Massive amount of reservation conflict for clustered VMDK datastore (Mode sense (0x1a))

Products

• VMware vSphere ESXi

Issue/Introduction

Symptoms:

• Massive amount of reservation conflict for clustered VMDK datastore
• Host not responding
• Reservation conflicts for Mode sense (0x1a) command

Examples:

475Z cpu4:2098225)NMP: nmp_ResetDeviceLogThrottling:3776: Error status H:0x0 D:0x18 P:0x0 Sense Data: 0x0 0x0 0x0 from dev "naa.#########################6142" occurred 2000 times(of 2003 commands)

And

509Z cpu72:2097292)ScsiDeviceIO: 4267: Cmd(0x45d9dfadb800) 0x1a, CmdSN 0x73b6 from world 0 to dev "naa.#########################6142" failed H:0x3 D:0x0 P:0x0

 

Environment

VMware vSphere ESXi 7

VMware vSphere ESXi 8

Cause

 As per the SCSI t10 spec SPC4r37 LUN which claims SPC-2 | SPC-3 support may return reservation conflict for Mode sense (0x1a) command even though the Mode sense command is not a write to media type request (Page 132 Table-66), this is allowed only in case of SPC4 Write Exclusive type reservation. By setting the perennially reserved flag as TRUE we can mitigate this issue.

Resolution

Not availabe

Workaround:
You can mitigate that by marking the LUN as "perennially reserved"

Additional Information

It is safe to ignore messages like "489Z cpu42:2100820)WARNING: Partition: 1437: Device "naa.#########################6142" with a VMFS partition is marked perennially reserved. This is not supported and may lead to data loss." for clustered VMDK datastore

Impact/Risks:
hostd service became unresponsive