[Enhancement] reduce lock granular of TabletStatMgr (#50668)

Signed-off-by: Murphy <[email protected]> (cherry picked from commit 0b796e0) # Conflicts: # fe/fe-core/src/main/java/com/starrocks/catalog/TabletStatMgr.java
StarRocks · Sep 4, 2024 · 5fe81ef · 5fe81ef
1 parent 4673054
commit 5fe81ef
Showing 1 changed file with 43 additions and 6 deletions.
diff --git a/fe/fe-core/src/main/java/com/starrocks/catalog/TabletStatMgr.java b/fe/fe-core/src/main/java/com/starrocks/catalog/TabletStatMgr.java
@@ -36,6 +36,7 @@
 
 import com.google.common.collect.ImmutableMap;
 import com.google.common.collect.Lists;
+import com.google.common.collect.Maps;
 import com.starrocks.catalog.MaterializedIndex.IndexExtState;
 import com.starrocks.common.Config;
 import com.starrocks.common.util.FrontendDaemon;
@@ -107,25 +108,38 @@ protected void runAfterCatalogReady() {
                 continue;
             }
             Locker locker = new Locker();
+<<<<<<< HEAD
             locker.lockDatabase(db, LockType.WRITE);
             try {
                 for (Table table : db.getTables()) {
                     long totalRowCount = 0L;
                     if (!table.isNativeTableOrMaterializedView()) {
                         continue;
                     }
+=======
+            for (Table table : GlobalStateMgr.getCurrentState().getLocalMetastore().getTables(db.getId())) {
+                long totalRowCount = 0L;
+                if (!table.isNativeTableOrMaterializedView()) {
+                    continue;
+                }
+>>>>>>> 0b796e0ba0 ([Enhancement] reduce lock granular of TabletStatMgr (#50668))
 
+                // NOTE: calculate the row first with read lock, then update the stats with write lock
+                locker.lockTableWithIntensiveDbLock(db, table.getId(), LockType.READ);
+                Map<Long, Long> indexRowCountMap = Maps.newHashMap();
+                try {
                     OlapTable olapTable = (OlapTable) table;
                     for (Partition partition : olapTable.getAllPartitions()) {
                         for (PhysicalPartition physicalPartition : partition.getSubPartitions()) {
                             long version = physicalPartition.getVisibleVersion();
                             for (MaterializedIndex index : physicalPartition.getMaterializedIndices(
                                     IndexExtState.VISIBLE)) {
                                 long indexRowCount = 0L;
+                                // NOTE: can take a rather long time to iterate lots of tablets
                                 for (Tablet tablet : index.getTablets()) {
                                     indexRowCount += tablet.getRowCount(version);
                                 } // end for tablets
-                                index.setRowCount(indexRowCount);
+                                indexRowCountMap.put(index.getId(), indexRowCount);
                                 if (!olapTable.isTempPartition(partition.getId())) {
                                     totalRowCount += indexRowCount;
                                 }
@@ -134,10 +148,29 @@ protected void runAfterCatalogReady() {
                     } // end for partitions
                     LOG.debug("finished to set row num for table: {} in database: {}",
                             table.getName(), db.getFullName());
+                } finally {
+                    locker.unLockTableWithIntensiveDbLock(db, table, LockType.READ);
+                }
+
+                // update
+                locker.lockTableWithIntensiveDbLock(db, table.getId(), LockType.WRITE);
+                try {
+                    OlapTable olapTable = (OlapTable) table;
+                    for (Partition partition : olapTable.getAllPartitions()) {
+                        for (PhysicalPartition physicalPartition : partition.getSubPartitions()) {
+                            for (MaterializedIndex index :
+                                    physicalPartition.getMaterializedIndices(IndexExtState.VISIBLE)) {
+                                Long indexRowCount = indexRowCountMap.get(index.getId());
+                                if (indexRowCount != null) {
+                                    index.setRowCount(indexRowCount);
+                                }
+                            }
+                        }
+                    }
                     adjustStatUpdateRows(table.getId(), totalRowCount);
+                } finally {
+                    locker.unLockTableWithIntensiveDbLock(db, table, LockType.WRITE);
                 }
-            } finally {
-                locker.unLockDatabase(db, LockType.WRITE);
             }
         }
         LOG.info("finished to update index row num of all databases. cost: {} ms",
@@ -149,7 +182,8 @@ private void updateLocalTabletStat() {
         if (!RunMode.isSharedNothingMode()) {
             return;
         }
-        ImmutableMap<Long, Backend> backends = GlobalStateMgr.getCurrentState().getNodeMgr().getClusterInfo().getIdToBackend();
+        ImmutableMap<Long, Backend> backends =
+                GlobalStateMgr.getCurrentState().getNodeMgr().getClusterInfo().getIdToBackend();
 
         long start = System.currentTimeMillis();
         for (Backend backend : backends.values()) {
@@ -354,10 +388,13 @@ private void sendTasks() {
                     LakeService lakeService = BrpcProxy.getLakeService(node.getHost(), node.getBrpcPort());
                     Future<TabletStatResponse> responseFuture = lakeService.getTabletStats(request);
                     responseList.add(responseFuture);
-                    LOG.debug("Sent tablet stat collection task to node {} for partition {} of version {}. tablet count={}",
+                    LOG.debug(
+                            "Sent tablet stat collection task to node {} for partition {} of version {}. tablet " +
+                                    "count={}",
                             node.getHost(), debugName(), version, entry.getValue().size());
                 } catch (Throwable e) {
-                    LOG.warn("Fail to send tablet stat task to host {} for partition {}: {}", node.getHost(), debugName(),
+                    LOG.warn("Fail to send tablet stat task to host {} for partition {}: {}", node.getHost(),
+                            debugName(),
                             e.getMessage());
                 }
             }