Revert "Reduce latency effects of non-interactive I/O"

Under certain conditions commit a3a4b8def appears to result in a
hang, or poor performance, when importing a pool.  Until the root
cause can be identified it has been reverted from the release branch.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #11245

include/sys/vdev_impl.h

@@ -148,9 +148,6 @@ struct vdev_queue {
 	avl_tree_t	vq_write_offset_tree;
 	avl_tree_t	vq_trim_offset_tree;
 	uint64_t	vq_last_offset;
-	zio_priority_t	vq_last_prio;	/* Last sent I/O priority. */
-	uint32_t	vq_ia_active;	/* Active interactive I/Os. */
-	uint32_t	vq_nia_credit;	/* Non-interactive I/Os credit. */
 	hrtime_t	vq_io_complete_ts; /* time last i/o completed */
 	hrtime_t	vq_io_delta_ts;
 	zio_t		vq_io_search; /* used as local for stack reduction */

man/man5/zfs-module-parameters.5

@@ -2011,7 +2011,8 @@ Default value: \fB1\fR.
 .ad
 .RS 12n
 The maximum number of I/Os active to each device.  Ideally, this will be >=
-the sum of each queue's max_active.  See the section "ZFS I/O SCHEDULER".
+the sum of each queue's max_active.  It must be at least the sum of each
+queue's min_active.  See the section "ZFS I/O SCHEDULER".
 .sp
 Default value: \fB1,000\fR.
 .RE
@@ -2160,42 +2161,6 @@ See the section "ZFS I/O SCHEDULER".
 Default value: \fB1\fR.
 .RE
 
-.sp
-.ne 2
-.na
-\fBzfs_vdev_nia_delay\fR (int)
-.ad
-.RS 12n
-For non-interactive I/O (scrub, resilver, removal, initialize and rebuild),
-the number of concurrently-active I/O's is limited to *_min_active, unless
-the vdev is "idle".  When there are no interactive I/Os active (sync or
-async), and zfs_vdev_nia_delay I/Os have completed since the last
-interactive I/O, then the vdev is considered to be "idle", and the number
-of concurrently-active non-interactive I/O's is increased to *_max_active.
-See the section "ZFS I/O SCHEDULER".
-.sp
-Default value: \fB5\fR.
-.RE
-
-.sp
-.ne 2
-.na
-\fBzfs_vdev_nia_credit\fR (int)
-.ad
-.RS 12n
-Some HDDs tend to prioritize sequential I/O so high, that concurrent
-random I/O latency reaches several seconds.  On some HDDs it happens
-even if sequential I/Os are submitted one at a time, and so setting
-*_max_active to 1 does not help.  To prevent non-interactive I/Os, like
-scrub, from monopolizing the device no more than zfs_vdev_nia_credit
-I/Os can be sent while there are outstanding incomplete interactive
-I/Os.  This enforced wait ensures the HDD services the interactive I/O
-within a reasonable amount of time.
-See the section "ZFS I/O SCHEDULER".
-.sp
-Default value: \fB5\fR.
-.RE
-
 .sp
 .ne 2
 .na

module/zfs/vdev_queue.c

@@ -121,17 +121,16 @@
 
 /*
  * The maximum number of i/os active to each device.  Ideally, this will be >=
- * the sum of each queue's max_active.
+ * the sum of each queue's max_active.  It must be at least the sum of each
+ * queue's min_active.
  */
 uint32_t zfs_vdev_max_active = 1000;
 
 /*
  * Per-queue limits on the number of i/os active to each device.  If the
  * number of active i/os is < zfs_vdev_max_active, then the min_active comes
- * into play.  We will send min_active from each queue round-robin, and then
+ * into play. We will send min_active from each queue, and then select from
- * send from queues in the order defined by zio_priority_t up to max_active.
+ * queues in the order defined by zio_priority_t.
- * Some queues have additional mechanisms to limit number of active I/Os in
- * addition to min_active and max_active, see below.
  *
  * In general, smaller max_active's will lead to lower latency of synchronous
  * operations.  Larger max_active's may lead to higher overall throughput,
@@ -152,7 +151,7 @@ uint32_t zfs_vdev_async_read_max_active = 3;
 uint32_t zfs_vdev_async_write_min_active = 2;
 uint32_t zfs_vdev_async_write_max_active = 10;
 uint32_t zfs_vdev_scrub_min_active = 1;
-uint32_t zfs_vdev_scrub_max_active = 3;
+uint32_t zfs_vdev_scrub_max_active = 2;
 uint32_t zfs_vdev_removal_min_active = 1;
 uint32_t zfs_vdev_removal_max_active = 2;
 uint32_t zfs_vdev_initializing_min_active = 1;
@@ -172,28 +171,6 @@ uint32_t zfs_vdev_rebuild_max_active = 3;
 int zfs_vdev_async_write_active_min_dirty_percent = 30;
 int zfs_vdev_async_write_active_max_dirty_percent = 60;
 
-/*
- * For non-interactive I/O (scrub, resilver, removal, initialize and rebuild),
- * the number of concurrently-active I/O's is limited to *_min_active, unless
- * the vdev is "idle".  When there are no interactive I/Os active (sync or
- * async), and zfs_vdev_nia_delay I/Os have completed since the last
- * interactive I/O, then the vdev is considered to be "idle", and the number
- * of concurrently-active non-interactive I/O's is increased to *_max_active.
- */
-uint_t zfs_vdev_nia_delay = 5;
-
-/*
- * Some HDDs tend to prioritize sequential I/O so high that concurrent
- * random I/O latency reaches several seconds.  On some HDDs it happens
- * even if sequential I/Os are submitted one at a time, and so setting
- * *_max_active to 1 does not help.  To prevent non-interactive I/Os, like
- * scrub, from monopolizing the device no more than zfs_vdev_nia_credit
- * I/Os can be sent while there are outstanding incomplete interactive
- * I/Os.  This enforced wait ensures the HDD services the interactive I/O
- * within a reasonable amount of time.
- */
-uint_t zfs_vdev_nia_credit = 5;
-
 /*
  * To reduce IOPs, we aggregate small adjacent I/Os into one large I/O.
  * For read I/Os, we also aggregate across small adjacency gaps; for writes
@@ -284,7 +261,7 @@ vdev_queue_timestamp_compare(const void *x1, const void *x2)
 }
 
 static int
-vdev_queue_class_min_active(vdev_queue_t *vq, zio_priority_t p)
+vdev_queue_class_min_active(zio_priority_t p)
 {
 	switch (p) {
 	case ZIO_PRIORITY_SYNC_READ:
@@ -296,19 +273,15 @@ vdev_queue_class_min_active(vdev_queue_t *vq, zio_priority_t p)
 	case ZIO_PRIORITY_ASYNC_WRITE:
 		return (zfs_vdev_async_write_min_active);
 	case ZIO_PRIORITY_SCRUB:
-		return (vq->vq_ia_active == 0 ? zfs_vdev_scrub_min_active :
+		return (zfs_vdev_scrub_min_active);
-		    MIN(vq->vq_nia_credit, zfs_vdev_scrub_min_active));
 	case ZIO_PRIORITY_REMOVAL:
-		return (vq->vq_ia_active == 0 ? zfs_vdev_removal_min_active :
+		return (zfs_vdev_removal_min_active);
-		    MIN(vq->vq_nia_credit, zfs_vdev_removal_min_active));
 	case ZIO_PRIORITY_INITIALIZING:
-		return (vq->vq_ia_active == 0 ?zfs_vdev_initializing_min_active:
+		return (zfs_vdev_initializing_min_active);
-		    MIN(vq->vq_nia_credit, zfs_vdev_initializing_min_active));
 	case ZIO_PRIORITY_TRIM:
 		return (zfs_vdev_trim_min_active);
 	case ZIO_PRIORITY_REBUILD:
-		return (vq->vq_ia_active == 0 ? zfs_vdev_rebuild_min_active :
+		return (zfs_vdev_rebuild_min_active);
-		    MIN(vq->vq_nia_credit, zfs_vdev_rebuild_min_active));
 	default:
 		panic("invalid priority %u", p);
 		return (0);
@@ -364,7 +337,7 @@ vdev_queue_max_async_writes(spa_t *spa)
 }
 
 static int
-vdev_queue_class_max_active(spa_t *spa, vdev_queue_t *vq, zio_priority_t p)
+vdev_queue_class_max_active(spa_t *spa, zio_priority_t p)
 {
 	switch (p) {
 	case ZIO_PRIORITY_SYNC_READ:
@@ -376,34 +349,14 @@ vdev_queue_class_max_active(spa_t *spa, vdev_queue_t *vq, zio_priority_t p)
 	case ZIO_PRIORITY_ASYNC_WRITE:
 		return (vdev_queue_max_async_writes(spa));
 	case ZIO_PRIORITY_SCRUB:
-		if (vq->vq_ia_active > 0) {
-			return (MIN(vq->vq_nia_credit,
-			    zfs_vdev_scrub_min_active));
-		} else if (vq->vq_nia_credit < zfs_vdev_nia_delay)
-			return (zfs_vdev_scrub_min_active);
 		return (zfs_vdev_scrub_max_active);
 	case ZIO_PRIORITY_REMOVAL:
-		if (vq->vq_ia_active > 0) {
-			return (MIN(vq->vq_nia_credit,
-			    zfs_vdev_removal_min_active));
-		} else if (vq->vq_nia_credit < zfs_vdev_nia_delay)
-			return (zfs_vdev_removal_min_active);
 		return (zfs_vdev_removal_max_active);
 	case ZIO_PRIORITY_INITIALIZING:
-		if (vq->vq_ia_active > 0) {
-			return (MIN(vq->vq_nia_credit,
-			    zfs_vdev_initializing_min_active));
-		} else if (vq->vq_nia_credit < zfs_vdev_nia_delay)
-			return (zfs_vdev_initializing_min_active);
 		return (zfs_vdev_initializing_max_active);
 	case ZIO_PRIORITY_TRIM:
 		return (zfs_vdev_trim_max_active);
 	case ZIO_PRIORITY_REBUILD:
-		if (vq->vq_ia_active > 0) {
-			return (MIN(vq->vq_nia_credit,
-			    zfs_vdev_rebuild_min_active));
-		} else if (vq->vq_nia_credit < zfs_vdev_nia_delay)
-			return (zfs_vdev_rebuild_min_active);
 		return (zfs_vdev_rebuild_max_active);
 	default:
 		panic("invalid priority %u", p);
@@ -419,24 +372,17 @@ static zio_priority_t
 vdev_queue_class_to_issue(vdev_queue_t *vq)
 {
 	spa_t *spa = vq->vq_vdev->vdev_spa;
-	zio_priority_t p, n;
+	zio_priority_t p;
 
 	if (avl_numnodes(&vq->vq_active_tree) >= zfs_vdev_max_active)
 		return (ZIO_PRIORITY_NUM_QUEUEABLE);
 
-	/*
+	/* find a queue that has not reached its minimum # outstanding i/os */
-	 * Find a queue that has not reached its minimum # outstanding i/os.
+	for (p = 0; p < ZIO_PRIORITY_NUM_QUEUEABLE; p++) {
-	 * Do round-robin to reduce starvation due to zfs_vdev_max_active
-	 * and vq_nia_credit limits.
-	 */
-	for (n = 0; n < ZIO_PRIORITY_NUM_QUEUEABLE; n++) {
-		p = (vq->vq_last_prio + n + 1) % ZIO_PRIORITY_NUM_QUEUEABLE;
 		if (avl_numnodes(vdev_queue_class_tree(vq, p)) > 0 &&
 		    vq->vq_class[p].vqc_active <
-		    vdev_queue_class_min_active(vq, p)) {
+		    vdev_queue_class_min_active(p))
-			vq->vq_last_prio = p;
 			return (p);
-		}
 	}
 
 	/*
@@ -446,10 +392,8 @@ vdev_queue_class_to_issue(vdev_queue_t *vq)
 	for (p = 0; p < ZIO_PRIORITY_NUM_QUEUEABLE; p++) {
 		if (avl_numnodes(vdev_queue_class_tree(vq, p)) > 0 &&
 		    vq->vq_class[p].vqc_active <
-		    vdev_queue_class_max_active(spa, vq, p)) {
+		    vdev_queue_class_max_active(spa, p))
-			vq->vq_last_prio = p;
 			return (p);
-		}
 	}
 
 	/* No eligible queued i/os */
@@ -549,20 +493,6 @@ vdev_queue_io_remove(vdev_queue_t *vq, zio_t *zio)
 	}
 }
 
-static boolean_t
-vdev_queue_is_interactive(zio_priority_t p)
-{
-	switch (p) {
-	case ZIO_PRIORITY_SCRUB:
-	case ZIO_PRIORITY_REMOVAL:
-	case ZIO_PRIORITY_INITIALIZING:
-	case ZIO_PRIORITY_REBUILD:
-		return (B_FALSE);
-	default:
-		return (B_TRUE);
-	}
-}
-
 static void
 vdev_queue_pending_add(vdev_queue_t *vq, zio_t *zio)
 {
@@ -572,12 +502,6 @@ vdev_queue_pending_add(vdev_queue_t *vq, zio_t *zio)
 	ASSERT(MUTEX_HELD(&vq->vq_lock));
 	ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE);
 	vq->vq_class[zio->io_priority].vqc_active++;
-	if (vdev_queue_is_interactive(zio->io_priority)) {
-		if (++vq->vq_ia_active == 1)
-			vq->vq_nia_credit = 1;
-	} else if (vq->vq_ia_active > 0) {
-		vq->vq_nia_credit--;
-	}
 	avl_add(&vq->vq_active_tree, zio);
 
 	if (shk->kstat != NULL) {
@@ -596,13 +520,6 @@ vdev_queue_pending_remove(vdev_queue_t *vq, zio_t *zio)
 	ASSERT(MUTEX_HELD(&vq->vq_lock));
 	ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE);
 	vq->vq_class[zio->io_priority].vqc_active--;
-	if (vdev_queue_is_interactive(zio->io_priority)) {
-		if (--vq->vq_ia_active == 0)
-			vq->vq_nia_credit = 0;
-		else
-			vq->vq_nia_credit = zfs_vdev_nia_credit;
-	} else if (vq->vq_ia_active == 0)
-		vq->vq_nia_credit++;
 	avl_remove(&vq->vq_active_tree, zio);
 
 	if (shk->kstat != NULL) {
@@ -1148,12 +1065,6 @@ ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, rebuild_max_active, INT, ZMOD_RW,
 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, rebuild_min_active, INT, ZMOD_RW,
 	"Min active rebuild I/Os per vdev");
 
-ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, nia_credit, INT, ZMOD_RW,
-	"Number of non-interactive I/Os to allow in sequence");
-
-ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, nia_delay, INT, ZMOD_RW,
-	"Number of non-interactive I/Os before _max_active");
-
 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, queue_depth_pct, INT, ZMOD_RW,
 	"Queue depth percentage for each top-level vdev");
 /* END CSTYLED */