Bug #52512
Updated by Konstantin Shalygin over 2 years ago
Our application use object size mostly 512 bytes 512k or lower, with 4K overhead we can't use standard stored or stored_raw metrics for pool usage <pre> # ceph df RAW STORAGE: CLASS SIZE AVAIL USED RAW USED %RAW USED hdd_pg 4.3 PiB 1.0 PiB 3.2 PiB 3.2 PiB 75.91 nvme_pg 161 TiB 44 TiB 96 TiB 117 TiB 72.65 TOTAL 4.4 PiB 1.1 PiB 3.3 PiB 3.4 PiB 75.79 POOLS: POOL ID PGS STORED OBJECTS USED %USED MAX AVAIL meta 17 2307 3.6 TiB 8.62G 96 TiB 85.58 5.4 TiB data 18 16384 1.1 PiB 3.16G 3.2 PiB 81.76 245 TiB </pre> As we can see for store 3.6TiB we use 96TiB We need expose avail_raw metrics and revert bytes_used metric back (((8620017891 * 512) * 3) = 13240347480576 (12.04TiB)) - (((8620017891 * 4096) * 3) = 105922779844608 (96.34TiB)) = 92682432364032 (84.29TiB) that we can't see