read-write Dataset gives Socket not connected errors after a cp -r

[starpit]

i am seeing issues when i try to mount a IBM Cloud COS bucket read-write. i see this error from pods that try to mount read-write Datasets:

Socket not connected

is it possible that csi-s3 only supports ReadWriteOnce? I see that datashim sets the PVC to ReadWriteMany, which seems not to be supported by csi-s3?

[starpit]

Update

now i'm less certain as to what is going on. (also updated issue title)

it turns out that the mount is fine... until my application does a cp -r src dst where both src and dst are on the mount in question. after that, subsequent reads result in Socket not connected.

[starpit]

Update 2

yes, if i update the code to avoid the s3-to-s3 copy, it works fine. how do we get to the root cause of this bug? i spent hours scouring the csi, kubelet, etc. logs, and there was nothing. not a single sign of why the mount disconnects after an s3-to-s3 copy.

[srikumar003]

@starpit we are looking into this issue but RWX has not (yet) caused an issue for csi-s3. We are duplicating the s3-s3 copy through the mount-points and be back with at least an understanding of what's happening

[srikumar003]

@starpit I was unable to replicate the issue either on kind or in an OpenShift cluster with small files (10s of KB) or comparatively larger files (approx 6 MBs). I can try again when I have more information about the context in which you got these errors