White Papers

Known issues with NVMe surprise removal
9 NVMe Surprise Removal on Dell EMC PowerEdge servers running Linux operating systems | 451
3.1.4 /proc/mdstat and mdadm -D commands display incorrect statuses when two
NVMe devices are surprise removed from a RAID 5 MD array
Description: When two of three NVMe devices are surprise removed from a RAID 5 MD array, the command
cat/proc/mdstat displays the array status incorrectly as active. Similarly, when the status of the MD
RAID is queried using the mdadm -D /dev/mdN command, the number of active and working devices
displayed is two. Only the status of the array reported is incorrect however, when I/O operations are
performed, I/O errors are observed as expected.
Cause: When the number of devices that are surprise removed exceeds the number of devices that are
required for the array to function, the MD status is not updated.
3.2 Ubuntu LTS 20.04.01
3.2.1 The name of the NVMe device may change when it is hot inserted after a
surprise removal
Description: If an NVMe device is hot inserted after it was previously surprise removed when I/O operations
are accessing the device, the name of the NVMe device may change or will not retain the same name that is
assigned prior to surprise removal. Dmesg displays the following messages:
kernel: nvme nvme3: failed to mark controller CONNECTING
kernel: nvme nvme3: Removing after probe failure status: -16
The functionality of the NVMe device is not affected.
3.2.2 NVMe devices are enumerated in namespace 2 when hot-inserted into the
server after being surprise removed
Description: When an NVMe device from a RAID 1 MD array is hot inserted after being surprise removed,
the device is enumerated in namespace 2 although only one namespace is enabled. The device is named as
nvme2n2 instead of nvme2n1. This issue is observed on Dell Express Flash PM1725a device. The
functionality of the NVMe device is not affected.
Workaround: Pass the multipath=N module parameter to the nvme_core driver.