Sysbench file I/O test on nvme crash

[kartsan]

My environment is as follows:

- Supermicro H11SSL-NC with EPYC 7232P CPU

- ESXi-7.0b-16324942-standard

- Western Digital SN750 1T disk

- Debian 10 guest

I'm testing WD disk with this configuration but I experience crash every time when testing file I/O with sysbench. I'm using the following command:

sysbench fileio --file-test-mode=seqwr --file-total-size=10G --file-block-size=16K --threads=8 --time=60 run

Please take a look at the following kernel log:

2020-09-07T09:10:19.407Z cpu8:1049379)WARNING: NVMEIO:2223 Controller 256 receiv - Pastebin.com

You can see that there is a critical warning 0x2 in the beginning and after that everything goes down. APD starts and All Paths Down is finally reached. Disk never recovers and I have to power off the guest. It is in invalid state after that and I need to reboot the host to recover.

It appears that increasing --file-block-size will make the system crash earlier. With 4K block size it seems to work.

Any ideas what is the issue and/or how to workaround this? What is that critical warning 0x2?

[scott28tt]

What type of disk controller is the WD connected to?

[kartsan]

Thanks for your reply. It is connected to the PCIe M.2 connector on the motherboard. According to the manual, that is directly connected to the CPU via G1 bus.

Ilkka

[scott28tt]

And have you validated your System and IO Devices hardware against this?

VMware Compatibility Guide - System Search​