Removing a nodemap fileset wipes the whole subid range reserved for that fileset instead of deleting the records the fileset actually occupies. Every fileset removal therefore declares LUSTRE_NODEMAP_FILESET_SUBID_RANGE (256) index deletes in a single transaction, whatever the length of the fileset path, and even when the nodemap has no fileset at all.

On ldiskfs each osd_index_declare_iam_delete() charges osd_dto_credits_noquota[DTO_INDEX_DELETE] (1) + 3 = 4 credits, so one removal declares 1024 credits. When that exceeds the maximum transaction size of the journal, osd_trans_start() clamps the transaction down to jbd2_journal_get_max_txn_bufs(journal) / 2 and osd_trans_exec_op() reports the mismatch:

  lustre-MDT0000: opcode 9: before 516 < left 1024, rollback = 8
    create: 0/0/0, destroy: 0/0/0
    attr_set: 0/0/0, xattr_set: 1/1/0
    write: 0/0/0, punch: 0/0/0, quota 1/3/0
    insert: 0/0/0, delete: 256/1024/0
    ref_add: 0/0/0, ref_del: 0/0/0