Jordan Rife
Jordan Rife
@dobsonj @divyenpatel I'm wondering if there has been any progress on this front? I have a scenario where we may not be able to disable `multipathd` to get the vSphere...
@dobsonj Thanks for looking into this. > > the correct solution might be instead be to fail in NodeStageVolume as we have in aws-ebs and gcp-pd today > > We...
Bump @xing-yang @dobsonj
> This difference between platforms is puzzling to me, but we don't necessarily need to solve that here. As far as I can tell, the symlink evaluation in GetDevMounts is...
@dobsonj OK, that works for me then. I have no strong opinion on the second commit here. I'll let @xing-yang and @divyenpatel chime in on the rest.
Here's a trace of one of the `CreateVolume()` requests in the `vsphere-csi-controller` (`v3.0.2`) logs from a similar scenario: ``` 2023-08-04T12:21:33.335367178Z {"level":"info","time":"2023-08-04T12:21:33.335207218Z","caller":"vanilla/controller.go:1806","msg":"CreateVolume: called with args {Name:pvc-777d04c4-454e-409e-9bd0-f5c0f8f3cbff CapacityRange:required_bytes:5368709120 VolumeCapabilities:[mount: access_mode: ] Parameters:map[datastoreURL:ds:///vmfs/volumes/vsan:52111bdcd5e647f9-6419d415990a43f9/]...
Here is a sample of what logs look like on the backend (couldn't find the exact log that corresponds to the PV above, but they all follow a similar pattern)....
This scenario can be reproduced by lowering CSI sidecar `--timeout` argument to a low value (e.g. `100ms`)
@jingxu97 This might reduce the chances of duplicate creations, but ultimately, the only solution that guarantees idempotency 100% is CNS API support either through: 1) Uniqueness guarantees for a particular...