misc: add NVMe device support to get_scratch_devices() - #2118
Conversation
| devs.remove(dev) | ||
| if not dev: | ||
| continue | ||
| if 'vda' in dev or 'nvme0n1' in dev: |
There was a problem hiding this comment.
For the configurations where are mixed devices installed, for example, where vda is boot and nvme are data, this code will ignore one device, which should not be ignored, right?
There was a problem hiding this comment.
Fixed root device filtering to only filter nvme0n1 when vda doesn't exist, preventing false positives in mixed configurations.
There was a problem hiding this comment.
hm, here are another suggestions,
first, if the devices is listed in the scratch_devs file, there should be no exclusion, from my understanding, because if it is manually defined per machine type or hardware layout operator knows why it is there;
second, in order to exclude root in case there is no scratch_devs file we need to check which device is actually mounted as a root, the best way I guess is to look into mount table or even use findmnt --real
For example:
findmnt --real --json
{
"filesystems": [
{
"target": "/",
"source": "/dev/vda1",
"fstype": "xfs",
"options": "rw,relatime,seclabel,attr2,inode64,logbufs=8,logbsize=32k,noquota"
}
]
}
Just looking for "/" will give us device where the root is mounted, right?
$ findmnt --real --json | jq -r '.filesystems[] | select(.target == "/")| .source'
/dev/vda1
When /scratch_devs file is missing, the function only detected traditional SCSI/VirtIO devices (/dev/sd? and /dev/vd?), causing failures on machines using NVMe devices. This change: - Adds NVMe device detection (/dev/nvme*n1) as fallback - Filters out nvme0n1 as root device (similar to vda) - Fixes iteration bug when removing root devices - Filters out empty strings from device list Signed-off-by: deepssin <deepssin@redhat.com>
593fb6a to
f84d2e3
Compare
zmc
left a comment
zmc
left a comment
There was a problem hiding this comment.
This looks good to me - could we see a small test run using it?
|
I tested this earlier on NVMe hardware, but I didn't keep the logs and no longer have access to that setup to rerun it. The change is minimal, but a quick verification on NVMe backed hardware would be helpful. |
3 participants
When /scratch_devs file is missing, the function only detected traditional SCSI/VirtIO devices (/dev/sd? and /dev/vd?), causing failures on machines using NVMe devices.
This change: