CephFS client evictions

When a Ceph File System (CephFS) client is unresponsive or misbehaving, it might be necessary to forcibly stop, or evict it from accessing the CephFS.

Evicting a CephFS client prevents it from communicating further with Metadata Server (MDS) daemons and Ceph OSD daemons. If a CephFS client is buffering I/O to the CephFS at the time of eviction, then any un-flushed data will be lost. The CephFS client eviction process applies to all client types: FUSE mounts, kernel mounts, NFS gateways, and any process that uses libcephfs API library.

You can evict CephFS clients automatically, if they fail to communicate promptly with the MDS daemon, or manually. For more information, see Manually evicting a CephFS client.

Automatic client eviction

These scenarios cause an automatic CephFS client eviction:

If a CephFS client has not communicated with the active MDS daemon for over the default of 300 seconds, or as set by the session_autoclose option.
If the mds_cap_revoke_eviction_timeout option is set, and a CephFS client has not responded to the cap revoke messages for over the set number of seconds. The mds_cap_revoke_eviction_timeout option is disabled by default.
During MDS startup or failover, the MDS daemon goes through a reconnect phase waiting for all the CephFS clients to connect to the new MDS daemon. If any CephFS clients fail to reconnect within the default time window of 45 seconds, or as set by the mds_reconnect_timeout option.