OpenZFS 7303 - dynamic metaslab selection

This change introduces a new weighting algorithm to improve
metaslab selection. The new weighting algorithm relies on the
SPACEMAP_HISTOGRAM feature. As a result, the metaslab weight
now encodes the type of weighting algorithm used (size-based
vs segment-based).

Porting Notes: The metaslab allocation tracing code is conditionally
removed on linux (dependent on mdb debugger).

Authored by: George Wilson <[email protected]>
Reviewed by: Alex Reece <[email protected]>
Reviewed by: Chris Siden <[email protected]>
Reviewed by: Dan Kimmel <[email protected]>
Reviewed by: Matthew Ahrens <[email protected]>
Reviewed by: Paul Dagnelie <[email protected]>
Reviewed by: Pavel Zakharov [email protected]
Reviewed by: Prakash Surya <[email protected]>
Reviewed by: Don Brady <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Ported-by: Don Brady <[email protected]>

OpenZFS-issue: https://www.illumos.org/issues/7303
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/d5190931bd
Closes #5404 

5 files changed, 131 insertions, 14 deletions

diff --git a/include/sys/metaslab.h b/include/sys/metaslab.h
index 408f6d333..be271c702 100644
--- a/include/sys/metaslab.h
+++ b/include/sys/metaslab.h
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
  */
 
 #ifndef _SYS_METASLAB_H
@@ -36,10 +36,12 @@
 extern "C" {
 #endif
 
+
 typedef struct metaslab_ops {
-	uint64_t (*msop_alloc)(metaslab_t *msp, uint64_t size);
+	uint64_t (*msop_alloc)(metaslab_t *, uint64_t);
 } metaslab_ops_t;
 
+
 extern metaslab_ops_t *zfs_metaslab_ops;
 
 int metaslab_init(metaslab_group_t *, uint64_t, uint64_t, uint64_t,
@@ -64,13 +66,18 @@ uint64_t metaslab_block_maxsize(metaslab_t *);
 #define	METASLAB_FASTWRITE		0x20
 
 int metaslab_alloc(spa_t *, metaslab_class_t *, uint64_t,
-    blkptr_t *, int, uint64_t, blkptr_t *, int, zio_t *);
+    blkptr_t *, int, uint64_t, blkptr_t *, int, zio_alloc_list_t *, zio_t *);
 void metaslab_free(spa_t *, const blkptr_t *, uint64_t, boolean_t);
 int metaslab_claim(spa_t *, const blkptr_t *, uint64_t);
 void metaslab_check_free(spa_t *, const blkptr_t *);
 void metaslab_fastwrite_mark(spa_t *, const blkptr_t *);
 void metaslab_fastwrite_unmark(spa_t *, const blkptr_t *);
 
+void metaslab_alloc_trace_init(void);
+void metaslab_alloc_trace_fini(void);
+void metaslab_trace_init(zio_alloc_list_t *);
+void metaslab_trace_fini(zio_alloc_list_t *);
+
 metaslab_class_t *metaslab_class_create(spa_t *, metaslab_ops_t *);
 void metaslab_class_destroy(metaslab_class_t *);
 int metaslab_class_validate(metaslab_class_t *);
diff --git a/include/sys/metaslab_impl.h b/include/sys/metaslab_impl.h
index 080a1db9a..7b68b51f6 100644
--- a/include/sys/metaslab_impl.h
+++ b/include/sys/metaslab_impl.h
@@ -24,7 +24,7 @@
  */
 
 /*
- * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
  */
 
 #ifndef _SYS_METASLAB_IMPL_H
@@ -42,6 +42,94 @@ extern "C" {
 #endif
 
 /*
+ * Metaslab allocation tracing record.
+ */
+typedef struct metaslab_alloc_trace {
+	list_node_t			mat_list_node;
+	metaslab_group_t		*mat_mg;
+	metaslab_t			*mat_msp;
+	uint64_t			mat_size;
+	uint64_t			mat_weight;
+	uint32_t			mat_dva_id;
+	uint64_t			mat_offset;
+} metaslab_alloc_trace_t;
+
+/*
+ * Used by the metaslab allocation tracing facility to indicate
+ * error conditions. These errors are stored to the offset member
+ * of the metaslab_alloc_trace_t record and displayed by mdb.
+ */
+typedef enum trace_alloc_type {
+	TRACE_ALLOC_FAILURE	= -1ULL,
+	TRACE_TOO_SMALL		= -2ULL,
+	TRACE_FORCE_GANG	= -3ULL,
+	TRACE_NOT_ALLOCATABLE	= -4ULL,
+	TRACE_GROUP_FAILURE	= -5ULL,
+	TRACE_ENOSPC		= -6ULL,
+	TRACE_CONDENSING	= -7ULL,
+	TRACE_VDEV_ERROR	= -8ULL
+} trace_alloc_type_t;
+
+#define	METASLAB_WEIGHT_PRIMARY		(1ULL << 63)
+#define	METASLAB_WEIGHT_SECONDARY	(1ULL << 62)
+#define	METASLAB_WEIGHT_TYPE		(1ULL << 61)
+#define	METASLAB_ACTIVE_MASK		\
+	(METASLAB_WEIGHT_PRIMARY | METASLAB_WEIGHT_SECONDARY)
+
+/*
+ * The metaslab weight is used to encode the amount of free space in a
+ * metaslab, such that the "best" metaslab appears first when sorting the
+ * metaslabs by weight. The weight (and therefore the "best" metaslab) can
+ * be determined in two different ways: by computing a weighted sum of all
+ * the free space in the metaslab (a space based weight) or by counting only
+ * the free segments of the largest size (a segment based weight). We prefer
+ * the segment based weight because it reflects how the free space is
+ * comprised, but we cannot always use it -- legacy pools do not have the
+ * space map histogram information necessary to determine the largest
+ * contiguous regions. Pools that have the space map histogram determine
+ * the segment weight by looking at each bucket in the histogram and
+ * determining the free space whose size in bytes is in the range:
+ *	[2^i, 2^(i+1))
+ * We then encode the largest index, i, that contains regions into the
+ * segment-weighted value.
+ *
+ * Space-based weight:
+ *
+ *      64      56      48      40      32      24      16      8       0
+ *      +-------+-------+-------+-------+-------+-------+-------+-------+
+ *      |PS1|                   weighted-free space                     |
+ *      +-------+-------+-------+-------+-------+-------+-------+-------+
+ *
+ *	PS - indicates primary and secondary activation
+ *	space - the fragmentation-weighted space
+ *
+ * Segment-based weight:
+ *
+ *      64      56      48      40      32      24      16      8       0
+ *      +-------+-------+-------+-------+-------+-------+-------+-------+
+ *      |PS0| idx|             count of segments in region              |
+ *      +-------+-------+-------+-------+-------+-------+-------+-------+
+ *
+ *	PS - indicates primary and secondary activation
+ *	idx - index for the highest bucket in the histogram
+ *	count - number of segments in the specified bucket
+ */
+#define	WEIGHT_GET_ACTIVE(weight)		BF64_GET((weight), 62, 2)
+#define	WEIGHT_SET_ACTIVE(weight, x)		BF64_SET((weight), 62, 2, x)
+
+#define	WEIGHT_IS_SPACEBASED(weight)		\
+	((weight) == 0 || BF64_GET((weight), 61, 1))
+#define	WEIGHT_SET_SPACEBASED(weight)		BF64_SET((weight), 61, 1, 1)
+
+/*
+ * These macros are only applicable to segment-based weighting.
+ */
+#define	WEIGHT_GET_INDEX(weight)		BF64_GET((weight), 55, 6)
+#define	WEIGHT_SET_INDEX(weight, x)		BF64_SET((weight), 55, 6, x)
+#define	WEIGHT_GET_COUNT(weight)		BF64_GET((weight), 0, 55)
+#define	WEIGHT_SET_COUNT(weight, x)		BF64_SET((weight), 0, 55, x)
+
+/*
  * A metaslab class encompasses a category of allocatable top-level vdevs.
  * Each top-level vdev is associated with a metaslab group which defines
  * the allocatable region for that vdev. Examples of these categories include
@@ -220,7 +308,6 @@ struct metaslab {
 	kmutex_t	ms_lock;
 	kcondvar_t	ms_load_cv;
 	space_map_t	*ms_sm;
-	metaslab_ops_t	*ms_ops;
 	uint64_t	ms_id;
 	uint64_t	ms_start;
 	uint64_t	ms_size;
@@ -233,12 +320,27 @@ struct metaslab {
 
 	boolean_t	ms_condensing;	/* condensing? */
 	boolean_t	ms_condense_wanted;
+
+	/*
+	 * We must hold both ms_lock and ms_group->mg_lock in order to
+	 * modify ms_loaded.
+	 */
 	boolean_t	ms_loaded;
 	boolean_t	ms_loading;
 
 	int64_t		ms_deferspace;	/* sum of ms_defermap[] space	*/
 	uint64_t	ms_weight;	/* weight vs. others in group	*/
-	uint64_t	ms_access_txg;
+	uint64_t	ms_activation_weight;	/* activation weight	*/
+
+	/*
+	 * Track of whenever a metaslab is selected for loading or allocation.
+	 * We use this value to determine how long the metaslab should
+	 * stay cached.
+	 */
+	uint64_t	ms_selected_txg;
+
+	uint64_t	ms_alloc_txg;	/* last successful alloc (debug only) */
+	uint64_t	ms_max_size;	/* maximum allocatable size	*/
 
 	/*
 	 * The metaslab block allocators can optionally use a size-ordered
diff --git a/include/sys/zfs_context.h b/include/sys/zfs_context.h
index a1abe20da..6e950716c 100644
--- a/include/sys/zfs_context.h
+++ b/include/sys/zfs_context.h
@@ -394,6 +394,7 @@ extern void cv_broadcast(kcondvar_t *cv);
  */
 extern kstat_t *kstat_create(const char *, int,
     const char *, const char *, uchar_t, ulong_t, uchar_t);
+extern void kstat_named_init(kstat_named_t *, const char *, uchar_t);
 extern void kstat_install(kstat_t *);
 extern void kstat_delete(kstat_t *);
 extern void kstat_waitq_enter(kstat_io_t *);
diff --git a/include/sys/zfs_debug.h b/include/sys/zfs_debug.h
index 2f0064ee0..a2f61e5b2 100644
--- a/include/sys/zfs_debug.h
+++ b/include/sys/zfs_debug.h
@@ -42,14 +42,15 @@ extern int zfs_flags;
 extern int zfs_recover;
 extern int zfs_free_leak_on_eio;
 
-#define	ZFS_DEBUG_DPRINTF		(1<<0)
-#define	ZFS_DEBUG_DBUF_VERIFY		(1<<1)
-#define	ZFS_DEBUG_DNODE_VERIFY		(1<<2)
-#define	ZFS_DEBUG_SNAPNAMES		(1<<3)
-#define	ZFS_DEBUG_MODIFY		(1<<4)
-#define	ZFS_DEBUG_SPA			(1<<5)
-#define	ZFS_DEBUG_ZIO_FREE		(1<<6)
-#define	ZFS_DEBUG_HISTOGRAM_VERIFY	(1<<7)
+#define	ZFS_DEBUG_DPRINTF		(1 << 0)
+#define	ZFS_DEBUG_DBUF_VERIFY		(1 << 1)
+#define	ZFS_DEBUG_DNODE_VERIFY		(1 << 2)
+#define	ZFS_DEBUG_SNAPNAMES		(1 << 3)
+#define	ZFS_DEBUG_MODIFY		(1 << 4)
+#define	ZFS_DEBUG_SPA			(1 << 5)
+#define	ZFS_DEBUG_ZIO_FREE		(1 << 6)
+#define	ZFS_DEBUG_HISTOGRAM_VERIFY	(1 << 7)
+#define	ZFS_DEBUG_METASLAB_VERIFY	(1 << 8)
 
 extern void __dprintf(const char *file, const char *func,
     int line, const char *fmt, ...);
diff --git a/include/sys/zio.h b/include/sys/zio.h
index 6c5153dcf..1f8708e5a 100644
--- a/include/sys/zio.h
+++ b/include/sys/zio.h
@@ -358,6 +358,11 @@ typedef int zio_pipe_stage_t(zio_t *zio);
 #define	ZIO_REEXECUTE_NOW	0x01
 #define	ZIO_REEXECUTE_SUSPEND	0x02
 
+typedef struct zio_alloc_list {
+	list_t  zal_list;
+	uint64_t zal_size;
+} zio_alloc_list_t;
+
 typedef struct zio_link {
 	zio_t		*zl_parent;
 	zio_t		*zl_child;
@@ -417,6 +422,7 @@ struct zio {
 	avl_node_t	io_queue_node;
 	avl_node_t	io_offset_node;
 	avl_node_t	io_alloc_node;
+	zio_alloc_list_t 	io_alloc_list;
 
 	/* Internal pipeline state */
 	enum zio_flag	io_flags;