OpenZFS 7004 - dmu_tx_hold_zap() does dnode_hold() 7x on same object

Using a benchmark which has 32 threads creating 2 million files in the
same directory, on a machine with 16 CPU cores, I observed poor
performance. I noticed that dmu_tx_hold_zap() was using about 30% of
all CPU, and doing dnode_hold() 7 times on the same object (the ZAP
object that is being held).

dmu_tx_hold_zap() keeps a hold on the dnode_t the entire time it is
running, in dmu_tx_hold_t:txh_dnode, so it would be nice to use the
dnode_t that we already have in hand, rather than repeatedly calling
dnode_hold(). To do this, we need to pass the dnode_t down through
all the intermediate calls that dmu_tx_hold_zap() makes, making these
routines take the dnode_t* rather than an objset_t* and a uint64_t
object number. In particular, the following routines will need to have
analogous *_by_dnode() variants created:

dmu_buf_hold_noread()
dmu_buf_hold()
zap_lookup()
zap_lookup_norm()
zap_count_write()
zap_lockdir()
zap_count_write()

This can improve performance on the benchmark described above by 100%,
from 30,000 file creations per second to 60,000. (This improvement is on
top of that provided by working around the object allocation issue. Peak
performance of ~90,000 creations per second was observed with 8 CPUs;
adding CPUs past that decreased performance due to lock contention.) The
CPU used by dmu_tx_hold_zap() was reduced by 88%, from 340 CPU-seconds
to 40 CPU-seconds.

Sponsored by: Intel Corp.

Signed-off-by: Matthew Ahrens <[email protected]>
Signed-off-by: Ned Bass <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>

OpenZFS-issue: https://www.illumos.org/issues/7004
OpenZFS-commit: https://github.com/openzfs/openzfs/pull/109
Closes #4641
Closes #4972

1 files changed, 48 insertions, 3 deletions

diff --git a/module/zfs/zap_micro.c b/module/zfs/zap_micro.c
index de646287e..bc0a5e71d 100644
--- a/module/zfs/zap_micro.c
+++ b/module/zfs/zap_micro.c
@@ -536,6 +536,24 @@ zap_lockdir_impl(dmu_buf_t *db, void *tag, dmu_tx_t *tx,
 	return (0);
 }
 
+static int
+zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx,
+    krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
+{
+	dmu_buf_t *db;
+	int err;
+
+	err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
+	if (err != 0) {
+		return (err);
+	}
+	err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
+	if (err != 0) {
+		dmu_buf_rele(db, tag);
+	}
+	return (err);
+}
+
 int
 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
     krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
@@ -928,6 +946,33 @@ zap_prefetch(objset_t *os, uint64_t zapobj, const char *name)
 }
 
 int
+zap_lookup_by_dnode(dnode_t *dn, const char *name,
+    uint64_t integer_size, uint64_t num_integers, void *buf)
+{
+	return (zap_lookup_norm_by_dnode(dn, name, integer_size,
+	    num_integers, buf, MT_EXACT, NULL, 0, NULL));
+}
+
+int
+zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
+    uint64_t integer_size, uint64_t num_integers, void *buf,
+    matchtype_t mt, char *realname, int rn_len,
+    boolean_t *ncp)
+{
+	zap_t *zap;
+	int err;
+
+	err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
+	    FTAG, &zap);
+	if (err != 0)
+		return (err);
+	err = zap_lookup_impl(zap, name, integer_size,
+	    num_integers, buf, mt, realname, rn_len, ncp);
+	zap_unlockdir(zap, FTAG);
+	return (err);
+}
+
+int
 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
     int key_numints)
 {
@@ -1460,7 +1505,7 @@ zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
 }
 
 int
-zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
+zap_count_write_by_dnode(dnode_t *dn, const char *name, int add,
     uint64_t *towrite, uint64_t *tooverwrite)
 {
 	zap_t *zap;
@@ -1488,7 +1533,7 @@ zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
 	 * At present we are just evaluating the possibility of this operation
 	 * and hence we do not want to trigger an upgrade.
 	 */
-	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE,
+	err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
 	    FTAG, &zap);
 	if (err != 0)
 		return (err);
@@ -1552,7 +1597,7 @@ EXPORT_SYMBOL(zap_lookup_uint64);
 EXPORT_SYMBOL(zap_contains);
 EXPORT_SYMBOL(zap_prefetch);
 EXPORT_SYMBOL(zap_prefetch_uint64);
-EXPORT_SYMBOL(zap_count_write);
+EXPORT_SYMBOL(zap_count_write_by_dnode);
 EXPORT_SYMBOL(zap_add);
 EXPORT_SYMBOL(zap_add_uint64);
 EXPORT_SYMBOL(zap_update);