HP-UX 11i v3 Crash Dump Improvements

Page 16

5 Availability and Manageability Improvements

Many of the availability and manageability improvements make use of the native multi-pathing

provided in HP-UX 11i v3. The native multi-pathing automatically correlates the paths to a LUN

and notifies the dump subsystem of path offline and device offline events and other hardware

events so that the dump configuration can be automatically adjusted as needed. The events and

reconfiguration are available at run time, not at dump time.

5.1 Path Failover and Auto-reconfiguration

At run time, HP-UX 11i v3 supports native multi-pathing for normal I/O operations. At dump time,

I/O will go through only one of the selected paths to the device. The path used is normally the

one chosen by the infrastructure during dump configuration as discussed in section 3.3.

However, if the selected path goes offline or is disabled during run time operation, the dump

subsystem will be notified and a different path will be automatically selected (using the HBA

selection rules in section 3.3 with respect to the remaining available paths) and the dump device

reconfigured. As noted in section 4.2, the –l option to crashconf(1m) can be used to display the

currently configured lun path for each dump device.

This auto-reconfiguration gets invoked only when a currently selected path goes offline or is

disabled. Thus, if the currently selected path goes offline or is disabled and then later comes

back online or is re-enabled it will not automatically be re-selected.

Note: Dump functionality may be affected if the system administrator removes the system definitions

associated with device special files of configured dump devices (e.g., using the –a or –H options to the

rmsf(1m) command) instead of disabling the corresponding LUN or LUN path. Prior to removing, the

administrator should verify that the device special file is not configured for crash dump.

5.2 Avoiding off-line devices in the dump

When a dump device goes offline at run-time, the dump subsystem is notified. A device can go

offline if all the paths available to the LUN go offline or the LUN itself goes offline. The dump

subsystem will mark the device as offline and it will not be used while dumping, and dump unit

allocation and other operations at dump time will take this into account. If all the dump devices

configured have gone off-line, dump will be aborted. The crashconf(1m) command’s –v option

has been enhanced to display offline device information to the user. When a dump device goes

offline the dump subsystem will also log a message to the syslog file.

The following example shows crashconf –v output with an offline dump device:

If a lun or lun path goes offline at any time after the system crashes, the offline device or path will not be avoided during dump and the

dump will fail. However, this is a significant improvement over previous releases of HP-UX in which the failure window was not only after

the system crashed, but anytime after the dump device was last manually configured.