Hi all, I've recently been playing around with re-enabling delayed allocations on the RT subvolume (for the rextentsize=1 case), and ran into an interesting bug where fsx unexpectedly returned -ENOSPC when testing this code on top of Darrick's rtgroups code. It turns out that this is because xfs_alloc_file_space does not retry when xfs_bmapi_write returns 0 with *nimaps = 0, which can happen if the allocator can't fill the entire space from the beginning of a delalloc extent to the start of the actually requested range in the call to xfs_bmapi_write.
