I recently ran into an issue where a customer was seeing “SCSI Reservation” messages on one of their hosts. I asked for a log bundle (KB 653), and they had sent me one. I was looking over the vmkernel logs (/var/log/vmkernel) and I saw the following:

vmkernel.9:Mar 10 01:00:45 esx_host vmkernel: 97:00:40:02.937 cpu15:14570)VMW_SATP_SVC: satp_svc_UpdatePath: Failed to update path "vmhba2:C0:T1:L80" state. Status=SCSI reservation conflict

Looking at the frequency of the message and what LUN kept having the issue, I saw the following:

me@my_server:log$ grep 'reservation conflict' vmkernel* | awk '{print $13}' | sort | uniq -c
5432 "vmhba1:C0:T0:L80"
5432 "vmhba1:C0:T1:L80"
5432 "vmhba1:C0:T2:L80"
5432 "vmhba1:C0:T3:L80"
5432 "vmhba2:C0:T1:L80"
5432 "vmhba2:C0:T2:L80"
5432 "vmhba2:C0:T3:L80"

It looks like it’s only happening on Lun 80 and here is the info for that LUN:

First here is the ‘esxcli nmp device’ output for that lun:

Runtime Name: vmhba1:C0:T3:L80
Device: naa.600507680180863d6800000000000062
Device Display Name: IBM Fibre Channel Disk (naa.600507680180863d6800000000000062)
Group State: active
Array Priority: 0
Storage Array Type Path Config: SATP VMW_SATP_SVC does not support path configuration.
Path Selection Policy Path Config: {current: no; preferred: no}

Secondly, here is the esxcfg-scsidevs output for that LUN:

Device Type: Direct-Access
Size: 30720 MB
Display Name: IBM Fibre Channel Disk (naa.600507680180863d6800000000000062)
Multipath Plugin: NMP
Console Device: /dev/sdcs
Devfs Path: /vmfs/devices/disks/naa.600507680180863d6800000000000062
Vendor: IBM Model: 2145 Revis: 0000
SCSI Level: 6 Is Pseudo: false Status: on
Is RDM Capable: true Is Removable: false
Is Local: false
Other Names:
VAAI Status: unknown

Lastly, here is the fdisk output for that LUN:

Disk /dev/sdcs: 32.2 GB, 32212254720 bytes
256 heads, 63 sectors/track, 3900 cylinders, total 62914560 sectors Units = sectors of 1 * 512 = 512 bytes

Device Boot Start End Blocks Id System
/dev/sdcs1 1 4294967295 2147483647+ ee EFI GPT

From above, we could gather that it’s a Fibre-Channel Lun of about 30GB, but the interesting thing came from the fdisk output. It had a GPT partition on that LUN, which usually means that this LUN is used as an RDM and is directly presented to the Guest OS.

I asked the customer what that RDM was used for, and to check if there is any special application within the Guest OS that would use SCSI Reservations. It turned out that it was a change within Veritas Enterprise Administrator, just had to click a button that said “Remove SCSI reservations”. After that, the messages stopped showing up in the vmkernel.

Published by Karim Elatov

16 April 2012