Allow txg_kick() even there is a txg still syncing

Current txg_kick() only works if there is no txg in either quiescing or syncing. This commit allow txg_kick() to work as long as there is no txg is quiescing. This keeps syncing stage busy, which should benefit write throughput. Also txg_kick() should be triggered on dp_dirty_pertxg[] instead of dp_dirty_total, because dp_dirty_total won't be decreased until the dirty txg is synced. This change prevent pushing a empty txg through pipeline in a row. Signed-off-by: jxdking <[email protected]>
openzfs · Apr 22, 2021 · 184f4cd · 184f4cd
1 parent d92af6f
commit 184f4cd
Show file tree

Hide file tree

Showing 5 changed files with 28 additions and 19 deletions.
diff --git a/include/sys/dsl_pool.h b/include/sys/dsl_pool.h
@@ -171,6 +171,7 @@ void dsl_pool_mos_diduse_space(dsl_pool_t *dp,
 void dsl_pool_ckpoint_diduse_space(dsl_pool_t *dp,
     int64_t used, int64_t comp, int64_t uncomp);
 boolean_t dsl_pool_need_dirty_delay(dsl_pool_t *dp);
+boolean_t dsl_pool_need_dirty_sync(dsl_pool_t *dp, uint64_t txg);
 void dsl_pool_config_enter(dsl_pool_t *dp, void *tag);
 void dsl_pool_config_enter_prio(dsl_pool_t *dp, void *tag);
 void dsl_pool_config_exit(dsl_pool_t *dp, void *tag);

diff --git a/include/sys/txg.h b/include/sys/txg.h
@@ -78,7 +78,7 @@ extern void txg_register_callbacks(txg_handle_t *txghp, list_t *tx_callbacks);
 
 extern void txg_delay(struct dsl_pool *dp, uint64_t txg, hrtime_t delta,
     hrtime_t resolution);
-extern void txg_kick(struct dsl_pool *dp);
+extern void txg_kick(struct dsl_pool *dp, uint64_t txg);
 
 /*
  * Wait until the given transaction group has finished syncing.

diff --git a/module/zfs/dmu_tx.c b/module/zfs/dmu_tx.c
@@ -38,6 +38,7 @@
 #include <sys/sa_impl.h>
 #include <sys/zfs_context.h>
 #include <sys/trace_zfs.h>
+#include <sys/txg.h>
 
 typedef void (*dmu_tx_hold_func_t)(dmu_tx_t *tx, struct dnode *dn,
     uint64_t arg1, uint64_t arg2);
@@ -1055,6 +1056,9 @@ dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how)
 
 	txg_rele_to_quiesce(&tx->tx_txgh);
 
+	if (dsl_pool_need_dirty_sync(tx->tx_pool, tx->tx_txg)) {
+		txg_kick(tx->tx_pool, tx->tx_txg);
+	}
 	return (0);
 }
 

diff --git a/module/zfs/dsl_pool.c b/module/zfs/dsl_pool.c
@@ -899,18 +899,29 @@ dsl_pool_need_dirty_delay(dsl_pool_t *dp)
 {
 	uint64_t delay_min_bytes =
 	    zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100;
-	uint64_t dirty_min_bytes =
-	    zfs_dirty_data_max * zfs_dirty_data_sync_percent / 100;
 	uint64_t dirty;
 
 	mutex_enter(&dp->dp_lock);
 	dirty = dp->dp_dirty_total;
 	mutex_exit(&dp->dp_lock);
-	if (dirty > dirty_min_bytes)
-		txg_kick(dp);
+
 	return (dirty > delay_min_bytes);
 }
 
+boolean_t
+dsl_pool_need_dirty_sync(dsl_pool_t *dp, uint64_t txg)
+{
+	uint64_t dirty;
+	uint64_t dirty_min_bytes =
+	    zfs_dirty_data_max * zfs_dirty_data_sync_percent / 100;
+
+	mutex_enter(&dp->dp_lock);
+	dirty = dp->dp_dirty_pertxg[txg & TXG_MASK];
+	mutex_exit(&dp->dp_lock);
+
+	return (dirty > dirty_min_bytes);
+}
+
 void
 dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
 {

diff --git a/module/zfs/txg.c b/module/zfs/txg.c
@@ -498,14 +498,6 @@ txg_wait_callbacks(dsl_pool_t *dp)
 		taskq_wait_outstanding(tx->tx_commit_cb_taskq, 0);
 }
 
-static boolean_t
-txg_is_syncing(dsl_pool_t *dp)
-{
-	tx_state_t *tx = &dp->dp_tx;
-	ASSERT(MUTEX_HELD(&tx->tx_sync_lock));
-	return (tx->tx_syncing_txg != 0);
-}
-
 static boolean_t
 txg_is_quiescing(dsl_pool_t *dp)
 {
@@ -784,22 +776,23 @@ txg_wait_open(dsl_pool_t *dp, uint64_t txg, boolean_t should_quiesce)
 }
 
 /*
- * If there isn't a txg syncing or in the pipeline, push another txg through
- * the pipeline by quiescing the open txg.
+ * If there isn't a txg quiescing in the pipeline, push the txg
+ * through the pipeline by quiescing the open txg.
+ * It is fine there is a txg still syncing.
  */
 void
-txg_kick(dsl_pool_t *dp)
+txg_kick(dsl_pool_t *dp, uint64_t txg)
 {
 	tx_state_t *tx = &dp->dp_tx;
 
 	ASSERT(!dsl_pool_config_held(dp));
 
 	mutex_enter(&tx->tx_sync_lock);
-	if (!txg_is_syncing(dp) &&
+	txg = txg == 0 ? tx->tx_open_txg : txg;
+	if (txg == tx->tx_open_txg &&
 	    !txg_is_quiescing(dp) &&
 	    tx->tx_quiesce_txg_waiting <= tx->tx_open_txg &&
-	    tx->tx_sync_txg_waiting <= tx->tx_synced_txg &&
-	    tx->tx_quiesced_txg <= tx->tx_synced_txg) {
+	    tx->tx_sync_txg_waiting <= tx->tx_synced_txg) {
 		tx->tx_quiesce_txg_waiting = tx->tx_open_txg + 1;
 		cv_broadcast(&tx->tx_quiesce_more_cv);
 	}