Main idea of zdb was initially to be efficient on HDD, current implementation was doing always-append but per namespace, which means random write to different namespace at the same time would eventually produce fragmentation.
New implementation ensure data are (lazily) pre-allocated using fallocate by reserving blocks when creating data file.
Main idea of zdb was initially to be efficient on HDD, current implementation was doing always-append but per namespace, which means random write to different namespace at the same time would eventually produce fragmentation.
New implementation ensure data are (lazily) pre-allocated using
fallocateby reserving blocks when creating data file.