From: Brian J. Watson <Brian.J.Watson@hp...> - 2004-07-07 23:37:12
The focus of the 1.0.0-rc6 release was bug fixes, of which there were
many. Some highlights include a bug that prevented X Windows from
starting on a node that crashed and rebooted, incorrect CPU times for
remote processes, incorrect ttys for migrated processes, various
problems with the migrate(), rexec() and rfork() system calls, issues
with the shutdown and clusternode_shutdown commands, a deadlock in NFS
server, and bugs in the clusterwide /dev/pts. Kernel debugging code has
been disabled, to prevent annoying messages. A troubleshooting guide for
CVIP and HA-LVS has been added.
As always, you can find this new release on OpenSSI.org.
Here are the release notes:
The focus of the 1.0.0-rc6 release was bug fixes, of which there were many.
Before discussing the bug fixes, it should be noted that Bruce Walker
added a troubleshooting guide for CVIP and HA-LVS. Also, several
enhancements were made to other documentation.
John Hughes, together with Laura Ramirez, fixed bug #948584, which was a
problem with shared memory segments that prevented X Windows from
starting on a node that crashed and rebooted. Laura also made sure that
the CLUSTERNODE_BEST argument now works with rfork(). Another bug she
resolved was causing incorrect CPU times to be shown for remote process.
John Byrne fixed a panic in the migrate() system call, hangs in the
rexec() call, and various error path bugs in migrate(), rexec() and
rfork(). He resolved a bug in /proc that caused ps to incorrectly report
the tty of migrated processes. He also fixed ptys, yet again. Another
bug he corrected was causing infinite recursion whenever /dev/zero or
/dev/null was opened remotely. Finally, he eliminated a panic that was
caused by calling shutdown on a non-init node.
Speaking of shutdown, David Zafman fixed the clusternode_shutdown
command so that it does not break network service. He also corrected a
bug in the loadlevel RC script, and a deadlock that could occur when
running NFS server.
Brian Watson fixed a bug that occurred when ssi-ksync was run without a
default stanza specified in /etc/grub.conf. He also fixed the OpenSSI
upgrade script so that it could be run on the base kernel, although in
general this is not recommended, particularly if you have more than one
node with a local boot partition. He turned off debug code in the
kernel, since it was occasionally causing nuisance messages for users.
Finally, he removed the default serial console configuration, since it
was preventing boot messages from coming out on the physical consoles of
network boot nodes. For users with serial consoles, he wrote a README
about how to add back this support.
Last but not least, before leaving the OpenSSI project for advanced
education, Vinod Vijayarajan fixed bug #932596 by adding a test for a
corrupted /cluster/rootdisknode file, and exiting with an informative
error message if the file is either corrupted or missing.