Remove limit on number of async zio_frees of non-dedup blocks

The module parameter zfs_async_block_max_blocks limits the number of
blocks that can be freed by the background freeing of filesystems and
snapshots (from "zfs destroy"), in one TXG.  This is useful when freeing
dedup blocks, becuase each zio_free() of a dedup block can require an
i/o to read the relevant part of the dedup table (DDT), and will also
dirty that block.

zfs_async_block_max_blocks is set to 100,000 by default.  For the more
typical case where dedup is not used, this can have a negative
performance impact on the rate of background freeing (from "zfs
destroy").  For example, with recordsize=8k, and TXG's syncing once
every 5 seconds, we can free only 160MB of data per second, which may be
much less than the rate we can write data.

This change increases zfs_async_block_max_blocks to be unlimited by
default.  To address the dedup freeing issue, a new tunable is
introduced, zfs_max_async_dedup_frees, which limits the number of
zio_free()'s of dedup blocks done by background destroys, per txg.  The
default is 100,000 free's (same as the old zfs_async_block_max_blocks
default).

Reviewed-by: Paul Dagnelie <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Matthew Ahrens <[email protected]>
Closes #10000 

1 files changed, 15 insertions, 1 deletions

diff --git a/module/zfs/dsl_scan.c b/module/zfs/dsl_scan.c
index 1c31261bf..3579f71d7 100644
--- a/module/zfs/dsl_scan.c
+++ b/module/zfs/dsl_scan.c
@@ -175,7 +175,9 @@ int zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */
 int zfs_no_scrub_prefetch = B_FALSE; /* set to disable scrub prefetch */
 enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE;
 /* max number of blocks to free in a single TXG */
-unsigned long zfs_async_block_max_blocks = 100000;
+unsigned long zfs_async_block_max_blocks = ULONG_MAX;
+/* max number of dedup blocks to free in a single TXG */
+unsigned long zfs_max_async_dedup_frees = 100000;
 
 int zfs_resilver_disable_defer = 0; /* set to disable resilver deferring */
 
@@ -3101,6 +3103,11 @@ dsl_scan_async_block_should_pause(dsl_scan_t *scn)
 		return (B_TRUE);
 	}
 
+	if (zfs_max_async_dedup_frees != 0 &&
+	    scn->scn_dedup_frees_this_txg >= zfs_max_async_dedup_frees) {
+		return (B_TRUE);
+	}
+
 	elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
 	return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
 	    (NSEC2MSEC(elapsed_nanosecs) > scn->scn_async_block_min_time_ms &&
@@ -3125,6 +3132,8 @@ dsl_scan_free_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
 	    -bp_get_dsize_sync(scn->scn_dp->dp_spa, bp),
 	    -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx);
 	scn->scn_visited_this_txg++;
+	if (BP_GET_DEDUP(bp))
+		scn->scn_dedup_frees_this_txg++;
 	return (0);
 }
 
@@ -3365,6 +3374,7 @@ dsl_process_async_destroys(dsl_pool_t *dp, dmu_tx_t *tx)
 		    NSEC2MSEC(gethrtime() - scn->scn_sync_start_time),
 		    (longlong_t)tx->tx_txg, err);
 		scn->scn_visited_this_txg = 0;
+		scn->scn_dedup_frees_this_txg = 0;
 
 		/*
 		 * Write out changes to the DDT that may be required as a
@@ -3493,6 +3503,7 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx)
 
 	/* reset scan statistics */
 	scn->scn_visited_this_txg = 0;
+	scn->scn_dedup_frees_this_txg = 0;
 	scn->scn_holes_this_txg = 0;
 	scn->scn_lt_min_this_txg = 0;
 	scn->scn_gt_max_this_txg = 0;
@@ -4327,6 +4338,9 @@ ZFS_MODULE_PARAM(zfs, zfs_, no_scrub_prefetch, INT, ZMOD_RW,
 ZFS_MODULE_PARAM(zfs, zfs_, async_block_max_blocks, ULONG, ZMOD_RW,
 	"Max number of blocks freed in one txg");
 
+ZFS_MODULE_PARAM(zfs, zfs_, max_async_dedup_frees, ULONG, ZMOD_RW,
+	"Max number of dedup blocks freed in one txg");
+
 ZFS_MODULE_PARAM(zfs, zfs_, free_bpobj_enabled, INT, ZMOD_RW,
 	"Enable processing of the free_bpobj");