Legacy storage subsystems contain two controllers for redundancy. An online upgrade is done one controller at a time. DMP fails over all I/O to the second controller while the first controller is undergoing an Online Controller Upgrade. After the first controller has completely staged the code, it reboots, resets, and comes online with the new version of the code. The second controller goes through the same process, and I/O fails over to the first controller.
A/A type arrays require no special handling during this online upgrade process. For A/P, A/PF, and ALUA type arrays, DMP performs array-specific handling through vendor-specific array policy modules (APMs) during an online controller code upgrade.
If the array does not fully support NDU, all paths to the controllers may be unavailable for I/O for a short period of time. Before beginning the upgrade, set the dmp_lun_retry_timeout tunable to a period greater than the time that you expect the controllers to be unavailable for I/O. DMP retries the I/Os until the end of the dmp_lun_retry_timeout period, or until the I/O succeeds, whichever happens first. Therefore, you can perform the firmware upgrade without interrupting the application I/Os.
# vxdmpadm settune dmp_lun_retry_timeout=300