Greetings, For anyone running or planning to run an MDS on a big-endian architecture, please be aware that when directories are fragmented (an automatic process when they become large) that the MDS will send byte-swapped fragment identifiers to clients. This confuses the client and can result in it entering an infinite loop trying to read the contents of a directory. The issue is tracked here: https://tracker.ceph.com/issues/73792 The fix includes code to automatically correct any persisted byte-swapped fragment identifiers. We expect it to be backported to Tentacle and Squid as soon as the fix is in the main branch. Regards, -- Patrick Donnelly, Ph.D. He / Him / His Red Hat Partner Engineer IBM, Inc. GPG: 19F28A586F808C2402351B93C3301A3E258DD79D