CephFS provides top(1) like utility to display various Ceph Filesystem metrics in realtime. cephfs-top is a curses based python script which makes use of stats plugin in Ceph Manager to fetch (and display) metrics.
Ceph Filesystem clients periodically forward various metrics to Ceph Metadata Servers (MDS) which in turn get forwarded to Ceph Manager by MDS rank zero. Each active MDS forward its respective set of metrics to MDS rank zero. Metrics are aggregated and forwarded to Ceph Manager.
Metrics are divided into two categories - global and per-mds. Global metrics represent set of metrics for the filesystem as a whole (e.g., client read latency) whereas per-mds metrics are for a particular MDS rank (e.g., number of subtrees handled by an MDS).
Note
Currently, only global metrics are tracked.
stats plugin is disabled by default and should be enabled via:
$ ceph mgr module enable stats
Once enabled, Ceph Filesystem metrics can be fetched via:
$ ceph fs perf stats
The output format is JSON and contains fields as follows:
- version: Version of stats output
- global_counters: List of global performance metrics
- counters: List of per-mds performance metrics
- client_metadata: Ceph Filesystem client metadata
- global_metrics: Global performance counters
- metrics: Per-MDS performance counters (currently, empty) and delayed ranks
Note
delayed_ranks is the set of active MDS ranks that are reporting stale metrics. This can happen in cases such as (temporary) network issue between MDS rank zero and other active MDSs.
Metrics can be fetched for a particular client and/or for a set of active MDSs. To fetch metrics for a particular client (e.g., for client-id: 1234):
$ ceph fs perf stats --client_id=1234
To fetch metrics only for a subset of active MDSs (e.g., MDS rank 1 and 2):
$ ceph fs perf stats --mds_rank=1,2
cephfs-top utility relies on stats plugin to fetch performance metrics and display in top(1) like format. cephfs-top is available as part of cephfs-top package.
By default, cephfs-top uses client.fstop user to connect to a Ceph cluster:
$ ceph auth get-or-create client.fstop mon 'allow r' mds 'allow r' osd 'allow r' mgr 'allow r' $ cephfs-top
- chit : Cap hit
- Percentage of file capability hits over total number of caps
- dlease : Dentry lease
- Percentage of dentry leases handed out over the total dentry lease requests
- ofiles : Opened files
- Number of opened files
- oicaps : Pinned caps
- Number of pinned caps
- oinodes : Opened inodes
- Number of opened inodes
- rtio : Total size of read IOs
- Number of bytes read in input/output operations generated by all process
- wtio : Total size of write IOs
- Number of bytes written in input/output operations generated by all processes
- raio : Average size of read IOs
- Mean of number of bytes read in input/output operations generated by all process over total IO done
- waio : Average size of write IOs
- Mean of number of bytes written in input/output operations generated by all process over total IO done
- rsp : Read speed
- Speed of read IOs with respect to the duration since the last refresh of clients
- wsp : Write speed
- Speed of write IOs with respect to the duration since the last refresh of clients
- rlatavg : Average read latency
- Mean value of the read latencies
- rlatsd : Standard deviation (variance) for read latency
- Dispersion of the metric for the read latency relative to its mean
- wlatavg : Average write latency
- Mean value of the write latencies
- wlatsd : Standard deviation (variance) for write latency
- Dispersion of the metric for the write latency relative to its mean
- mlatavg : Average metadata latency
- Mean value of the metadata latencies
- mlatsd : Standard deviation (variance) for metadata latency
- Dispersion of the metric for the metadata latency relative to its mean
To use a non-default user (other than client.fstop) use:
$ cephfs-top --id <name>
By default, cephfs-top connects to cluster name ceph. To use a non-default cluster name:
$ cephfs-top --cluster <cluster>
cephfs-top refreshes stats every second by default. To choose a different refresh interval use:
$ cephfs-top -d <seconds>
Refresh interval should be a positive integer.
To dump the metrics to stdout without creating a curses display use:
$ cephfs-top --dump
To dump the metrics of the given filesystem to stdout without creating a curses display use:
$ cephfs-top --dumpfs <fs_name>
- m : Filesystem selection
- Displays a menu of filesystems for selection.
- s : Sort field selection
- Designates the sort field. 'cap_hit' is the default.
- l : Client limit
- Sets the limit on the number of clients to be displayed.
- r : Reset
- Resets the sort field and limit value to the default.
- q : Quit
- Exit the utility if you are at the home screen (all filesystem info), otherwise escape back to the home screen.
The metrics display can be scrolled using the Arrow Keys, PgUp/PgDn, Home/End and mouse.
Sample screenshot running cephfs-top with 2 filesystems:
Note
Minimum compatible python version for cephfs-top is 3.6.0. cephfs-top is supported on distros RHEL 8, Ubuntu 18.04, CentOS 8 and above.