CON-9040 driver can accidentally format existing volume #478
Conversation
|
Let's also make sure to document the new flag in the README. |
…efault value for the
|
Forgot to mention: we can already extend the CHANGELOG file for this feature. |
CHANGELOG.md
Outdated
| @@ -1,5 +1,9 @@ | |||
| ## unreleased | |||
|
|
|||
| ## v4.5.1 - 2023.03.03 | |||
There was a problem hiding this comment.
Please omit the header -- that's going to be added at the time we do the release for sure.
There was a problem hiding this comment.
I've removed the version header.
driver/mounter.go
Outdated
| return false, fmt.Errorf("error comparing the state file content, expected: %s, got: %s", runningState, string(deviceStateFileContent)) | ||
| } | ||
|
|
||
| return true, nil |
There was a problem hiding this comment.
At this point, we only return true when err == nil, which means we could drop the boolean return value and rely on the error return value exclusively.
There was a problem hiding this comment.
I've removed the boolean return and only kept the error. I've refactored the interface method with the new structure, refactored the implementation and the tests.
driver/mounter_test.go
Outdated
| name string | ||
| av AttachmentValidator | ||
| errorMsg string | ||
| wantErr bool |
There was a problem hiding this comment.
wantErr is not really needed since it derives from errorMsg != "".
There was a problem hiding this comment.
I've removed the wantErr and want properties and now I'm only relying on errorMsg.
driver/mounter_test.go
Outdated
| if tt.wantErr { | ||
| assert.ErrorContains(t, err, tt.errorMsg) | ||
| } else { | ||
| assert.Equal(t, tt.want, got) |
There was a problem hiding this comment.
nit / personal note: Go stdlib testing prefers outputting got first, then want. I'd match the order here given how thin of a layer gotest.tools is.
There was a problem hiding this comment.
I refactored the test so now it uses the got first and want after.
llDrLove
deleted the
CON-9040-driver-can-accidentally-format-existing-volume
branch
There is a race condition where DOBS volumes look to our indication method as if they'd fully be attached, but aren't completely.
The result can be a misinterpretation by CSI which causes a force-format of the volume.
This PR uses a new method to check the attachment of the volume by looking at the
runningstate in the/sys/class/block/<device name like sda>/device/statefile. This can be enabled using the new flag--validate-attachment=trueflag.