MFV r274271:

Improve zdb -b performance:

 - Reduce gethrtime() call to 1/100th of blkptr's;
 - Skip manipulating the size-ordered tree;
 - Issue more (10, previously 3) async reads;
 - Use lighter weight testing in traverse_visitbp();

Illumos issue:
    5243 zdb -b could be much faster

MFC after:	2 weeks

diff --git a/cddl/contrib/opensolaris/cmd/zdb/zdb.c b/cddl/contrib/opensolaris/cmd/zdb/zdb.c
index 1fce5b658d64a823598c9e0da088e40c0c77a0c5..c2c47d511c03b2bfa50560bb56810c6378518b8a 100644 (file)
--- a/cddl/contrib/opensolaris/cmd/zdb/zdb.c
+++ b/cddl/contrib/opensolaris/cmd/zdb/zdb.c
@@ -77,9 +77,11 @@
 #ifndef lint
 extern boolean_t zfs_recover;
 extern uint64_t zfs_arc_max, zfs_arc_meta_limit;
+extern int zfs_vdev_async_read_max_active;
 #else
 boolean_t zfs_recover;
 uint64_t zfs_arc_max, zfs_arc_meta_limit;
+int zfs_vdev_async_read_max_active;
 #endif
 
 const char cmdname[] = "zdb";
@@ -2384,8 +2386,14 @@ zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 
        zcb->zcb_readfails = 0;
 
-       if (dump_opt['b'] < 5 &&
-           gethrtime() > zcb->zcb_lastprint + NANOSEC) {
+       /* only call gethrtime() every 100 blocks */
+       static int iters;
+       if (++iters > 100)
+               iters = 0;
+       else
+               return (0);
+
+       if (dump_opt['b'] < 5 && gethrtime() > zcb->zcb_lastprint + NANOSEC) {
                uint64_t now = gethrtime();
                char buf[10];
                uint64_t bytes = zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL].zb_asize;
@@ -2494,6 +2502,14 @@ zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
                                            (longlong_t)vd->vdev_ms_count);
 
                                        msp->ms_ops = &zdb_metaslab_ops;
+
+                                       /*
+                                        * We don't want to spend the CPU
+                                        * manipulating the size-ordered
+                                        * tree, so clear the range_tree
+                                        * ops.
+                                        */
+                                       msp->ms_tree->rt_ops = NULL;
                                        VERIFY0(space_map_load(msp->ms_sm,
                                            msp->ms_tree, SM_ALLOC));
                                        msp->ms_loaded = B_TRUE;
@@ -3508,6 +3524,13 @@ main(int argc, char **argv)
         */
        zfs_arc_max = zfs_arc_meta_limit = 256 * 1024 * 1024;
 
+       /*
+        * "zdb -c" uses checksum-verifying scrub i/os which are async reads.
+        * "zdb -b" uses traversal prefetch which uses async reads.
+        * For good performance, let several of them be active at once.
+        */
+       zfs_vdev_async_read_max_active = 10;
+
        kernel_init(FREAD);
        g_zfs = libzfs_init();
        ASSERT(g_zfs != NULL);

diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_traverse.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_traverse.c
index fb1ec0755cf179b9e7dfc465d9f62a59a76fb408..5836549d200ac844e7dd4daca8f7843ef243959c 100644 (file)
--- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_traverse.c
+++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_traverse.c
@@ -59,6 +59,7 @@ typedef struct traverse_data {
        int td_flags;
        prefetch_data_t *td_pfd;
        boolean_t td_paused;
+       uint64_t td_hole_birth_enabled_txg;
        blkptr_cb_t *td_func;
        void *td_arg;
 } traverse_data_t;
@@ -229,25 +230,20 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
        }
 
        if (bp->blk_birth == 0) {
-               if (spa_feature_is_active(td->td_spa, SPA_FEATURE_HOLE_BIRTH)) {
-                       /*
-                        * Since this block has a birth time of 0 it must be a
-                        * hole created before the SPA_FEATURE_HOLE_BIRTH
-                        * feature was enabled.  If SPA_FEATURE_HOLE_BIRTH
-                        * was enabled before the min_txg for this traveral we
-                        * know the hole must have been created before the
-                        * min_txg for this traveral, so we can skip it. If
-                        * SPA_FEATURE_HOLE_BIRTH was enabled after the min_txg
-                        * for this traveral we cannot tell if the hole was
-                        * created before or after the min_txg for this
-                        * traversal, so we cannot skip it.
-                        */
-                       uint64_t hole_birth_enabled_txg;
-                       VERIFY(spa_feature_enabled_txg(td->td_spa,
-                           SPA_FEATURE_HOLE_BIRTH, &hole_birth_enabled_txg));
-                       if (hole_birth_enabled_txg < td->td_min_txg)
-                               return (0);
-               }
+               /*
+                * Since this block has a birth time of 0 it must be a
+                * hole created before the SPA_FEATURE_HOLE_BIRTH
+                * feature was enabled.  If SPA_FEATURE_HOLE_BIRTH
+                * was enabled before the min_txg for this traveral we
+                * know the hole must have been created before the
+                * min_txg for this traveral, so we can skip it. If
+                * SPA_FEATURE_HOLE_BIRTH was enabled after the min_txg
+                * for this traveral we cannot tell if the hole was
+                * created before or after the min_txg for this
+                * traversal, so we cannot skip it.
+                */
+               if (td->td_hole_birth_enabled_txg < td->td_min_txg)
+                       return (0);
        } else if (bp->blk_birth <= td->td_min_txg) {
                return (0);
        }
@@ -523,6 +519,13 @@ traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp,
        td.td_flags = flags;
        td.td_paused = B_FALSE;
 
+       if (spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) {
+               VERIFY(spa_feature_enabled_txg(spa,
+                   SPA_FEATURE_HOLE_BIRTH, &td.td_hole_birth_enabled_txg));
+       } else {
+               td.td_hole_birth_enabled_txg = 0;
+       }
+
        pd.pd_blks_max = zfs_pd_blks_max;
        pd.pd_flags = flags;
        mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL);