the first_trans flag, which is set when a node first encounters another node in the cluster. Only if first_trans is set does cman now consider the node as joining the cluster. (BZ#518061)
Also, this update includes the following enhancement:
First, if a node was asked to remove a key (fence) for a device that it was not registered with, the node attempted to register with that device on-the-fly. With this update, when nodes are asked to remove a key from devices with which they are not registered, the fencing fails.
Second, for the common case of SAN environments with multiple Logical Unit Numbers (LUNs), the devices (LUNs) that can be unregistered must be ordered consistently on all nodes. Consistent ordering is not guaranteed by the Logical Volume Manager (LVM), however; device names can vary from node to node to prevent interleaving of fence operation among devices. With this update, the fence_scsi agent extracts the device name (pv_name) and Universally Unique Identifier (pv_uuid) and builds a hash keyed on the UUID (which is consistent on all nodes). This ensures devices are ordered consistently on each node.
Consequent to these two changes, the first node to fence removes the other node's key from the device or devices. The second node, now not registered with the device, is not able to fence the first. This allows fence_scsi to work in a 2-node cluster. (BZ#520823)
All cman users should install this updated package, which fixes this bug and enables users to use fence_scsi in a 2-node environment.
1.19.2. RHBA-2009:1516: bug-fix update
Note
This update has already been released (prior to the GA of this release) as errata RHBA-2009:1516
The Cluster Manager (cman) utility provides user-level services for managing a Linux cluster.
This update applies the following bug fix:
Add support for power cycle command to fence_ipmi, which doesn't shut down the BMC controller.
Old behavior is still the default, so nothing changes without a configuration change. Now there is a new method option that can have the value "cycle", which uses the ipmi power cycle command.
Example of usage:...
<fencedevices> <fencedevice agent="fence_ipmilan_new" ipaddr="1.2.3.4" login="root" name="ipmifd1" passwd="password" method="cycle" /> ...
Users are advised to upgrade to these updated cman packages, which resolve this issue.
1.19.3. RHBA-2009:1598: bug fix update
Note
This update has already been released (prior to the GA of this release) as errata RHBA-2009:1598
[Updated 4 Jan 2009] This update provides improved descriptions of both bug fixes included in this advisory, and especially the description for bug 529712. The packages included in this revised update have not been changed in any way from those included in the original advisory.
The Cluster Manager (cman) utility provides user-level services for managing a Linux cluster.
These updated cman packages provide fixes for the following bugs:
* when using device-mapper-multipath devices, registrations were only sent to the active path, which meant that, in the event of path failure, the node would be unable to access the device via the secondary path or paths because the device would not be registered with the secondary path(s). With this update, the presence of device-mapper-multipath devices is detected correctly, the right paths are discovered, and each path is registered, including secondary paths. (BZ#529712)
* when running the /etc/init.d/scsi_reserve init script to check for errors, such as an incorrect cluster.conf configuration, among others, upon finding an error the script did not print "[FAILED]" to standard output, as is convention for system services which encounter startup errors. With this update, the scsi_reserve init script has been fixed so that it prints "[FAILED]" to standard output when an error is encountered, and "[OK]" otherwise. Any errors encountered are logged to the system log. (BZ#530400)
All users of cman are advised to upgrade to these updated packages, which resolve these issues.
1.19.4. RHBA-2009:1622: bug-fix update
Note
This update has already been released (prior to the GA of this release) as errata RHBA-2009:1622
The Cluster Manager (cman) utility provides user-level services for managing a Linux cluster.
This update fixes the following bugs:
* qdiskd erroneously writes the message "qdiskd: read (system call) has hung for X seconds". (BZ 537157)
* Crash in fence_ipmilan when -M switch was not present on the command-line. (BZ 537157)
Users are advised to upgrade to these updated cman packages, which resolve these issues.
1.19.5. RHBA-2010:0266: bug fix and enhancement update
Updated cman packages that fix bugs and add enhancements are now available.
The Cluster Manager (cman) utility provides user-level services for managing a Linux cluster.
Changes in this update:
* fence_rsa fails to login with new RSA II firmware. (BZ#549473)
* fence_virsh reports vm status incorrectly. (BZ#544664)
* improve error messages from ccsd if there is a network problem. (BZ#517399)
* new fence agent for VMWare. (BZ#548577)
Note: this is a Tech Preview only.
* fence agent for HP iLO2 MP. (BZ#508722)
* fence agent for RSB ends with traceback. (BZ#545054)
* security feature for SNMP based agent: apc_snmp & ibmblade. (BZ#532922)
* change default timeout values for various fence agents. (BZ#549124)
* "Option -V" (show version) was not working in all fence agents. (BZ#549113)
* automatically configure consensus based on token timeout. (BZ#544482)
* add readconfig & dumpconfig to fence_tool. (BZ#514662)
* make groupd handle partition merges. (BZ#546082)
* groupd: clean up leaving failed node. (BZ#521817)
* scsi_reserve should always echo after failure. (BZ#514260)
* fence_scsi_test: add debug information. (BZ#516763)
* fence_scsi_test should not allow -c & -s options together. (BZ#528832)
* fix fence_ipmilan read from unitialized memory. (BZ#532138)
* make qdiskd stop crying wolf. (BZ#532773)
* fencing failed when used without telnet or ssh. (BZ#512343)
* APC changed product name (MasterSwitch -> Switched Rack PDU). (BZ#447481)
* fix invalid initalization introduced by retry-on option.
* broken device detection for DRAC3 ERA/O. (BZ#489809)
* fix case sensitivities in action parameter. (BZ#528938)
* fencing_snmp failed on all operations & traceback fix. (BZ#528916)
* accept unknown options from standard input. (BZ#532920)
* fence_apc unable to obtain plug status. (BZ#532916)
* timeout options added. (BZ#507514)
* better default timeout for bladecenter. (BZ#526806)
* the LOGIN_TIMEOUT value was too short for fence_lpar & the SSH login timed out before the connection could be completed. (BZ#546340)
* add missing-as-off option (missing blade/device is always OFF). (BZ#248006)
* make qdiskd "master-wins" node work. (BZ#372901)
* make qdisk self-fence system if write errors take longer than interval*tko. (BZ#511113)
* make service_cman.lcrso executable, so RPM adds it to the debuginfo pkg. (BZ#511346)
* don't check for xm command in cman init script: virsh is more appropriate. (BZ#516111)
* allow re-registering of a quorum device. (BZ#525270)
* fix fence_scsi, multipath & persistent reservations. (BZ#516625)
* cman_tool leave remove reduces quorum when no services are connected. (BZ#515446)
* fence_sanbox2 unable to retrieve status. (BZ#512947)
* gfs_controld: GETLK should free unused resource. (BZ#513285)
* allow IP addresses as node names. (BZ#504158)
* fence_scsi man page contains invalid option. (BZ#515731)
* fence_scsi support for 2 node clusters. (BZ#516085)
* Support for power cycle in fence ipmi. (BZ#482913)
* add option 'list devices' for fencing agents. (BZ#519697)
* add support for switching IPv4/IPv6. (BZ#520458)
* fence agent ends with traceback if option is missing. (BZ#508262)
* command line options to override default ports for different services, such as SSH & Telnet (i.e. -u option) were added. ( BZ#506928 )
Note: "-u" does not currently work with fence_wti. Other agents honor the port override command line options properly, however. ( BZ#506928 )
* force stdout close for fencing agents. (BZ#518622)
* support for long options. (BZ#519670)
* fix a situation where cman could kill the wrong nodes. (BZ#513260)
* fix support for >100 gfs & gfs2 file systems. (BZ#561892)
* fix a problem where 'dm suspend' would hang a withdrawn GFS file system. (BZ#570530)
* fix a problem where fence_snmp returned success when the operation failed. (BZ#573834)
* fencing support for the new iDRAC interface included with Dell PowerEdge R710 & R910 blade servers was added. (BZ#496748)
All cman users should install this update which makes these changes.