Allow and prefer special vdevs as ZIL

Before this change ZIL blocks were allocated only from normal or
SLOG vdevs.  In typical situation when special vdevs are SSDs and
normal are HDDs it could cause weird inversions when data blocks
are written to SSDs, but ZIL referencing them to HDDs.

This change assumes that special vdevs typically have much better
(or at least not worse) latency than normal, and so in absence of
SLOGs should store ZIL blocks.  It means similar to normal vdevs
introduction of special embedded log allocation class and updating
the allocation fallback order to: SLOG -> special embedded log ->
special -> normal embedded log -> normal.

The code tries to guess whether data block is going to be written
to normal or special vdev (it can not be done precisely before
compression) and prefer indirect writes for blocks written to a
special vdev to avoid double-write.  For blocks that are going to
be written to normal vdev, special vdev by default plays as SLOG,
reducing write latency by the cost of higher special vdev wear,
but it is tunable via module parameter.

This should allow HDD pools with decent SSD as special vdev to
work under synchronous workloads without requiring additional
SLOG SSD, impractical in many scenarios.

Reviewed-by: Brian Behlendorf <[email protected]>
Reviewed-by: Rob Norris <[email protected]>
Reviewed-by: Paul Dagnelie <[email protected]>
Signed-off-by:	Alexander Motin <[email protected]>
Sponsored by:	iXsystems, Inc.
Closes #17505

Author: amotin

cmd/zdb/zdb.c

@@ -6750,6 +6750,7 @@ zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
 	spa->spa_normal_class->mc_ops = &zdb_metaslab_ops;
 	spa->spa_log_class->mc_ops = &zdb_metaslab_ops;
 	spa->spa_embedded_log_class->mc_ops = &zdb_metaslab_ops;
+	spa->spa_special_embedded_log_class->mc_ops = &zdb_metaslab_ops;
 
 	zcb->zcb_vd_obsolete_counts =
 	    umem_zalloc(rvd->vdev_children * sizeof (uint32_t *),
@@ -6887,7 +6888,9 @@ zdb_leak_fini(spa_t *spa, zdb_cb_t *zcb)
 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
 			metaslab_t *msp = vd->vdev_ms[m];
 			ASSERT3P(msp->ms_group, ==, (msp->ms_group->mg_class ==
-			    spa_embedded_log_class(spa)) ?
+			    spa_embedded_log_class(spa) ||
+			    msp->ms_group->mg_class ==
+			    spa_special_embedded_log_class(spa)) ?
 			    vd->vdev_log_mg : vd->vdev_mg);
 
 			/*
@@ -7121,6 +7124,8 @@ dump_block_stats(spa_t *spa)
 	zcb->zcb_totalasize += metaslab_class_get_alloc(spa_dedup_class(spa));
 	zcb->zcb_totalasize +=
 	    metaslab_class_get_alloc(spa_embedded_log_class(spa));
+	zcb->zcb_totalasize +=
+	    metaslab_class_get_alloc(spa_special_embedded_log_class(spa));
 	zcb->zcb_start = zcb->zcb_lastprint = gethrtime();
 	err = traverse_pool(spa, 0, flags, zdb_blkptr_cb, zcb);
 
@@ -7169,6 +7174,7 @@ dump_block_stats(spa_t *spa)
 	total_alloc = norm_alloc +
 	    metaslab_class_get_alloc(spa_log_class(spa)) +
 	    metaslab_class_get_alloc(spa_embedded_log_class(spa)) +
+	    metaslab_class_get_alloc(spa_special_embedded_log_class(spa)) +
 	    metaslab_class_get_alloc(spa_special_class(spa)) +
 	    metaslab_class_get_alloc(spa_dedup_class(spa)) +
 	    get_unflushed_alloc_space(spa);
@@ -7252,6 +7258,18 @@ dump_block_stats(spa_t *spa)
 		    100.0 * alloc / space);
 	}
 
+	if (spa_special_embedded_log_class(spa)->mc_allocator[0].mca_rotor
+	    != NULL) {
+		uint64_t alloc = metaslab_class_get_alloc(
+		    spa_special_embedded_log_class(spa));
+		uint64_t space = metaslab_class_get_space(
+		    spa_special_embedded_log_class(spa));
+
+		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
+		    "Special embedded log", (u_longlong_t)alloc,
+		    100.0 * alloc / space);
+	}
+
 	for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
 		if (zcb->zcb_embedded_blocks[i] == 0)
 			continue;

include/sys/spa.h

@@ -1065,6 +1065,7 @@ extern metaslab_class_t *spa_normal_class(spa_t *spa);
 extern metaslab_class_t *spa_log_class(spa_t *spa);
 extern metaslab_class_t *spa_embedded_log_class(spa_t *spa);
 extern metaslab_class_t *spa_special_class(spa_t *spa);
+extern metaslab_class_t *spa_special_embedded_log_class(spa_t *spa);
 extern metaslab_class_t *spa_dedup_class(spa_t *spa);
 extern metaslab_class_t *spa_preferred_class(spa_t *spa, const zio_t *zio);
 extern boolean_t spa_special_has_ddt(spa_t *spa);

include/sys/spa_impl.h

@@ -246,6 +246,7 @@ struct spa {
 	metaslab_class_t *spa_log_class;	/* intent log data class */
 	metaslab_class_t *spa_embedded_log_class; /* log on normal vdevs */
 	metaslab_class_t *spa_special_class;	/* special allocation class */
+	metaslab_class_t *spa_special_embedded_log_class; /* log on special */
 	metaslab_class_t *spa_dedup_class;	/* dedup allocation class */
 	uint64_t	spa_first_txg;		/* first txg after spa_open() */
 	uint64_t	spa_final_txg;		/* txg of export/destroy */

include/sys/zil.h

@@ -635,13 +635,17 @@ extern void	zil_set_logbias(zilog_t *zilog, uint64_t slogval);
 
 extern uint64_t	zil_max_copied_data(zilog_t *zilog);
 extern uint64_t	zil_max_log_data(zilog_t *zilog, size_t hdrsize);
+extern itx_wr_state_t zil_write_state(zilog_t *zilog, uint64_t size,
+    uint32_t blocksize, boolean_t o_direct, boolean_t commit);
 
 extern void zil_sums_init(zil_sums_t *zs);
 extern void zil_sums_fini(zil_sums_t *zs);
 extern void zil_kstat_values_update(zil_kstat_values_t *zs,
     zil_sums_t *zil_sums);
 
 extern int zil_replay_disable;
+extern uint_t zfs_immediate_write_sz;
+extern int zil_special_is_slog;
 
 #ifdef	__cplusplus
 }

man/man4/zfs.4

@@ -1713,10 +1713,18 @@ Similar to
 but for cleanup of old indirection records for removed vdevs.
 .
 .It Sy zfs_immediate_write_sz Ns = Ns Sy 32768 Ns B Po 32 KiB Pc Pq s64
-Largest data block to write to the ZIL.
-Larger blocks will be treated as if the dataset being written to had the
-.Sy logbias Ns = Ns Sy throughput
-property set.
+Largest write size to store the data directly into the ZIL if
+.Sy logbias Ns = Ns Sy latency .
+Larger writes may be written indirectly similar to
+.Sy logbias Ns = Ns Sy throughput .
+In presence of SLOG this parameter is ignored, as if it was set to infinity,
+storing all written data into ZIL to not depend on regular vdev latency.
+.
+.It Sy zil_special_is_slog Ns = Ns Sy 1 Ns | Ns 0 Pq int
+When enabled, and written blocks go to normal vdevs, treat present special
+vdevs as SLOGs, storing all synchronously written data into ZIL directly.
+Disabling this forces the indirect writes to preserve special vdev write
+throughput and endurance, likely at the cost of normal vdev latency.
 .
 .It Sy zfs_initialize_value Ns = Ns Sy 16045690984833335022 Po 0xDEADBEEFDEADBEEE Pc Pq u64
 Pattern written to vdev free space by

module/zfs/metaslab.c

@@ -750,7 +750,8 @@ metaslab_class_histogram_verify(metaslab_class_t *mc)
 		}
 
 		IMPLY(mg == mg->mg_vd->vdev_log_mg,
-		    mc == spa_embedded_log_class(mg->mg_vd->vdev_spa));
+		    mc == spa_embedded_log_class(mg->mg_vd->vdev_spa) ||
+		    mc == spa_special_embedded_log_class(mg->mg_vd->vdev_spa));
 
 		for (i = 0; i < ZFS_RANGE_TREE_HISTOGRAM_SIZE; i++)
 			mc_hist[i] += mg->mg_histogram[i];
@@ -1288,7 +1289,8 @@ metaslab_group_histogram_add(metaslab_group_t *mg, metaslab_t *msp)
 	mutex_enter(&mc->mc_lock);
 	for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
 		IMPLY(mg == mg->mg_vd->vdev_log_mg,
-		    mc == spa_embedded_log_class(mg->mg_vd->vdev_spa));
+		    mc == spa_embedded_log_class(mg->mg_vd->vdev_spa) ||
+		    mc == spa_special_embedded_log_class(mg->mg_vd->vdev_spa));
 		mg->mg_histogram[i + ashift] +=
 		    msp->ms_sm->sm_phys->smp_histogram[i];
 		mc->mc_histogram[i + ashift] +=
@@ -1316,7 +1318,8 @@ metaslab_group_histogram_remove(metaslab_group_t *mg, metaslab_t *msp)
 		ASSERT3U(mc->mc_histogram[i + ashift], >=,
 		    msp->ms_sm->sm_phys->smp_histogram[i]);
 		IMPLY(mg == mg->mg_vd->vdev_log_mg,
-		    mc == spa_embedded_log_class(mg->mg_vd->vdev_spa));
+		    mc == spa_embedded_log_class(mg->mg_vd->vdev_spa) ||
+		    mc == spa_special_embedded_log_class(mg->mg_vd->vdev_spa));
 
 		mg->mg_histogram[i + ashift] -=
 		    msp->ms_sm->sm_phys->smp_histogram[i];

module/zfs/spa.c

@@ -417,11 +417,15 @@ spa_prop_get_config(spa_t *spa, nvlist_t *nv)
 		alloc += metaslab_class_get_alloc(spa_special_class(spa));
 		alloc += metaslab_class_get_alloc(spa_dedup_class(spa));
 		alloc += metaslab_class_get_alloc(spa_embedded_log_class(spa));
+		alloc += metaslab_class_get_alloc(
+		    spa_special_embedded_log_class(spa));
 
 		size = metaslab_class_get_space(mc);
 		size += metaslab_class_get_space(spa_special_class(spa));
 		size += metaslab_class_get_space(spa_dedup_class(spa));
 		size += metaslab_class_get_space(spa_embedded_log_class(spa));
+		size += metaslab_class_get_space(
+		    spa_special_embedded_log_class(spa));
 
 		spa_prop_add_list(nv, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
 		spa_prop_add_list(nv, ZPOOL_PROP_SIZE, NULL, size, src);
@@ -1679,6 +1683,8 @@ spa_activate(spa_t *spa, spa_mode_t mode)
 	    "embedded_log", msp, B_TRUE);
 	spa->spa_special_class = metaslab_class_create(spa, "special",
 	    msp, B_FALSE);
+	spa->spa_special_embedded_log_class = metaslab_class_create(spa,
+	    "special_embedded_log", msp, B_TRUE);
 	spa->spa_dedup_class = metaslab_class_create(spa, "dedup",
 	    msp, B_FALSE);
 
@@ -1853,6 +1859,9 @@ spa_deactivate(spa_t *spa)
 	metaslab_class_destroy(spa->spa_special_class);
 	spa->spa_special_class = NULL;
 
+	metaslab_class_destroy(spa->spa_special_embedded_log_class);
+	spa->spa_special_embedded_log_class = NULL;
+
 	metaslab_class_destroy(spa->spa_dedup_class);
 	spa->spa_dedup_class = NULL;
 
@@ -9092,6 +9101,8 @@ spa_async_thread(void *arg)
 		old_space += metaslab_class_get_space(spa_dedup_class(spa));
 		old_space += metaslab_class_get_space(
 		    spa_embedded_log_class(spa));
+		old_space += metaslab_class_get_space(
+		    spa_special_embedded_log_class(spa));
 
 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
 
@@ -9100,6 +9111,8 @@ spa_async_thread(void *arg)
 		new_space += metaslab_class_get_space(spa_dedup_class(spa));
 		new_space += metaslab_class_get_space(
 		    spa_embedded_log_class(spa));
+		new_space += metaslab_class_get_space(
+		    spa_special_embedded_log_class(spa));
 		mutex_exit(&spa_namespace_lock);
 
 		/*
@@ -10309,7 +10322,7 @@ spa_sync(spa_t *spa, uint64_t txg)
 
 	metaslab_class_evict_old(spa->spa_normal_class, txg);
 	metaslab_class_evict_old(spa->spa_log_class, txg);
-	/* spa_embedded_log_class has only one metaslab per vdev. */
+	/* Embedded log classes have only one metaslab per vdev. */
 	metaslab_class_evict_old(spa->spa_special_class, txg);
 	metaslab_class_evict_old(spa->spa_dedup_class, txg);
 

module/zfs/spa_misc.c

@@ -1308,6 +1308,7 @@ spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error,
 	metaslab_class_validate(spa_log_class(spa));
 	metaslab_class_validate(spa_embedded_log_class(spa));
 	metaslab_class_validate(spa_special_class(spa));
+	metaslab_class_validate(spa_special_embedded_log_class(spa));
 	metaslab_class_validate(spa_dedup_class(spa));
 
 	spa_config_exit(spa, SCL_ALL, spa);
@@ -1896,6 +1897,8 @@ spa_get_slop_space(spa_t *spa)
 	 */
 	uint64_t embedded_log =
 	    metaslab_class_get_dspace(spa_embedded_log_class(spa));
+	embedded_log += metaslab_class_get_dspace(
+	    spa_special_embedded_log_class(spa));
 	slop -= MIN(embedded_log, slop >> 1);
 
 	/*
@@ -2000,6 +2003,12 @@ spa_special_class(spa_t *spa)
 	return (spa->spa_special_class);
 }
 
+metaslab_class_t *
+spa_special_embedded_log_class(spa_t *spa)
+{
+	return (spa->spa_special_embedded_log_class);
+}
+
 metaslab_class_t *
 spa_dedup_class(spa_t *spa)
 {

module/zfs/vdev.c

@@ -282,12 +282,15 @@ vdev_getops(const char *type)
  * Given a vdev and a metaslab class, find which metaslab group we're
  * interested in. All vdevs may belong to two different metaslab classes.
  * Dedicated slog devices use only the primary metaslab group, rather than a
- * separate log group. For embedded slogs, the vdev_log_mg will be non-NULL.
+ * separate log group.  For embedded slogs, vdev_log_mg will be non-NULL and
+ * will point to a metaslab group of either embedded_log_class (for normal
+ * vdevs) or special_embedded_log_class (for special vdevs).
  */
 metaslab_group_t *
 vdev_get_mg(vdev_t *vd, metaslab_class_t *mc)
 {
-	if (mc == spa_embedded_log_class(vd->vdev_spa) &&
+	if ((mc == spa_embedded_log_class(vd->vdev_spa) ||
+	    mc == spa_special_embedded_log_class(vd->vdev_spa)) &&
 	    vd->vdev_log_mg != NULL)
 		return (vd->vdev_log_mg);
 	else
@@ -1508,8 +1511,13 @@ vdev_metaslab_group_create(vdev_t *vd)
 		vd->vdev_mg = metaslab_group_create(mc, vd);
 
 		if (!vd->vdev_islog) {
-			vd->vdev_log_mg = metaslab_group_create(
-			    spa_embedded_log_class(spa), vd);
+			if (mc == spa_special_class(spa)) {
+				vd->vdev_log_mg = metaslab_group_create(
+				    spa_special_embedded_log_class(spa), vd);
+			} else {
+				vd->vdev_log_mg = metaslab_group_create(
+				    spa_embedded_log_class(spa), vd);
+			}
 		}
 
 		/*
@@ -1624,9 +1632,10 @@ vdev_metaslab_init(vdev_t *vd, uint64_t txg)
 	/*
 	 * Find the emptiest metaslab on the vdev and mark it for use for
 	 * embedded slog by moving it from the regular to the log metaslab
-	 * group.
+	 * group.  This works for normal and special vdevs.
 	 */
-	if (vd->vdev_mg->mg_class == spa_normal_class(spa) &&
+	if ((vd->vdev_mg->mg_class == spa_normal_class(spa) ||
+	    vd->vdev_mg->mg_class == spa_special_class(spa)) &&
 	    vd->vdev_ms_count > zfs_embedded_slog_min_ms &&
 	    avl_is_empty(&vd->vdev_log_mg->mg_metaslab_tree)) {
 		uint64_t slog_msid = 0;

module/zfs/zfs_log.c

@@ -607,8 +607,6 @@ zfs_log_rename_whiteout(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
  * called as soon as the write is on stable storage (be it via a DMU sync or a
  * ZIL commit).
  */
-static uint_t zfs_immediate_write_sz = 32768;
-
 void
 zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype,
     znode_t *zp, offset_t off, ssize_t resid, boolean_t commit,
@@ -626,15 +624,8 @@ zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype,
 		return;
 	}
 
-	if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT || o_direct)
-		write_state = WR_INDIRECT;
-	else if (!spa_has_slogs(zilog->zl_spa) &&
-	    resid >= zfs_immediate_write_sz)
-		write_state = WR_INDIRECT;
-	else if (commit)
-		write_state = WR_COPIED;
-	else
-		write_state = WR_NEED_COPY;
+	write_state = zil_write_state(zilog, resid, blocksize, o_direct,
+	    commit);
 
 	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(ZTOZSB(zp)), &gen,
 	    sizeof (gen));
@@ -938,6 +929,3 @@ zfs_log_clone_range(zilog_t *zilog, dmu_tx_t *tx, int txtype, znode_t *zp,
 		len -= partlen;
 	}
 }
-
-ZFS_MODULE_PARAM(zfs, zfs_, immediate_write_sz, UINT, ZMOD_RW,
-	"Largest data block to write to zil");