c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2026-05-24 11:18:52 +03:00
Code Issues Packages Projects Releases Wiki Activity

Detect a slow raidz child during reads

Browse Source 
A single slow responding disk can affect the overall read
performance of a raidz group.  When a raidz child disk is
determined to be a persistent slow outlier, then have it
sit out during reads for a period of time. The raidz group
can use parity to reconstruct the data that was skipped.

Each time a slow disk is placed into a sit out period, its
`vdev_stat.vs_slow_ios count` is incremented and a zevent
class `ereport.fs.zfs.delay` is posted.

The length of the sit out period can be changed using the
`raid_read_sit_out_secs` module parameter.  Setting it to
zero disables slow outlier detection.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Paul Dagnelie <paul.dagnelie@klarasystems.com>
Contributions-by: Don Brady <don.brady@klarasystems.com>
Contributions-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #17227
This commit is contained in:
Paul Dagnelie
2025-08-27 16:41:48 -07:00
committed by Brian Behlendorf
parent 0df85ec27c
commit df55ba7c49
28 changed files with 1399 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files
module/zfs/vdev.c
+116 -1
View File
@@ -29,7 +29,7 @@
* Copyright 2017 Joyent, Inc.
* Copyright (c) 2017, Intel Corporation.
* Copyright (c) 2019, Datto Inc. All rights reserved.
* Copyright (c) 2021, Klara Inc.
* Copyright (c) 2021, 2025, Klara, Inc.
* Copyright (c) 2021, 2023 Hewlett Packard Enterprise Development LP.
*/
@@ -1086,6 +1086,10 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
}
}
if (top_level && (ops == &vdev_raidz_ops || ops == &vdev_draid_ops))
vd->vdev_autosit =
vdev_prop_default_numeric(VDEV_PROP_AUTOSIT);
/*
* Add ourselves to the parent's list of children.
*/
@@ -1187,6 +1191,9 @@ vdev_free(vdev_t *vd)
spa_spare_remove(vd);
if (vd->vdev_isl2cache)
spa_l2cache_remove(vd);
if (vd->vdev_prev_histo)
kmem_free(vd->vdev_prev_histo,
sizeof (uint64_t) * VDEV_L_HISTO_BUCKETS);
txg_list_destroy(&vd->vdev_ms_list);
txg_list_destroy(&vd->vdev_dtl_list);
@@ -3857,6 +3864,26 @@ vdev_load(vdev_t *vd)
}
}
if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
spa_t *spa = vd->vdev_spa;
uint64_t autosit;
error = zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap,
vdev_prop_to_name(VDEV_PROP_AUTOSIT), sizeof (autosit),
1, &autosit);
if (error == 0) {
vd->vdev_autosit = autosit == 1;
} else if (error == ENOENT) {
vd->vdev_autosit = vdev_prop_default_numeric(
VDEV_PROP_AUTOSIT);
} else {
vdev_dbgmsg(vd,
"vdev_load: zap_lookup(top_zap=%llu) "
"failed [error=%d]",
(u_longlong_t)vd->vdev_top_zap, error);
}
}
/*
* Load any rebuild state from the top-level vdev zap.
*/
@@ -4616,6 +4643,8 @@ vdev_clear(spa_t *spa, vdev_t *vd)
vd->vdev_stat.vs_checksum_errors = 0;
vd->vdev_stat.vs_dio_verify_errors = 0;
vd->vdev_stat.vs_slow_ios = 0;
atomic_store_64(&vd->vdev_outlier_count, 0);
vd->vdev_read_sit_out_expire = 0;
for (int c = 0; c < vd->vdev_children; c++)
vdev_clear(spa, vd->vdev_child[c]);
@@ -6107,6 +6136,56 @@ vdev_prop_set(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl)
}
vd->vdev_failfast = intval & 1;
break;
case VDEV_PROP_SIT_OUT:
/* Only expose this for a draid or raidz leaf */
if (!vd->vdev_ops->vdev_op_leaf ||
vd->vdev_top == NULL ||
(vd->vdev_top->vdev_ops != &vdev_raidz_ops &&
vd->vdev_top->vdev_ops != &vdev_draid_ops)) {
error = ENOTSUP;
break;
}
if (nvpair_value_uint64(elem, &intval) != 0) {
error = EINVAL;
break;
}
if (intval == 1) {
vdev_t *ancestor = vd;
while (ancestor->vdev_parent != vd->vdev_top)
ancestor = ancestor->vdev_parent;
vdev_t *pvd = vd->vdev_top;
uint_t sitouts = 0;
for (int i = 0; i < pvd->vdev_children; i++) {
if (pvd->vdev_child[i] == ancestor)
continue;
if (vdev_sit_out_reads(
pvd->vdev_child[i], 0)) {
sitouts++;
}
}
if (sitouts >= vdev_get_nparity(pvd)) {
error = ZFS_ERR_TOO_MANY_SITOUTS;
break;
}
if (error == 0)
vdev_raidz_sit_child(vd,
INT64_MAX - gethrestime_sec());
} else {
vdev_raidz_unsit_child(vd);
}
break;
case VDEV_PROP_AUTOSIT:
if (vd->vdev_ops != &vdev_raidz_ops &&
vd->vdev_ops != &vdev_draid_ops) {
error = ENOTSUP;
break;
}
if (nvpair_value_uint64(elem, &intval) != 0) {
error = EINVAL;
break;
}
vd->vdev_autosit = intval == 1;
break;
case VDEV_PROP_CHECKSUM_N:
if (nvpair_value_uint64(elem, &intval) != 0) {
error = EINVAL;
@@ -6456,6 +6535,19 @@ vdev_prop_get(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl)
ZPROP_SRC_NONE);
}
continue;
case VDEV_PROP_SIT_OUT:
/* Only expose this for a draid or raidz leaf */
if (vd->vdev_ops->vdev_op_leaf &&
vd->vdev_top != NULL &&
(vd->vdev_top->vdev_ops ==
&vdev_raidz_ops ||
vd->vdev_top->vdev_ops ==
&vdev_draid_ops)) {
vdev_prop_add_list(outnvl, propname,
NULL, vdev_sit_out_reads(vd, 0),
ZPROP_SRC_NONE);
}
continue;
case VDEV_PROP_TRIM_SUPPORT:
/* only valid for leaf vdevs */
if (vd->vdev_ops->vdev_op_leaf) {
@@ -6506,6 +6598,29 @@ vdev_prop_get(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl)
vdev_prop_add_list(outnvl, propname, strval,
intval, src);
break;
case VDEV_PROP_AUTOSIT:
/* Only raidz vdevs cannot have this property */
if (vd->vdev_ops != &vdev_raidz_ops &&
vd->vdev_ops != &vdev_draid_ops) {
src = ZPROP_SRC_NONE;
intval = ZPROP_BOOLEAN_NA;
} else {
err = vdev_prop_get_int(vd, prop,
&intval);
if (err && err != ENOENT)
break;
if (intval ==
vdev_prop_default_numeric(prop))
src = ZPROP_SRC_DEFAULT;
else
src = ZPROP_SRC_LOCAL;
}
vdev_prop_add_list(outnvl, propname, NULL,
intval, src);
break;
case VDEV_PROP_CHECKSUM_N:
case VDEV_PROP_CHECKSUM_T:
case VDEV_PROP_IO_N: