Making this optional is fine, but when lastSyncTime is not present in status, what is the assumption consumers would have in terms of the time it was last synced? IOW, if a driver does not support this then this cannot be used. A driver that supports this, may not update it at which point there needs to be an assumption made by consumers that this is not yet synced. How would consumers know the capability and act accordingly is the key question.

Making this optional is fine, but when lastSyncTime is not present in status, what is the assumption consumers would have in terms of the time it was last synced? IOW, if a driver does not support this then this cannot be used. A driver that supports this, may not update it at which point there needs to be an assumption made by consumers that this is not yet synced. How would consumers know the capability and act accordingly is the key question.

When the last sync time is not present in the current request, the consumer should reuse the last known time for previous RPC calls.

Making this optional is fine, but when lastSyncTime is not present in status, what is the assumption consumers would have in terms of the time it was last synced? IOW, if a driver does not support this then this cannot be used. A driver that supports this, may not update it at which point there needs to be an assumption made by consumers that this is not yet synced. How would consumers know the capability and act accordingly is the key question.

When the last sync time is not present in the current request, the consumer should reuse the last known time for previous RPC calls.

Which is fine, question is when a driver does not support this there is no prior value as well, and how does the consumer know and communicate to the user or other consumers that this is not supported?

Which is fine, question is when a driver does not support this there is no prior value as well, and how does the consumer know and communicate to the user or other consumers that this is not supported?

In that case, no LastSyncTime will be added and it's based on the capability.

• The consumer first checks whether the driver supports the GetReplicationInfo API call, if yes use lastSyncTime, if the lastSyncTime is not present in the RPC call use last known LastSyncTime
• If the capability itself is not supported by the Driver, dont make RPC call nor advertise lastSyncTime to others.

If a driver does not ever return the optional LastSyncTime in the GetReplicationInfo response. What are consumers expected to use as LastSyncTime?

If a driver does not ever return the optional LastSyncTime in the GetReplicationInfo response. What are consumers expected to use as LastSyncTime?

If the driver is not returning, user will use last known LastSyncTime

If a driver does not ever return the optional LastSyncTime in the GetReplicationInfo response. What are consumers expected to use as LastSyncTime?

If the driver is not returning, user will use last known LastSyncTime

I still must be missing something... GetVolumeReplicationInfo is the only place where LastSyncTime could be returned. There is no other way to find out when the syncing was last finished, I think? Consumers may not rely on optional attributes for their functioning, so the LastSyncTime in GetVolumeReplicationInfo can only be used informative?

If a driver does not ever return the optional LastSyncTime in the GetReplicationInfo response. What are consumers expected to use as LastSyncTime?

If the driver is not returning, user will use last known LastSyncTime

I still must be missing something... GetVolumeReplicationInfo is the only place where LastSyncTime could be returned. There is no other way to find out when the syncing was last finished, I think? Consumers may not rely on optional attributes for their functioning, so the LastSyncTime in GetVolumeReplicationInfo can only be used informative?

Yes, its for information only the consumer cannot decide based on it.

There is no other way to find out when the syncing was last finished

The final sync complete is based on the readyToUse flag in the resync RPC call.

I don't see it being useful for a driver to implement GetVolumeReplicationInfo, but unable to return the last_sync_time. At the moment it is the only attribute of the response. Returning the "last known LastSyncTime" seems wrong, it may not always be the correct last time if some error occurred, better to return an error in that case and retry later.

I'd like to see a description of errors that can be returned on a GetVolumeReplicationInfo call, with notes that some errors can be retried later, whereas others are final failures and should not be retried.

Keeping the last_sync_time REQUIRED makes it clear to driver implementers and consumers of the API what the expectations are.

@nixpanic PTAL