ioztat is a storage load analysis tool for OpenZFS. It provides iostat-like statistics at an individual dataset/zvol level, complementing the pool/vdev level statistics offered by zpool iostat.
The statistics offered are per-second or per-interval read/write operations and throughput, plus optional average operation size and ZFS file unlink queue depths.
Viewing these statistics at the individual dataset level allows system administrators to identify storage "hot spots" in larger multi-tenant systems — particularly those with many VMs or containers operating essentially independent workloads.
This sample output shows activity which has taken place in the most recent second, on the ssd zpool of a ZFS virtualization host, with extended statistics showing the average I/O size:
root@redacted-prod0:~# ioztat -yx ssd
operations throughput opsize
dataset read write read write read write
---------------- ----- ----- ----- ----- ----- -----
ssd 0 0 0 0 0 0
images 0 0 0 0 0 0
DC1 4 18 49.0K 98.0K 12.5K 5.56K
DC2 0 22 0 137K 0 6.36K
QB 0 2 0 9.80K 0 5.00K
SAP-TC 0 3 0 19.6K 0 6.67K
SAP4-WIN2019 0 3 0 49.0K 0 16.7K
nagios 0 0 0 0 0 0
qemu 0 0 0 0 0 0
autostart 0 0 0 0 0 0
iso 0 0 0 0 0 0
unsnapped 0 0 0 0 0 0
rp9 0 0 0 0 0 0
ioztat behaves similarly to standard Unix iostat tools and the ZFS zpool iostat command:
usage: ioztat [-c COUNT] [-D] [-e] [-H] [-h] [-I] [-i INTERVAL] [-N] [-n] [-o]
[-P | -p] [-S]
[-s {name,operations,reads,writes,throughput,nread,nwritten}]
[-T {u,d}] [-V] [-x] [-y] [-z]
[dataset [dataset ...]] [interval [count]]
iostat for ZFS datasets
positional arguments:
dataset ZFS dataset
interval [count] seconds between reports and number of reports
optional arguments:
-c COUNT number of reports generated
-D display size in decimal powers of 1000 instead of 1024
-e display exact values without truncation or scaling
-H scripted mode, omit headers and tab-separate fields
-h, --help display this help message and exit
-I display totals since the last report rather than
averaged per-second
-i INTERVAL interval between reports in seconds
-N display headers at most once
-n omit child datasets when filtering
-o overwrite old reports in terminal
-P display dataset names on a single line
-p display dataset names as an abbreviated tree
-S include statistics for child datasets in parents
-s {name,operations,reads,writes,throughput,nread,nwritten}
sort by the specified field
-T {u,d} prefix reports with a Unix timestamp or formatted date
-V, --version display version number and exit
-x display extended statistics: once for average I/O
size, twice for unlink queue
-y omit the initial "summary" report
-z omit datasets with zero activity
Without arguments, ioztat prints a summary of activity for each mounted dataset since the most recent system boot and exits. This initial summary can be skipped with the -y flag.
With an optional interval, ioztat will repeat reports on that schedule until interrupted, or up to a specified count. If only a count is specified, the interval defaults to one second. Interval and count can be specified with -i and -c or as positional arguments at the very end of the argument list.
The -o flag will overwrite prior output and limit the display to the terminal height. This can be combined with an interval and sorting options for an iotop-like experience.
ioztat -z 5
Display statistics for every dataset with activity, giving per-second averages over 5 second intervals.
ioztat -ozs operations 5
As above, but order output by read/write operations and overwrite previous reports.
ioztat -I
Display sum totals of dataset activity since boot.
ioztat -SIn rpool/USERDATA
As above, but only display rpool/USERDATA, and combine statistics for any child datasets with it.
- Python 3.7 or later
- Linux with OpenZFS 0.8 or later
- FreeBSD 12.2 or later (available in ports as
sysutils/py-ioztat)