After upgrading one of our hosts from 5.1 U1 to ESXi 5.5, I noticed server had lost access to its local volume. Server has 2x300GB SAS disks configured with two RAID 1 volumes. One of them is a 10GB VMFS where ESXi is installed. The other volume is just sitting there as empty space.
Although I can't see the disks, server boots up fine and VMs are able to be powered up and run. They all sit on shared storage. The host has PowerPath/VE 5.8 installed, which is supported on both ESXi 5.1 and ESXi 5.5. This is a HP DL380p gen 8 server, with latest SPP installed.
When you select host / Configuration / Storage adapters and go to Smart Array P420i ( server's storage controller model), ESXi sees vmhba1 and no devices are shown. When I click Paths, I see vmhba1:C0:T0:L1 as DEAD.
During ESXi upgrade (done interactively for troubleshooting purposes) I can see both volumes as potential devices for installing ESXi.
As part of troubleshooting I tried to get rid of PowerPath, but I was thrown errors mentioning VIBs could not be removed live. This happened regardless of host being on maintenance mode or not.
It turns out, after so much THIS and THAT, that ESXi 5.5 does not work well with PowerPath running on HP or Hitachi host (issue with local storage) or PowerPath Unsupported array models (issue with SAN volumes), due to a change in the VMware PSA.
Thanks to this a nice article I could find a way around it: http://www.virten.net/2013/11/esxi-5-5-with-powerpathve-5-9-inaccessible-local-datastores/#comment-84947
EMC has an article (000173816) which explains the issue a bit futher, and states: PowerPath attempts to claim all the devices that matches Vendor and product ID. During claim, PPVE internally matches exact array model number of supported devices (As per PowerPath Support Matrix, ESM), if exact model number of the device is not matched then PowerPath will not claim that device.When PSA found that a device is not claimed by PPVE, it offers that device to another MPP. However, ESXi 5.5 PSA doesn't offer unclaimed devices to other Multipath-plugin - this results in data unavailability for those devices, as no multipathing plugin will manage those devices.
As per same article, there currently is no fix for this issue and VMware says it will be fixed when U1 is out. The only workaround is to create a claimrule which will claim such devices prior to PowerPath trying to do the same. Another option would be removing PowerPath rules for affected vendor, but this would mean storage from those vendors would no longer be able to use PowerPath.
I couldn't find any VMware articles mentioning this issue so far.