4.13. cman

Bug Fix

BZ#986981

Due to incorrect detection of newline characters during an SSH connection, the fence_drac5 agent could terminate the connection with a traceback error when fencing a Red Hat Enterprise Linux cluster node. Only the first fencing action completed successfully but the status of the node was not checked correctly. Consequently, the fence agent failed to report successful fencing. When the "reboot" operation was called, the node was only powered off. With this update, the newline characters are correctly detected and fencing works as expected.

Bug Fix

BZ#951049

Under some circumstances cman can return the fence daemon for the /dev/zero file, which is always active, and if the client application stores this instead of the one it expects then it will loop forever. This update addresses the problem by making sure that the fence daemon refreshes the file descriptor on each operation.

Bug Fixes

BZ#714042

Prior to this update, ccs_tool could fail due to a segmentation fault in two specific events. First, the input file was the same as the output file and options were specified with a subcommand. Second, the update command was used without an input file. With this update, ccs_tool has been fixed to report the appropriate error message without a segmentation fault.

BZ#854183

Previously, fence_xvm, a fencing agent commonly used in virtual deployments, did not respect the delay parameter as specified in the cluster.conf file. With this update, the delay support has been added and the fence_xvm agent now works only after the specified delay.

BZ#856214

Under some circumstances, cman returned the file descriptor for the /dev/zero file, which is always active. If the client application stored this /dev/zero file instead of the one it expected, cman entered an infinite loop. This update addresses the bug by making sure that the fence daemon refreshes the file descriptor on each operation.

BZ#876731

In a two node cluster where the network was not always reliable, and a token was lost but recovered before fencing was complete, the two nodes simultaneously killed each other. To fix this bug, only one node has been allowed to kill the other in a deterministic fashion. A cluster restart, one node at a time, is required for this change to take effect.

BZ#881217

Prior to this update, fenced, the daemon in charge of evicting cluster nodes, could suffer from a time-out while communicating with the ccsd cluster configuration daemon. The fenced daemon has been reconnected to ccsd when certain operations take too long to complete, thus fixing this bug.

BZ#883816

Previously, man pages had executable flags set. The manual pages permissions have been edited to agree with FHS standard, thus fixing this bug.

BZ#904195

Prior to this update, the end-of-line automatic character detection worked properly only when using ssh on certain Dell DRAC devices. With this update, the end-of-line character detection has been fixed and fencing works as expected.

BZ#886612

Prior to this update, APC power switches with firmware version 5.x were not supported. Consequently, the fence_apc fencing script could not login to the device. With this update, new firmware is supported and the fencing script is able to log in.

BZ#961119

Previously, the fence agent for Intel IPMILAN standard did not return exit code correctly when using the 'cycle' method. A patch has been provided to fix this bug and exit codes are now returned as expected.

BZ#963251

Prior to this update, the cman(5) man page did not document how to enable detailed logging in the CMAN subsystem. This update documents this facility in more detail.

Bug Fix

BZ#923861

In a two node cluster, under some circumstances where the network was not always reliable, and a token was lost but recovered before fencing was complete, then the two nodes simultaneously killed each other. With this update, the problem is solved by allowing only one node to kill the other in a deterministic fashion. A cluster restart, one node at a time, is required for this change to take effect.