You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
With regards to the size of a volume, the CSI spec supports specifying two parameters: required_bytes (which is really the minimum) and limit_bytes (the maximum). The driver is free to choose a value within the given range; specifically, if limit_bytes is not specified and required_bytes is below what is supported by the storage provider, then the driver should round up to the minimum supported size (1 GiB in DigitalOcean). See also the CSI spec on CreateVolume, especially the CapacityRange message.
This change is best shipped as a major release to make it clear to users that the behavior is going to change somewhat (though practically, most users would have presumably just increased the value anyway after they received an error).
We should also sanity-check if this change could have any negative impact when creating volumes from existing sources (e.g., snapshots).
With regards to the size of a volume, the CSI spec supports specifying two parameters:
required_bytes
(which is really the minimum) andlimit_bytes
(the maximum). The driver is free to choose a value within the given range; specifically, iflimit_bytes
is not specified andrequired_bytes
is below what is supported by the storage provider, then the driver should round up to the minimum supported size (1 GiB in DigitalOcean). See also the CSI spec onCreateVolume
, especially theCapacityRange
message.Today, our CSI driver does not adhere to the spec and instead errors out if
required_bytes
is set below the minimum whilelimit_bytes
is unset (which is always the case in Kubernetes as of today). The behavior should be fixed by rounding up accordingly.This change is best shipped as a major release to make it clear to users that the behavior is going to change somewhat (though practically, most users would have presumably just increased the value anyway after they received an error).
We should also sanity-check if this change could have any negative impact when creating volumes from existing sources (e.g., snapshots).
The issue was surfaced along kubernetes/kubernetes#110277.
The text was updated successfully, but these errors were encountered: