NAME

ztest — ZFS stress test utility

SYNOPSIS

DESCRIPTION

The ztest utility stress tests the DMU, ZAP, SPA components of ZFS from user space.

ztest provides a simple routine to test the functionality of a component task. These simple routines can then be used to stress test through parallel execution. Tests are run as child processes of the main process.

The checksum and compression functions are changed each time a dataset is opened to introduce varying combinations of checksum and compression from block to block among objects.

While tests are running, faults are injected into the pool to verify self-healing ability.

To verify that the on-disk consistency is never lost after a crash, child processes are killed at random with a SIGKILL signal, after which the parent process invokes a new child to run the test again on the same storage pool.

Many of the tests record the transaction group number as part of their data. When reading old data, tests verify that the transaction group number is less than the current, open state, transaction group to ensure the consistency of tests and detect unaccounted changes. It is advised that any new tests added to ztest should also perform the same check with transaction group numbers, if applicable.

The main ztest process uses an mmap(2) temporary file to pass information to child processes which allows shared memory to survive exec(2) syscall. A copy of the ztest_shared_hdr_t struct containing information on the size and number of shared structures in the file is always stored at offset 0 of the file.

For backwards compatibility testing ztest can invoke an alternative version of ztest after a SIGKILL signal using the -B flag.

The following options are available:

-a shift

alignment shift (default: 9) use 0 for random

-B path

alt ztest (default: <none>) alternate ztest path

-d datasets

datasets (default: 7)

-E

use existing pool instead of creating new one

-F loops

freezeloops (default: 50) max loops in spa_freeze()

-f path

dir (default: /tmp) file directory for vdev files

-g threshold

gang block threshold (default: 32K)

-h

print help

-i count

init count (default: 1) initialize pool count times

-k percent

kill percentage (default: 70%)

-m copies

mirror copies (default: 2)

-P passtime

passtime (default: 60 sec) time per pass

-p name

pool name (default: ztest)

-R parity

raidz parity (default: 1)

-r disks

raidz disks (default: 4)

-s size

size of each vdev (default: 256M)

-T time

time (default: 300 sec) total run time

-t threads

threads (default: 23)

-V

verbose (use multiple times to increase verbosity)

-v vdevs

number of vdevs (default: 5)

AUTHORS

This man page was written by Sevan Janiyan <sevan@NetBSD.org>.

CAVEATS

To allow for backward compatibility testing using older versions of ztest the information stored in the mmap(2) temporary file passed from parent process to child must remain compatible with older versions of ztest.