Open mnaamani opened 9 months ago
If an operator runs out of balance, they could build up a stack of uploaded objects that don't get accepted. If a node like that disappears for some reason, currently all those objects would get lost since other operators won't sync until object is moved to main uploads/
.
This is possibly blocked by the fact that storage nodes will currently not synch objects that haven't been accepted on-chain
Currently on request for an object that is still being processed in the pending folder, we return a 404. We were concerned with what would happen if file was being streamed to the response, while the
AcceptPendingObjects
service tries to move/copy the file from pending folder to the uploads folder.If we find that it is safe and we are okay with the resulting behaviour, and the potential latency between uploading a new object and being able to fetch it from the same node is not acceptable, then implement this change.