When uploading data the true size of the upload is not always known up front. Thus in some cases the entire image must arrive before the image size is known. This causes a problem when trying to implement quota and when trying to limit the size of any given image
In order to implement a size limit on upload of a given image the iterator (ie: image_size_cap). To deal with this the iterator LimitingReader was created. This is a wrapper around the incoming data that throws a ImageSizeLimitExceeded exception when a liit is reach. This effectively protects the system from receiving more than the alloted ammount.
However, when that error occurs the already received data is never cleaned up. The fix for this must be in each data store. The higher level abstractions do not know the location of the data until it is all successfully written.
After I checking from the details, currently gridfs, rbd, sheepdog store drivers have this leaking issue, and filesystem, swift and s3 store drivers is OK.