Viewing: mdt.recovery_reconnect_histogram.4
.TH RECOVERY_RECONNECT_HISTOGRAM 4 2026-05-30 Lustre "Lustre Kernel Interfaces"
.SH NAME
recovery_reconnect_histogram \- reconnect-delay histogram per recovery
.SH SYNOPSIS
.SY
.BI "lctl get_param mdt." TARGET ".recovery_reconnect_histogram"
.SY
.BI "lctl get_param obdfilter." TARGET ".recovery_reconnect_histogram"
.SY
.BI "lctl set_param mdt." TARGET ".recovery_reconnect_histogram=clear"
.YS
.SS PROPERTIES
.TP
.B Access Permissions
.BR 644 " | " -rw-r--r--
.TP
.B Scope
.br
Per MDT or OST target device
.SH DESCRIPTION
The
.B recovery_reconnect_histogram
parameter reports a log2-bucketed histogram of how many seconds each client
took to reconnect to this target during the current recovery window. The
histogram is reset when recovery begins, so it always reflects the most
recent recovery. Writing any value to the parameter also clears it.
.PP
Output is YAML. Leading scalars report the recovery window
.RB ( recovery_start ", " recovery_finish ", " recovery_time " in seconds)"
and the total sample count, followed by a
.B client_reconnect_histogram
list. Each list entry reports a bucket upper bound
.RB ( delay_sec ),
the number of clients whose reconnect delay fell into that bucket
.RB ( clients ),
that bucket's percentage of the total
.RB ( pct ),
and the cumulative percentage
.RB ( cum_pct ).
Buckets with zero clients before the first non-zero bucket are elided.
.PP
Per-client
.BR mdt.reconnect_delay (4)
parameters are also exposed individually under each client mountpoint's
.BI mdt. TARGET .exports. NID .reconnect_delay
and
.BI obdfilter. TARGET .exports. NID .reconnect_delay
parameter. This is useful for isolating a specific client's recovery delay,
which is otherwise shown in aggregate form in the full histogram,
or may not be in the
.BR mdt.recovery_reconnect_top (4)
list.
.SH MODULES
This parameter is in the following modules:
.EX
.B mdt.*.recovery_reconnect_histogram
.B obdfilter.*.recovery_reconnect_histogram
.EE
.SH EXAMPLES
Show the histogram for an MDT after a recovery:
.EX
.RB mds# " lctl get_param mdt.lfs-MDT0000.recovery_reconnect_histogram"
recovery_start: 1780002682
recovery_finish: 1780002729
recovery_time: 47
reconnect_delay_seconds_samples: 58
client_reconnect_histogram:
- { delay_sec: 1, clients: 45, pct: 77, cum_pct: 77 }
- { delay_sec: 2, clients: 6, pct: 10, cum_pct: 87 }
- { delay_sec: 4, clients: 4, pct: 7, cum_pct: 94 }
- { delay_sec: 8, clients: 2, pct: 3, cum_pct: 97 }
- { delay_sec: 16, clients: 0, pct: 0, cum_pct: 97 }
- { delay_sec: 32, clients: 0, pct: 0, cum_pct: 97 }
- { delay_sec: 64, clients: 1, pct: 3, cum_pct: 100 }
.EE
Clear the histogram:
.EX
.RB mds# " lctl set_param mdt.lfs-MDT0000.recovery_reconnect_histogram=clear"
.EE
.SH AVAILABILITY
.B recovery_reconnect_histogram
is part of the
.BR lustre (7)
filesystem package since release 2.18.0.
.\" commit v2_17_53
.SH SEE ALSO
.BR mdt.reconnect_delay (4),
.BR mdt.recovery_reconnect_top (4),
.BR lctl-get_param (8),
.BR lctl-set_param (8)
