Troubleshooting the SDK installation

This topic describes what to do if things go wrong when using the cellsdk script.

YUM continues to install the SDK packages even if some of the RPMs were not completely downloaded from the BSC Web site. The failure messages from YUM do not clearly state this failure. To verify if all files were installed correctly, type the command:

# ./cellsdk verify

In the list output by this command, verify that all default RPMs were installed. If they were not, retype the ./cellsdk install command. YUM will attempt to download any required RPMs that were not downloaded during a past attempt to install the SDK. YUM resumes the download process from the previous failure point.

Sometimes YUM operates incorrectly. It writes files in /var/cache/yum, and sometimes these no longer reflect the correct state of the command. If this happens, type the command:

# yum clean metadata

This will remove the incorrect status files.

There are other options to YUM that are useful to use if things go wrong. If the previous command did not restore correct operation of YUM, try typing the command:

# yum clean all

This will remove additional state files, and might cause the YUM installation process to succeed on the next invocation.

If the preceding commands do not restore correct operation of the YUM installation process, manually remove the cached state files. To do this, type the following command:

# rm -rf /var/cache/yum/CellSDK

To see more information about YUM, set the options:

debuglevel=10
errorlevel=10

in /etc/yum.conf. You an get an equivalent result by adding the string -d 10 -e 10 to any YUM command.

You can specify the flags -v for verbose output, or -vv for very verbose output when typing the cellsdk script. YUM writes a log to /var/log/yum.log. The cellsdk install script writes a log to /var/log/cellsdk.log. Looking at these files might provide helpful information about what went wrong.

As the last step of the install, the cellsdk script locates the SDK version of some open source RPMs. The RPMs may include some, none or all of the blas, blas-debuginfo, blas-devel, numactl, numactl-devel, oprofile, and oprofile-debuginfo packages. It places the RPMs it finds in the /tmp/cellsdk/openSrc directory. Then cellsdk removes the installed version of each RPM, and installs the SDK version of the RPM in its place. Finally, it creates a done file in the /tmp/cellsdk/openSrc directory that prevents this install step from repeating if cellsdk is run again. If you want to reinstall the SDK version of the RPMs at a later time, you can remove the done file and run the cellsdk install command again. See Preventing automatic updates from overwriting SDK components for more information.

A potential problem is that the Fedora 7 version of blas is installed instead of the SDK version because YUM sees that it has a higher version number. To fix this problem, delete the /tmp/cellsdk/openSrc/done file and reinstall blas as described above.