====== [TROUBLESHOOT] MDS cache too large ====== ^ Documentation ^| ^Name:| [TROUBLESHOOT] MDS cache too large | ^Description:| how to solve this "issue" | ^Modification date :|31/07/2019| ^Owner:|dodger| ^Notify changes to:|Owner | ^Tags:|ceph, object storage | ^Scalate to:|The_fucking_bofh| ====== The Warning ====== avmlp-osm-001 /home/ceph # ceph health detail HEALTH_WARN 1 MDSs report oversized cache MDS_CACHE_OVERSIZED 1 MDSs report oversized cache mdsavmlp-osfs-004.ciberterminal.net(mds.0): MDS cache is too large (2GB/1GB); 903372 inodes in use by clients, 0 stray files ====== The solution ====== Increase MDS cache, this is a **soft** warning, just reporting high memory cache usage by MDS. \\ The setting must be changed on **every** MDS server, for example: avmlp-osfs-004 /var/log/ceph # ceph daemon mds.avmlp-osfs-004.ciberterminal.net config get mds_cache_memory_limit { "mds_cache_memory_limit": "4294967296" } And from a monitor node, with the server list exported: ceph@avmlp-osm-001 ~/ceph-deploy :( $ for i in ${MDSSERVERS} ; do echo "###### ${i}" ; ssh ${i} "sudo ceph daemon mds.${i} config set mds_cache_memory_limit 4294967296 " ; done ###### avmlp-osfs-002.ciberterminal.net { "success": "mds_cache_memory_limit = '4294967296' " } ###### avmlp-osfs-001.ciberterminal.net { "success": "mds_cache_memory_limit = '4294967296' " } ###### avmlp-osfs-004.ciberterminal.net { "success": "mds_cache_memory_limit = '4294967296' " } ###### avmlp-osfs-003.ciberterminal.net { "success": "mds_cache_memory_limit = '4294967296' " } And add the corresponding option to ''ceph.conf'': [mds] # https://docs.ceph.com/ceph-prs/19690/cephfs/cache-size-limits/ mds cache memory limit = 4294967296 ====== Official documentation ====== * [[https://ceph.com/community/new-luminous-cephfs-metadata-server-memory-limits/]]