DLPX-51539 Improve I/O performance of metaslab spacemaps with a log s…

…pacemap

DLPX-41227 metaslab_sync_done should not wait for metaslab_load to complete

Reviewed at: http://reviews.delphix.com/r/37308/

usr/src/cmd/ztest/ztest.c

@@ -2563,24 +2563,12 @@ vdev_lookup_by_path(vdev_t *vd, const char *path)
 	return (NULL);
 }
 
-/*
- * Find the first available hole which can be used as a top-level.
- */
-int
-find_vdev_hole(spa_t *spa)
+static int
+spa_num_top_vdevs(spa_t *spa)
 {
 	vdev_t *rvd = spa->spa_root_vdev;
-	int c;
-
-	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
-
-	for (c = 0; c < rvd->vdev_children; c++) {
-		vdev_t *cvd = rvd->vdev_child[c];
-
-		if (cvd->vdev_ishole)
-			break;
-	}
-	return (c);
+	ASSERT3U(spa_config_held(spa, SCL_VDEV, RW_READER), ==, SCL_VDEV);
+	return (rvd->vdev_children);
 }
 
 /*
@@ -2602,7 +2590,7 @@ ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
 
 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 
-	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
+	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
 
 	/*
 	 * If we have slogs then remove them 1/4 of the time.
@@ -6527,6 +6515,15 @@ ztest_init(ztest_shared_t *zs)
 	props = make_random_props();
 	for (int i = 0; i < SPA_FEATURES; i++) {
 		char buf[1024];
+
+		/*
+		 * 75% chance of using the log space map feature. We want ztest
+		 * to exercise both the code paths that use the log space map
+		 * feature and the ones that don't.
+		 */
+		if (i == SPA_FEATURE_LOG_SPACEMAP && ztest_random(4) == 0)
+			continue;
+
 		(void) snprintf(buf, sizeof (buf), "feature@%s",
 		    spa_feature_table[i].fi_uname);
 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));

usr/src/common/zfs/zfeature_common.c

@@ -275,6 +275,17 @@ zpool_feature_init(void)
 	    "Improved clone deletion performance.",
 	    ZFEATURE_FLAG_READONLY_COMPAT, ZFEATURE_TYPE_BOOLEAN, NULL);
 
+	static const spa_feature_t log_spacemap_deps[] = {
+		SPA_FEATURE_SPACEMAP_V2,
+		SPA_FEATURE_NONE
+	};
+	zfeature_register(SPA_FEATURE_LOG_SPACEMAP,
+	    "com.delphix:log_spacemap", "log_spacemap",
+	    "Log metaslab changes on a single spacemap and "
+	    "flush them periodically.",
+	    ZFEATURE_FLAG_READONLY_COMPAT, ZFEATURE_TYPE_BOOLEAN,
+	    log_spacemap_deps);
+
 	static const spa_feature_t large_blocks_deps[] = {
 		SPA_FEATURE_EXTENSIBLE_DATASET,
 		SPA_FEATURE_NONE

usr/src/common/zfs/zfeature_common.h

@@ -65,6 +65,7 @@ typedef enum spa_feature {
 	SPA_FEATURE_POOL_CHECKPOINT,
 	SPA_FEATURE_SPACEMAP_V2,
 	SPA_FEATURE_LIVELIST,
+	SPA_FEATURE_LOG_SPACEMAP,
 	SPA_FEATURES
 } spa_feature_t;
 

usr/src/man/man1m/zdb.1m

@@ -10,7 +10,7 @@
 .\"
 .\"
 .\" Copyright 2012, Richard Lowe.
-.\" Copyright (c) 2012, 2017 by Delphix. All rights reserved.
+.\" Copyright (c) 2012, 2018 by Delphix. All rights reserved.
 .\" Copyright 2017 Nexenta Systems, Inc.
 .\"
 .Dd April 14, 2017
@@ -190,7 +190,7 @@ If the
 .Fl u
 option is also specified, also display the uberblocks on this device.
 .It Fl L
-Disable leak tracing and the loading of space maps.
+Disable leak detection and the loading of space maps.
 By default,
 .Nm
 verifies that all non-free blocks are referenced, which can be very expensive.

usr/src/test/zfs-tests/tests/functional/cli_root/zpool_import/import_rewind_config_changed.ksh

@@ -48,7 +48,7 @@ function custom_cleanup
 {
 	set_vdev_validate_skip 0
 	cleanup
-	log_must mdb_ctf_set_int vdev_min_ms_count 0t16
+	log_must mdb_ctf_set_int zfs_vdev_min_ms_count 0t16
 	log_must mdb_ctf_set_int spa_allocators 0t4
 }
 
@@ -206,7 +206,8 @@ increase_device_sizes $(( FILE_SIZE * 4 ))
 
 # Increase the number of metaslabs for small pools temporarily to
 # reduce the chance of reusing a metaslab that holds old MOS metadata.
-log_must mdb_ctf_set_int vdev_min_ms_count 0t150
+log_must mdb_ctf_set_int zfs_vdev_min_ms_count 0t150
+
 # Decrease the number of allocators for pools created during this test,
 # to increase the odds that metadata survives from old txgs.
 log_must mdb_ctf_set_int spa_allocators 0t1

usr/src/test/zfs-tests/tests/functional/cli_root/zpool_import/import_rewind_device_replaced.ksh

@@ -12,7 +12,7 @@
 #
 
 #
-# Copyright (c) 2016 by Delphix. All rights reserved.
+# Copyright (c) 2016, 2018 by Delphix. All rights reserved.
 #
 
 . $STF_SUITE/tests/functional/cli_root/zpool_import/zpool_import.kshlib
@@ -64,6 +64,8 @@ function custom_cleanup
 	    log_must set_zfs_txg_timeout $ZFS_TXG_TIMEOUT
 	log_must rm -rf $BACKUP_DEVICE_DIR
 	cleanup
+	log_must mdb_ctf_set_int zfs_vdev_min_ms_count 0t16
+	log_must mdb_ctf_set_int spa_allocators 0t4
 }
 
 log_onexit custom_cleanup
@@ -157,6 +159,14 @@ log_must mkdir $BACKUP_DEVICE_DIR
 # Make the devices bigger to reduce chances of overwriting MOS metadata.
 increase_device_sizes $(( FILE_SIZE * 4 ))
 
+# Increase the number of metaslabs for small pools temporarily to
+# reduce the chance of reusing a metaslab that holds old MOS metadata.
+log_must mdb_ctf_set_int zfs_vdev_min_ms_count 0t150
+
+# Decrease the number of allocators for pools created during this test,
+# to increase the odds that metadata survives from old txgs.
+log_must mdb_ctf_set_int spa_allocators 0t1
+
 # We set zfs_txg_timeout to 1 to reduce resilvering time at each sync.
 ZFS_TXG_TIMEOUT=$(get_zfs_txg_timeout)
 set_zfs_txg_timeout 1

usr/src/test/zfs-tests/tests/functional/pool_checkpoint/checkpoint_sm_scale.ksh

@@ -12,21 +12,21 @@
 #
 
 #
-# Copyright (c) 2017 by Delphix. All rights reserved.
+# Copyright (c) 2017, 2018 by Delphix. All rights reserved.
 #
 
 . $STF_SUITE/tests/functional/pool_checkpoint/pool_checkpoint.kshlib
 
 #
 # DESCRIPTION:
-#	The maximum address that can be described by the current space
-#	map design (assuming the minimum 512-byte addressable storage)
-#	limits the maximum allocatable space of any top-level vdev to
-#	64PB whenever a vdev-wide space map is used.
+#	The maximum address that can be described by a single-word
+#	space map entry	limits the maximum allocatable space of any
+#	top-level vdev to 64PB whenever a vdev-wide space map is used.
 #
 #	Since a vdev-wide space map is introduced for the checkpoint
-#	we want to ensure that we cannot checkpoint a pool that has a
-#	top-level vdev with more than 64PB of allocatable space.
+#	we want to ensure that we cannot checkpoint a pool that does
+#	not use the new space map encoding (V2) and has a top-level
+#	vdev with more than 64PB of allocatable space.
 #
 #	Note: Since this is a pool created from file-based vdevs we
 #	      are guaranteed that vdev_ashift  is SPA_MINBLOCKSHIFT
@@ -35,12 +35,25 @@
 #
 # STRATEGY:
 #	1. Create pool with a disk of exactly 64PB
-#	   (so ~63.5PB of allocatable space)
+#	   (so ~63.5PB of allocatable space) and
+#	   ensure that has the checkpoint feature
+#	   enabled but not space map V2.
 #	2. Ensure that you can checkpoint it
 #	3. Create pool with a disk of exactly 65PB
-#	   (so ~64.5PB of allocatable space)
+#	   (so ~64.5PB of allocatable space) with
+#	   the same setup/
 #	4. Ensure we fail trying to checkpoint it
 #
+# Note:
+# This test used to create the two pools and attempt to checkpoint
+# them at the same time, then destroy them. We later had to change
+# this to test one pool and then destroy it at a time, as the
+# metaslabs (even though empty) consumed a lot of memory, especially
+# on a machine that has been running with kmem_flags on. To give
+# an example, each metaslab structure is 1712 bytes (at the time of
+# this writing), and each vdev has 128K metaslabs, which means that
+# just the structures consume 131071 * 1712 = ~224M.
+#
 
 verify_runnable "global"
 
@@ -65,10 +78,14 @@ log_must zfs create $DISKFS
 log_must mkfile -n $((64 * 1024 * 1024))g $DISK64PB
 log_must mkfile -n $((65 * 1024 * 1024))g $DISK65PB
 
-log_must zpool create $TESTPOOL1 $DISK64PB
-log_must zpool create $TESTPOOL2 $DISK65PB
-
+log_must zpool create -d $TESTPOOL1 $DISK64PB
+log_must zpool set feature@zpool_checkpoint=enabled $TESTPOOL1
 log_must zpool checkpoint $TESTPOOL1
+destroy_pool $TESTPOOL1
+
+log_must zpool create -d $TESTPOOL2 $DISK65PB
+log_must zpool set feature@zpool_checkpoint=enabled $TESTPOOL2
 log_mustnot zpool checkpoint $TESTPOOL2
+destroy_pool $TESTPOOL2
 
 log_pass "Attempting to checkpoint a pool with a vdev that's more than 64PB."

usr/src/test/zfs-tests/tests/functional/pool_checkpoint/pool_checkpoint.kshlib

@@ -314,8 +314,19 @@ function fragment_after_checkpoint_and_verify
 	#
 	log_must zpool list -v
 
-	log_must zdb $NESTEDPOOL
-	log_must zdb -kc $NESTEDPOOL
+	#
+	# Typically we would just run zdb at this point and things
+	# would be fine. Unfortunately, if there is still any
+	# background I/O in the pool the zdb command can fail with
+	# checksum errors temporarily.
+	#
+	# Export the pool when running zdb so the pool is idle and
+	# the verification results are consistent.
+	#
+	log_must zpool export $NESTEDPOOL
+	log_must zdb -e -p $FILEDISKDIR $NESTEDPOOL
+	log_must zdb -e -p $FILEDISKDIR -kc $NESTEDPOOL
+	log_must zpool import -d $FILEDISKDIR $NESTEDPOOL
 }
 
 #

usr/src/test/zfs-tests/tests/functional/removal/removal_remap.ksh

@@ -15,7 +15,7 @@
 #
 
 #
-# Copyright (c) 2015, 2016 by Delphix. All rights reserved.
+# Copyright (c) 2015, 2018 by Delphix. All rights reserved.
 #
 
 . $STF_SUITE/include/libtest.shlib
@@ -91,13 +91,13 @@ mapping_size_after=$(indirect_vdev_mapping_size $TESTPOOL)
 
 #
 # After the remap, there should not be very many blocks referenced. The reason
-# why our threshold is as high as 512 is because our ratio of metadata to
+# why our threshold is as high as 2048 is because our ratio of metadata to
 # user data is relatively high, with only 64M of user data on the file system.
 #
 (( mapping_size_after < mapping_size_before )) || \
     log_fail "Mapping size did not decrease after remap: " \
     "$mapping_size_before before to $mapping_size_after after."
-(( mapping_size_after < 512 )) || \
+(( mapping_size_after < 2048 )) || \
     log_fail "Mapping size not small enough after remap: " \
     "$mapping_size_before before to $mapping_size_after after."
 

usr/src/uts/common/Makefile.files

@@ -21,7 +21,7 @@
 
 #
 # Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
-# Copyright (c) 2011, 2017 by Delphix. All rights reserved.
+# Copyright (c) 2011, 2018 by Delphix. All rights reserved.
 # Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
 # Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
 # Copyright 2016 Garrett D'Amore <[email protected]>
@@ -1439,6 +1439,7 @@ ZFS_COMMON_OBJS +=		\
 	spa_config.o		\
 	spa_errlog.o		\
 	spa_history.o		\
+	spa_log_spacemap.o	\
 	spa_misc.o		\
 	space_map.o		\
 	space_reftree.o		\

usr/src/uts/common/fs/zfs/arc.c

@@ -4351,7 +4351,6 @@ arc_adapt(int bytes, arc_state_t *state)
 		return;
 	}
 
-
 	if (arc_no_grow)
 		return;
 

usr/src/uts/common/fs/zfs/dbuf.c

@@ -3448,6 +3448,33 @@ dbuf_sync_indirect(dbuf_dirty_record_t *dr, dmu_tx_t *tx)
 	zio_nowait(zio);
 }
 
+/*
+ * Verify that the size of the data in our bonus buffer does not exceed
+ * its recorded size.
+ *
+ * The purpose of this verification is to catch any cases in development
+ * where the size of a phys structure (i.e space_map_phys_t) grows and,
+ * due to incorrect feature management, older pools expect to read more
+ * data even though they didn't actually write it to begin with.
+ *
+ * For a example, this would catch an error in the feature logic where we
+ * open an older pool and we expect to write the space map histogram of
+ * a space map with size SPACE_MAP_SIZE_V0.
+ */
+static void
+dbuf_sync_leaf_verify_bonus_dnode(dbuf_dirty_record_t *dr)
+{
+	arc_buf_t *datap = dr->dt.dl.dr_data;
+	uint16_t bonuslen = DB_DNODE(dr->dr_dbuf)->dn_phys->dn_bonuslen;
+
+	ASSERT3U(DN_MAX_BONUSLEN, >=, bonuslen);
+	char *datap_end = ((char *)datap) + bonuslen;
+	char *datap_max = ((char *)datap) + DN_MAX_BONUSLEN;
+
+	for (; datap_end < datap_max; datap_end++)
+		ASSERT(*datap_end == 0);
+}
+
 static void
 dbuf_sync_leaf(dbuf_dirty_record_t *dr, dmu_tx_t *tx)
 {
@@ -3498,9 +3525,14 @@ dbuf_sync_leaf(dbuf_dirty_record_t *dr, dmu_tx_t *tx)
 		ASSERT(*datap != NULL);
 		ASSERT0(db->db_level);
 		ASSERT3U(dn->dn_phys->dn_bonuslen, <=, DN_MAX_BONUSLEN);
+
 		bcopy(*datap, DN_BONUS(dn->dn_phys), dn->dn_phys->dn_bonuslen);
 		DB_DNODE_EXIT(db);
 
+#ifdef DEBUG
+		dbuf_sync_leaf_verify_bonus_dnode(dr);
+#endif
+
 		if (*datap != db->db.db_data) {
 			zio_buf_free(*datap, DN_MAX_BONUSLEN);
 			arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER);

usr/src/uts/common/fs/zfs/dmu_objset.c

@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
@@ -1296,7 +1296,7 @@ dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx)
 		ASSERT(dn->dn_dbuf->db_data_pending);
 		/*
 		 * Initialize dn_zio outside dnode_sync() because the
-		 * meta-dnode needs to set it ouside dnode_sync().
+		 * meta-dnode needs to set it outside dnode_sync().
 		 */
 		dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
 		ASSERT(dn->dn_zio);

usr/src/uts/common/fs/zfs/dnode.c

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  * Copyright (c) 2014 Integros [integros.com]
  * Copyright 2017 RackTop Systems.
@@ -362,6 +362,14 @@ dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx)
 	rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
 	ASSERT3U(newsize, <=, DN_MAX_BONUSLEN -
 	    (dn->dn_nblkptr-1) * sizeof (blkptr_t));
+
+	if (newsize < dn->dn_bonuslen) {
+		/* clear any data after the end of the new size */
+		size_t diff = dn->dn_bonuslen - newsize;
+		char *data_end = ((char *)dn->dn_bonus->db.db_data) + newsize;
+		bzero(data_end, diff);
+	}
+
 	dn->dn_bonuslen = newsize;
 	if (newsize == 0)
 		dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = DN_ZERO_BONUSLEN;

usr/src/uts/common/fs/zfs/dsl_pool.c

@@ -689,7 +689,7 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
 		dp->dp_mos_uncompressed_delta = 0;
 	}
 
-	if (!multilist_is_empty(mos->os_dirty_dnodes[txg & TXG_MASK])) {
+	if (dmu_objset_is_dirty(mos, txg)) {
 		dsl_pool_sync_mos(dp, tx);
 	}