Fix two issues encountered when testing LHv2 provisioning #142
Conversation
Previously, every call to LonghornV2Provisioner.Update() would trigger an udpate of the Longhorn node object and a requeue even if nothing had changed. This resulted in the BD getting suck with the AddedToNode condition set to False and an error: admission webhook "validator.longhorn.io" denied the request: spec and status of disks on node harvester-node-0 are being syncing and please retry later. Now we only try to sync if the DiskDriver has actually changed, and everything behaves correctly. Signed-off-by: Tim Serong <tserong@suse.com>
When Longhorn V2 volumes are created and attached to VMs, they appear on Harvester hosts as /dev/dm-* and /dev/nvme* devices. This is problematic, because NDM thinks those things are actual disks and creates BD CRs from them. This change is a bit of a hack until I finish the work for harvester/harvester#5059. Signed-off-by: Tim Serong <tserong@suse.com>
| ) | ||
|
|
||
| var ( | ||
| defaultExcludedVendors = []string{vendorFilterDefaultLonghorn} | ||
| defaultExcludedVendors = []string{vendorFilterDefaultLonghorn, modelFilterDefaultSPDK} |
There was a problem hiding this comment.
Would you like to change the name of this variable?
Now, we mixed the vendor/module into this slice.
I do not have a strong opinion about changing the name. If you want to keep it, I am also okay with that.
Thanks!
There was a problem hiding this comment.
I'd like to keep it for now - I know it's a bit silly/incorrect, I just didn't rename it given that I want to rework all this filter stuff anyway separately for harvester/harvester#5059, and keeping the name avoids also tweaking the SetExcludeFilters() function in this commit.
|
@Mergifyio backport v0.7.x |
✅ Backports have been created
|
|
Failure is the side-effect that comes from https://github.com/harvester/node-disk-manager/blob/master/pkg/controller/blockdevice/scanner.go#L287 The autoProvision should not update both We can ignore this error at this PR. I will file another PR to fix it. |
Problem:
/dev/dm-*and/dev/nvme*devices on the host. NDM sees these and thinks they're real disks, and then creates new BD CRs from those devices.Solution:
reflect.DeepEqual()to avoid unnecessary updates and resultant weird state./dev/dm-devices inApplyExcludeFiltersForDisk().Note: I intend to handle this more cleanly later as part of harvester/harvester#5059
Related Issue:
harvester/harvester#5274
Test plan:
N/A