c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2026-01-14 01:02:04 +03:00
Code Issues Packages Projects Releases Wiki Activity

Add knob to disable slow io notifications

Browse Source 
Introduce a new vdev property `VDEV_PROP_SLOW_IO_REPORTING` that
allows users to disable notifications for slow devices.
This prevents ZED and/or ZFSD from degrading the pool due to slow
I/O.

Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Mariusz Zaborski <oshogbo@FreeBSD.org>
Closes 17477
This commit is contained in:
Mariusz Zaborski 2025-11-11 19:42:17 +01:00 committed by Brian Behlendorf
parent 41878d57ea
commit 1e8c96d7d5
10 changed files with 127 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
include/sys/fs/zfs.h
View File

@ -387,6 +387,7 @@ typedef enum {
VDEV_PROP_SLOW_IOS,
VDEV_PROP_SIT_OUT,
VDEV_PROP_AUTOSIT,
VDEV_PROP_SLOW_IO_EVENTS,
VDEV_NUM_PROPS
} vdev_prop_t;

1
include/sys/vdev_impl.h
View File

@ -470,6 +470,7 @@ struct vdev {
uint64_t vdev_checksum_t;
uint64_t vdev_io_n;
uint64_t vdev_io_t;
boolean_t vdev_slow_io_events;
uint64_t vdev_slow_io_n;
uint64_t vdev_slow_io_t;
};

3
lib/libzfs/libzfs.abi
View File

@ -6026,7 +6026,8 @@
<enumerator name='VDEV_PROP_SLOW_IOS' value='51'/>
<enumerator name='VDEV_PROP_SIT_OUT' value='52'/>
<enumerator name='VDEV_PROP_AUTOSIT' value='53'/>
<enumerator name='VDEV_NUM_PROPS' value='54'/>
<enumerator name='VDEV_PROP_SLOW_IO_EVENTS' value='54'/>
<enumerator name='VDEV_NUM_PROPS' value='55'/>
</enum-decl>
<typedef-decl name='vdev_prop_t' type-id='1573bec8' id='5aa5c90c'/>
<class-decl name='zpool_load_policy' size-in-bits='256' is-struct='yes' visibility='default' id='2f65b36f'>

8
man/man7/vdevprops.7
View File

@ -45,7 +45,7 @@ section, below.
Every vdev has a set of properties that export statistics about the vdev
as well as control various behaviors.
Properties are not inherited from top-level vdevs, with the exception of
checksum_n, checksum_t, io_n, io_t, slow_io_n, and slow_io_t.
checksum_n, checksum_t, io_n, io_t, slow_io_events, slow_io_n, and slow_io_t.
.Pp
The values of numeric properties can be specified using human-readable suffixes
.Po for example,
@ -149,6 +149,12 @@ For
.Sy OpenZFS on FreeBSD
defaults see
.Xr zfsd 8 .
The
.It Sy slow_io_events
property controls whether slow I/O events are generated.
Even when disabled, slow I/Os will be included in the
.Nm zpool Cm status Fl s
output.
.It Sy comment
A text comment up to 8192 characters long
.It Sy bootsize

3
module/zcommon/zpool_prop.c
View File

@ -481,6 +481,9 @@ vdev_prop_init(void)
zprop_register_index(VDEV_PROP_FAILFAST, "failfast", B_TRUE,
PROP_DEFAULT, ZFS_TYPE_VDEV, "on | off", "FAILFAST", boolean_table,
sfeatures);
zprop_register_index(VDEV_PROP_SLOW_IO_EVENTS, "slow_io_events",
B_TRUE, PROP_DEFAULT, ZFS_TYPE_VDEV, "on | off",
"SLOW_IO_EVENTS", boolean_table, sfeatures);
/* hidden properties */
zprop_register_hidden(VDEV_PROP_NAME, "name", PROP_TYPE_STRING,

86
module/zfs/vdev.c
View File

@ -448,6 +448,23 @@ vdev_get_nparity(vdev_t *vd)
return (nparity);
}
static int
vdev_prop_get_objid(vdev_t *vd, uint64_t *objid)
{
if (vd->vdev_root_zap != 0) {
*objid = vd->vdev_root_zap;
} else if (vd->vdev_top_zap != 0) {
*objid = vd->vdev_top_zap;
} else if (vd->vdev_leaf_zap != 0) {
*objid = vd->vdev_leaf_zap;
} else {
return (EINVAL);
}
return (0);
}
static int
vdev_prop_get_int(vdev_t *vd, vdev_prop_t prop, uint64_t *value)
{
@ -456,25 +473,29 @@ vdev_prop_get_int(vdev_t *vd, vdev_prop_t prop, uint64_t *value)
uint64_t objid;
int err;
if (vd->vdev_root_zap != 0) {
objid = vd->vdev_root_zap;
} else if (vd->vdev_top_zap != 0) {
objid = vd->vdev_top_zap;
} else if (vd->vdev_leaf_zap != 0) {
objid = vd->vdev_leaf_zap;
} else {
if (vdev_prop_get_objid(vd, &objid) != 0)
return (EINVAL);
}
err = zap_lookup(mos, objid, vdev_prop_to_name(prop),
sizeof (uint64_t), 1, value);
if (err == ENOENT)
*value = vdev_prop_default_numeric(prop);
return (err);
}
static int
vdev_prop_get_bool(vdev_t *vd, vdev_prop_t prop, boolean_t *bvalue)
{
int err;
uint64_t ivalue;
err = vdev_prop_get_int(vd, prop, &ivalue);
*bvalue = ivalue != 0;
return (err);
}
/*
* Get the number of data disks for a top-level vdev.
*/
@ -737,8 +758,12 @@ vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
*/
vd->vdev_checksum_n = vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_N);
vd->vdev_checksum_t = vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_T);
vd->vdev_io_n = vdev_prop_default_numeric(VDEV_PROP_IO_N);
vd->vdev_io_t = vdev_prop_default_numeric(VDEV_PROP_IO_T);
vd->vdev_slow_io_events = vdev_prop_default_numeric(
VDEV_PROP_SLOW_IO_EVENTS);
vd->vdev_slow_io_n = vdev_prop_default_numeric(VDEV_PROP_SLOW_IO_N);
vd->vdev_slow_io_t = vdev_prop_default_numeric(VDEV_PROP_SLOW_IO_T);
@ -3931,6 +3956,11 @@ vdev_load(vdev_t *vd)
vdev_dbgmsg(vd, "vdev_load: zap_lookup(zap=%llu) "
"failed [error=%d]", (u_longlong_t)zapobj, error);
error = vdev_prop_get_bool(vd, VDEV_PROP_SLOW_IO_EVENTS,
&vd->vdev_slow_io_events);
if (error && error != ENOENT)
vdev_dbgmsg(vd, "vdev_load: zap_lookup(zap=%llu) "
"failed [error=%d]", (u_longlong_t)zapobj, error);
error = vdev_prop_get_int(vd, VDEV_PROP_SLOW_IO_N,
&vd->vdev_slow_io_n);
if (error && error != ENOENT)
@ -5980,15 +6010,8 @@ vdev_props_set_sync(void *arg, dmu_tx_t *tx)
/*
* Set vdev property values in the vdev props mos object.
*/
if (vd->vdev_root_zap != 0) {
objid = vd->vdev_root_zap;
} else if (vd->vdev_top_zap != 0) {
objid = vd->vdev_top_zap;
} else if (vd->vdev_leaf_zap != 0) {
objid = vd->vdev_leaf_zap;
} else {
if (vdev_prop_get_objid(vd, &objid) != 0)
panic("unexpected vdev type");
}
mutex_enter(&spa->spa_props_lock);
@ -6215,6 +6238,13 @@ vdev_prop_set(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl)
}
vd->vdev_io_t = intval;
break;
case VDEV_PROP_SLOW_IO_EVENTS:
if (nvpair_value_uint64(elem, &intval) != 0) {
error = EINVAL;
break;
}
vd->vdev_slow_io_events = intval != 0;
break;
case VDEV_PROP_SLOW_IO_N:
if (nvpair_value_uint64(elem, &intval) != 0) {
error = EINVAL;
@ -6256,6 +6286,7 @@ vdev_prop_get(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl)
nvpair_t *elem = NULL;
nvlist_t *nvprops = NULL;
uint64_t intval = 0;
boolean_t boolval = 0;
char *strval = NULL;
const char *propname = NULL;
vdev_prop_t prop;
@ -6269,15 +6300,8 @@ vdev_prop_get(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl)
nvlist_lookup_nvlist(innvl, ZPOOL_VDEV_PROPS_GET_PROPS, &nvprops);
if (vd->vdev_root_zap != 0) {
objid = vd->vdev_root_zap;
} else if (vd->vdev_top_zap != 0) {
objid = vd->vdev_top_zap;
} else if (vd->vdev_leaf_zap != 0) {
objid = vd->vdev_leaf_zap;
} else {
if (vdev_prop_get_objid(vd, &objid) != 0)
return (SET_ERROR(EINVAL));
}
ASSERT(objid != 0);
mutex_enter(&spa->spa_props_lock);
@ -6622,6 +6646,18 @@ vdev_prop_get(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl)
intval, src);
break;
case VDEV_PROP_SLOW_IO_EVENTS:
err = vdev_prop_get_bool(vd, prop, &boolval);
if (err && err != ENOENT)
break;
src = ZPROP_SRC_LOCAL;
if (boolval == vdev_prop_default_numeric(prop))
src = ZPROP_SRC_DEFAULT;
vdev_prop_add_list(outnvl, propname, NULL,
boolval, src);
break;
case VDEV_PROP_CHECKSUM_N:
case VDEV_PROP_CHECKSUM_T:
case VDEV_PROP_IO_N:

3
module/zfs/zfs_fm.c
View File

@ -223,6 +223,9 @@ vdev_prop_get_inherited(vdev_t *vd, vdev_prop_t prop)
case VDEV_PROP_IO_T:
propval = vd->vdev_io_t;
break;
case VDEV_PROP_SLOW_IO_EVENTS:
propval = vd->vdev_slow_io_events;
break;
case VDEV_PROP_SLOW_IO_N:
propval = vd->vdev_slow_io_n;
break;

9
module/zfs/zio.c
View File

@ -5569,9 +5569,12 @@ zio_done(zio_t *zio)
zio->io_vd->vdev_stat.vs_slow_ios++;
mutex_exit(&zio->io_vd->vdev_stat_lock);
(void) zfs_ereport_post(FM_EREPORT_ZFS_DELAY,
zio->io_spa, zio->io_vd, &zio->io_bookmark,
zio, 0);
if (zio->io_vd->vdev_slow_io_events) {
(void) zfs_ereport_post(
FM_EREPORT_ZFS_DELAY,
zio->io_spa, zio->io_vd,
&zio->io_bookmark, zio, 0);
}
}
}
}

1
tests/zfs-tests/tests/functional/cli_root/zpool_get/vdev_get.cfg
View File

@ -71,6 +71,7 @@ typeset -a properties=(
checksum_t
io_n
io_t
slow_io_events
slow_io_n
slow_io_t
trim_support

44
tests/zfs-tests/tests/functional/events/zed_slow_io.ksh
View File

@ -23,6 +23,7 @@
#
# Copyright (c) 2023, Klara Inc.
# Copyright (c) 2025, Mariusz Zaborski <oshogbo@FreeBSD.org>
#
# DESCRIPTION:
@ -140,8 +141,8 @@ function slow_io_degrade
{
do_setup
zpool set slow_io_n=5 $TESTPOOL $VDEV
zpool set slow_io_t=60 $TESTPOOL $VDEV
log_must zpool set slow_io_n=5 $TESTPOOL $VDEV
log_must zpool set slow_io_t=60 $TESTPOOL $VDEV
start_slow_io
for i in {1..16}; do
@ -193,6 +194,44 @@ function slow_io_no_degrade
do_clean
}
# Change slow_io_n, slow_io_t to 5 events in 60 seconds
# fire more than 5 events. Disable slow io events.
# Should not degrade.
function slow_io_degrade_disabled
{
do_setup
log_must zpool set slow_io_n=5 $TESTPOOL $VDEV
log_must zpool set slow_io_t=60 $TESTPOOL $VDEV
log_must zpool set slow_io_events=off $TESTPOOL $VDEV
start_slow_io
for i in {1..16}; do
dd if=${FILEPATH}$i of=/dev/null count=1 bs=512 2>/dev/null
sleep 0.5
done
stop_slow_io
zpool sync
#
# wait 60 seconds to confirm that zfs.delay was not generated.
#
typeset -i i=0
typeset -i events=0
while [[ $i -lt 60 ]]; do
events=$(zpool events | grep "ereport\.fs\.zfs.delay" | wc -l)
i=$((i+1))
sleep 1
done
log_note "$events delay events found"
[ $events -eq "0" ] || \
log_fail "expecting no delay events, found $events"
log_mustnot wait_vdev_state $TESTPOOL $VDEV "DEGRADED" 45
do_clean
}
log_assert "Test ZED slow io configurability"
log_onexit cleanup
@ -202,5 +241,6 @@ log_must zed_start
default_degrade
slow_io_degrade
slow_io_no_degrade
slow_io_degrade_disabled
log_pass "Test ZED slow io configurability"