Skip to content
This repository has been archived by the owner on Nov 7, 2019. It is now read-only.

Commit

Permalink
3329 spa_sync() spends 10-20% of its time in spa_free_sync_cb()
Browse files Browse the repository at this point in the history
3330 space_seg_t should have its own kmem_cache
3331 deferred frees should happen after sync_pass 1
3335 make SYNC_PASS_* constants tunable
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com>
Reviewed by: Christopher Siden <chris.siden@delphix.com>
Reviewed by: Eric Schrock <eric.schrock@delphix.com>
Reviewed by: Richard Lowe <richlowe@richlowe.net>
Reviewed by: Dan McDonald <danmcd@nexenta.com>
Approved by: Eric Schrock <eric.schrock@delphix.com>
  • Loading branch information
grwilson authored and Christopher Siden committed Nov 7, 2012
1 parent c523716 commit 01f55e4
Show file tree
Hide file tree
Showing 9 changed files with 75 additions and 23 deletions.
5 changes: 3 additions & 2 deletions usr/src/uts/common/fs/zfs/metaslab.c
Expand Up @@ -875,8 +875,9 @@ metaslab_activate(metaslab_t *msp, uint64_t activation_weight)
if ((msp->ms_weight & METASLAB_ACTIVE_MASK) == 0) {
space_map_load_wait(sm);
if (!sm->sm_loaded) {
int error = space_map_load(sm, sm_ops, SM_FREE,
&msp->ms_smo,
space_map_obj_t *smo = &msp->ms_smo;

int error = space_map_load(sm, sm_ops, SM_FREE, smo,
spa_meta_objset(msp->ms_group->mg_vd->vdev_spa));
if (error) {
metaslab_group_sort(msp->ms_group, msp, 0);
Expand Down
3 changes: 2 additions & 1 deletion usr/src/uts/common/fs/zfs/spa.c
Expand Up @@ -130,6 +130,7 @@ boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
uint_t zio_taskq_basedc = 80; /* base duty cycle */

boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
extern int zfs_sync_pass_deferred_free;

/*
* This (illegal) pool name is used when temporarily importing a spa_t in order
Expand Down Expand Up @@ -6040,7 +6041,7 @@ spa_sync(spa_t *spa, uint64_t txg)
spa_errlog_sync(spa, txg);
dsl_pool_sync(dp, txg);

if (pass <= SYNC_PASS_DEFERRED_FREE) {
if (pass < zfs_sync_pass_deferred_free) {
zio_t *zio = zio_root(spa, NULL, NULL, 0);
bplist_iterate(free_bpl, spa_free_sync_cb,
zio, tx);
Expand Down
2 changes: 2 additions & 0 deletions usr/src/uts/common/fs/zfs/spa_misc.c
Expand Up @@ -1687,6 +1687,7 @@ spa_init(int mode)

refcount_init();
unique_init();
space_map_init();
zio_init();
dmu_init();
zil_init();
Expand All @@ -1709,6 +1710,7 @@ spa_fini(void)
zil_fini();
dmu_fini();
zio_fini();
space_map_fini();
unique_fini();
refcount_fini();

Expand Down
42 changes: 34 additions & 8 deletions usr/src/uts/common/fs/zfs/space_map.c
Expand Up @@ -32,6 +32,23 @@
#include <sys/zio.h>
#include <sys/space_map.h>

static kmem_cache_t *space_seg_cache;

void
space_map_init(void)
{
ASSERT(space_seg_cache == NULL);
space_seg_cache = kmem_cache_create("space_seg_cache",
sizeof (space_seg_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
}

void
space_map_fini(void)
{
kmem_cache_destroy(space_seg_cache);
space_seg_cache = NULL;
}

/*
* Space map routines.
* NOTE: caller is responsible for all locking.
Expand Down Expand Up @@ -124,7 +141,7 @@ space_map_add(space_map_t *sm, uint64_t start, uint64_t size)
avl_remove(sm->sm_pp_root, ss_after);
}
ss_after->ss_start = ss_before->ss_start;
kmem_free(ss_before, sizeof (*ss_before));
kmem_cache_free(space_seg_cache, ss_before);
ss = ss_after;
} else if (merge_before) {
ss_before->ss_end = end;
Expand All @@ -137,7 +154,7 @@ space_map_add(space_map_t *sm, uint64_t start, uint64_t size)
avl_remove(sm->sm_pp_root, ss_after);
ss = ss_after;
} else {
ss = kmem_alloc(sizeof (*ss), KM_SLEEP);
ss = kmem_cache_alloc(space_seg_cache, KM_SLEEP);
ss->ss_start = start;
ss->ss_end = end;
avl_insert(&sm->sm_root, ss, where);
Expand Down Expand Up @@ -184,7 +201,7 @@ space_map_remove(space_map_t *sm, uint64_t start, uint64_t size)
avl_remove(sm->sm_pp_root, ss);

if (left_over && right_over) {
newseg = kmem_alloc(sizeof (*newseg), KM_SLEEP);
newseg = kmem_cache_alloc(space_seg_cache, KM_SLEEP);
newseg->ss_start = end;
newseg->ss_end = ss->ss_end;
ss->ss_end = start;
Expand All @@ -197,7 +214,7 @@ space_map_remove(space_map_t *sm, uint64_t start, uint64_t size)
ss->ss_start = end;
} else {
avl_remove(&sm->sm_root, ss);
kmem_free(ss, sizeof (*ss));
kmem_cache_free(space_seg_cache, ss);
ss = NULL;
}

Expand Down Expand Up @@ -237,7 +254,7 @@ space_map_vacate(space_map_t *sm, space_map_func_t *func, space_map_t *mdest)
while ((ss = avl_destroy_nodes(&sm->sm_root, &cookie)) != NULL) {
if (func != NULL)
func(mdest, ss->ss_start, ss->ss_end - ss->ss_start);
kmem_free(ss, sizeof (*ss));
kmem_cache_free(space_seg_cache, ss);
}
sm->sm_space = 0;
}
Expand Down Expand Up @@ -409,7 +426,7 @@ space_map_sync(space_map_t *sm, uint8_t maptype,
spa_t *spa = dmu_objset_spa(os);
void *cookie = NULL;
space_seg_t *ss;
uint64_t bufsize, start, size, run_len;
uint64_t bufsize, start, size, run_len, delta, sm_space;
uint64_t *entry, *entry_map, *entry_map_end;

ASSERT(MUTEX_HELD(sm->sm_lock));
Expand Down Expand Up @@ -438,11 +455,13 @@ space_map_sync(space_map_t *sm, uint8_t maptype,
SM_DEBUG_SYNCPASS_ENCODE(spa_sync_pass(spa)) |
SM_DEBUG_TXG_ENCODE(dmu_tx_get_txg(tx));

delta = 0;
sm_space = sm->sm_space;
while ((ss = avl_destroy_nodes(&sm->sm_root, &cookie)) != NULL) {
size = ss->ss_end - ss->ss_start;
start = (ss->ss_start - sm->sm_start) >> sm->sm_shift;

sm->sm_space -= size;
delta += size;
size >>= sm->sm_shift;

while (size) {
Expand All @@ -464,7 +483,7 @@ space_map_sync(space_map_t *sm, uint8_t maptype,
start += run_len;
size -= run_len;
}
kmem_free(ss, sizeof (*ss));
kmem_cache_free(space_seg_cache, ss);
}

if (entry != entry_map) {
Expand All @@ -476,8 +495,15 @@ space_map_sync(space_map_t *sm, uint8_t maptype,
smo->smo_objsize += size;
}

/*
* Ensure that the space_map's accounting wasn't changed
* while we were in the middle of writing it out.
*/
VERIFY3U(sm->sm_space, ==, sm_space);

zio_buf_free(entry_map, bufsize);

sm->sm_space -= delta;
VERIFY0(sm->sm_space);
}

Expand Down
5 changes: 4 additions & 1 deletion usr/src/uts/common/fs/zfs/sys/metaslab_impl.h
Expand Up @@ -21,7 +21,10 @@
/*
* Copyright 2009 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
* Copyright (c) 2011 by Delphix. All rights reserved.
*/

/*
* Copyright (c) 2012 by Delphix. All rights reserved.
*/

#ifndef _SYS_METASLAB_IMPL_H
Expand Down
8 changes: 0 additions & 8 deletions usr/src/uts/common/fs/zfs/sys/spa.h
Expand Up @@ -486,14 +486,6 @@ extern int spa_scan_stop(spa_t *spa);
extern void spa_sync(spa_t *spa, uint64_t txg); /* only for DMU use */
extern void spa_sync_allpools(void);

/*
* DEFERRED_FREE must be large enough that regular blocks are not
* deferred. XXX so can't we change it back to 1?
*/
#define SYNC_PASS_DEFERRED_FREE 2 /* defer frees after this pass */
#define SYNC_PASS_DONT_COMPRESS 4 /* don't compress after this pass */
#define SYNC_PASS_REWRITE 1 /* rewrite new bps after this pass */

/* spa namespace global mutex */
extern kmutex_t spa_namespace_lock;

Expand Down
6 changes: 6 additions & 0 deletions usr/src/uts/common/fs/zfs/sys/space_map.h
Expand Up @@ -23,6 +23,10 @@
* Use is subject to license terms.
*/

/*
* Copyright (c) 2012 by Delphix. All rights reserved.
*/

#ifndef _SYS_SPACE_MAP_H
#define _SYS_SPACE_MAP_H

Expand Down Expand Up @@ -136,6 +140,8 @@ struct space_map_ops {

typedef void space_map_func_t(space_map_t *sm, uint64_t start, uint64_t size);

extern void space_map_init(void);
extern void space_map_fini(void);
extern void space_map_create(space_map_t *sm, uint64_t start, uint64_t size,
uint8_t shift, kmutex_t *lp);
extern void space_map_destroy(space_map_t *sm);
Expand Down
5 changes: 5 additions & 0 deletions usr/src/uts/common/fs/zfs/sys/zio_impl.h
Expand Up @@ -23,6 +23,10 @@
* Use is subject to license terms.
*/

/*
* Copyright (c) 2012 by Delphix. All rights reserved.
*/

#ifndef _ZIO_IMPL_H
#define _ZIO_IMPL_H

Expand Down Expand Up @@ -143,6 +147,7 @@ enum zio_stage {
#define ZIO_FREE_PIPELINE \
(ZIO_INTERLOCK_STAGES | \
ZIO_STAGE_FREE_BP_INIT | \
ZIO_STAGE_ISSUE_ASYNC | \
ZIO_STAGE_DVA_FREE)

#define ZIO_DDT_FREE_PIPELINE \
Expand Down
22 changes: 19 additions & 3 deletions usr/src/uts/common/fs/zfs/zio.c
Expand Up @@ -82,6 +82,22 @@ extern vmem_t *zio_alloc_arena;
#endif
extern int zfs_mg_alloc_failures;

/*
* The following actions directly effect the spa's sync-to-convergence logic.
* The values below define the sync pass when we start performing the action.
* Care should be taken when changing these values as they directly impact
* spa_sync() performance. Tuning these values may introduce subtle performance
* pathologies and should only be done in the context of performance analysis.
* These tunables will eventually be removed and replaced with #defines once
* enough analysis has been done to determine optimal values.
*
* The 'zfs_sync_pass_deferred_free' pass must be greater than 1 to ensure that
* regular blocks are not deferred.
*/
int zfs_sync_pass_deferred_free = 2; /* defer frees starting in this pass */
int zfs_sync_pass_dont_compress = 5; /* don't compress starting in this pass */
int zfs_sync_pass_rewrite = 2; /* rewrite new bps starting in this pass */

/*
* An allocating zio is one that either currently has the DVA allocate
* stage set or will have it later in its lifetime.
Expand Down Expand Up @@ -690,7 +706,7 @@ zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,

ASSERT(!BP_IS_HOLE(bp));
ASSERT(spa_syncing_txg(spa) == txg);
ASSERT(spa_sync_pass(spa) <= SYNC_PASS_DEFERRED_FREE);
ASSERT(spa_sync_pass(spa) < zfs_sync_pass_deferred_free);

zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp),
NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_FREE, flags,
Expand Down Expand Up @@ -987,7 +1003,7 @@ zio_write_bp_init(zio_t *zio)
ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
ASSERT(!BP_GET_DEDUP(bp));

if (pass > SYNC_PASS_DONT_COMPRESS)
if (pass >= zfs_sync_pass_dont_compress)
compress = ZIO_COMPRESS_OFF;

/* Make sure someone doesn't change their mind on overwrites */
Expand Down Expand Up @@ -1016,7 +1032,7 @@ zio_write_bp_init(zio_t *zio)
* There should only be a handful of blocks after pass 1 in any case.
*/
if (bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == psize &&
pass > SYNC_PASS_REWRITE) {
pass >= zfs_sync_pass_rewrite) {
ASSERT(psize != 0);
enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES;
zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages;
Expand Down

0 comments on commit 01f55e4

Please sign in to comment.