[optimize](table stat) Reduce lock when get table statistics (#39807)

## Proposed changes Issue Number: this pr will pick part of #35457 to reduce the potential of dead lock with following example: 时间点 | create table线程1 | create table线程2 | table stat线程 | truncate线程 -- | -- | -- | -- | -- t1 | add table1 read lock | | | t2 | | add table2 read lock | | t3 | | | try add table1 write lock，blocked | t4 | | | | try add table2 write lock，blocked t5 | try add table2 read lock，Due to a thread attempting to add the table2 write lock earlier, it was forced to queue up | | | t6 | | try add table1 read lock，Due to a thread attempting to add the table1 write lock earlier, it was forced to queue up | | This deadlock situation usually occurs when it is a fair lock. So we should optimize the read lock. This PR places the updates of table statistics(such as data size/replica count) in the Table Stats Thread to be done on a scheduled basis, so that there is no need to add a read lock when obtaining table statistics.
apache · Aug 28, 2024 · 367d264 · 367d264
1 parent 299d221
commit 367d264
Show file tree

Hide file tree

Showing 4 changed files with 98 additions and 56 deletions.
diff --git a/fe/fe-core/src/main/java/org/apache/doris/analysis/ShowDataStmt.java b/fe/fe-core/src/main/java/org/apache/doris/analysis/ShowDataStmt.java
@@ -151,13 +151,8 @@ public int compare(Table t1, Table t2) {
                     OlapTable olapTable = (OlapTable) table;
                     long tableSize = 0;
                     long replicaCount = 0;
-                    olapTable.readLock();
-                    try {
-                        tableSize = olapTable.getDataSize();
-                        replicaCount = olapTable.getReplicaCount();
-                    } finally {
-                        olapTable.readUnlock();
-                    }
+                    tableSize = olapTable.getDataSize();
+                    replicaCount = olapTable.getReplicaCount();
                     //|TableName|Size|ReplicaCount|
                     List<Object> row = Arrays.asList(table.getName(), tableSize, replicaCount);
                     totalRowsObject.add(row);

diff --git a/fe/fe-core/src/main/java/org/apache/doris/catalog/Database.java b/fe/fe-core/src/main/java/org/apache/doris/catalog/Database.java
@@ -231,48 +231,42 @@ public void setDbProperties(DatabaseProperty dbProperties) {
 
     public long getUsedDataQuotaWithLock() {
         long usedDataQuota = 0;
+        List<Table> tables = new ArrayList<>();
         readLock();
         try {
-            for (Table table : this.idToTable.values()) {
-                if (table.getType() != TableType.OLAP) {
-                    continue;
-                }
-
-                OlapTable olapTable = (OlapTable) table;
-                olapTable.readLock();
-                try {
-                    usedDataQuota = usedDataQuota + olapTable.getDataSize();
-                } finally {
-                    olapTable.readUnlock();
-                }
-            }
-            return usedDataQuota;
+            tables.addAll(this.idToTable.values());
         } finally {
             readUnlock();
         }
+        for (Table table : tables) {
+            if (table.getType() != TableType.OLAP) {
+                continue;
+            }
+
+            OlapTable olapTable = (OlapTable) table;
+            usedDataQuota = usedDataQuota + olapTable.getDataSize();
+        }
+        return usedDataQuota;
     }
 
     public long getReplicaCountWithLock() {
+        long usedReplicaCount = 0;
+        List<Table> tables = new ArrayList<>();
         readLock();
         try {
-            long usedReplicaCount = 0;
-            for (Table table : this.idToTable.values()) {
-                if (table.getType() != TableType.OLAP) {
-                    continue;
-                }
-
-                OlapTable olapTable = (OlapTable) table;
-                olapTable.readLock();
-                try {
-                    usedReplicaCount = usedReplicaCount + olapTable.getReplicaCount();
-                } finally {
-                    olapTable.readUnlock();
-                }
-            }
-            return usedReplicaCount;
+            tables.addAll(this.idToTable.values());
         } finally {
             readUnlock();
         }
+        for (Table table : tables) {
+            if (table.getType() != TableType.OLAP) {
+                continue;
+            }
+
+            OlapTable olapTable = (OlapTable) table;
+            usedReplicaCount = usedReplicaCount + olapTable.getReplicaCount();
+        }
+        return usedReplicaCount;
     }
 
     public long getReplicaQuotaLeftWithLock() {

diff --git a/fe/fe-core/src/main/java/org/apache/doris/catalog/OlapTable.java b/fe/fe-core/src/main/java/org/apache/doris/catalog/OlapTable.java
@@ -86,6 +86,8 @@
 import java.util.Set;
 import java.util.stream.Collectors;
 
+import lombok.Getter;
+
 /**
  * Internal representation of tableFamilyGroup-related metadata. A OlaptableFamilyGroup contains several tableFamily.
  * Note: when you add a new olap table property, you should modify TableProperty class
@@ -148,6 +150,8 @@ public enum OlapTableState {
 
     private TableProperty tableProperty;
 
+    private volatile Statistics statistics = new Statistics();
+
     public OlapTable() {
         // for persist
         super(TableType.OLAP);
@@ -1393,26 +1397,6 @@ public Partition replacePartition(Partition newPartition) {
         return oldPartition;
     }
 
-    public long getDataSize(boolean singleReplica) {
-        long dataSize = 0;
-        for (Partition partition : getAllPartitions()) {
-            dataSize += partition.getDataSize(singleReplica);
-        }
-        return dataSize;
-    }
-
-    public long getDataSize() {
-        return getDataSize(false);
-    }
-
-    public long getReplicaCount() {
-        long replicaCount = 0;
-        for (Partition partition : getAllPartitions()) {
-            replicaCount += partition.getReplicaCount();
-        }
-        return replicaCount;
-    }
-
     public void checkStableAndNormal(String clusterName) throws DdlException {
         if (state != OlapTableState.NORMAL) {
             throw new DdlException("Table[" + name + "]'s state is not NORMAL. "
@@ -1995,4 +1979,59 @@ public boolean isDupKeysOrMergeOnWrite() {
                 || (getKeysType() == KeysType.UNIQUE_KEYS
                 && getEnableUniqueKeyMergeOnWrite());
     }
+
+    public void setStatistics(Statistics statistics) {
+        this.statistics = statistics;
+    }
+
+    public static class Statistics {
+        @Getter
+        private String dbName;
+        @Getter
+        private String tableName;
+
+        @Getter
+        private Long dataSize; // single replica data size
+        @Getter
+        private Long totalReplicaDataSize;
+
+        @Getter
+        private Long replicaCount;
+
+        public Statistics() {
+            this.dbName = null;
+            this.tableName = null;
+
+            this.dataSize = 0L;
+            this.totalReplicaDataSize = 0L;
+
+            this.replicaCount = 0L;
+        }
+
+        public Statistics(String dbName, String tableName,
+                          Long dataSize, Long totalReplicaDataSize, Long replicaCount) {
+            this.dbName = dbName;
+            this.tableName = tableName;
+
+            this.dataSize = dataSize;
+            this.totalReplicaDataSize = totalReplicaDataSize;
+
+            this.replicaCount = replicaCount;
+        }
+    }
+
+    public long getDataSize() {
+        return getDataSize(false);
+    }
+
+    public long getDataSize(boolean singleReplica) {
+        if (singleReplica) {
+            return statistics.getDataSize();
+        }
+        return statistics.getTotalReplicaDataSize();
+    }
+
+    public long getReplicaCount() {
+        return statistics.getReplicaCount();
+    }
 }
diff --git a/fe/fe-core/src/main/java/org/apache/doris/catalog/TabletStatMgr.java b/fe/fe-core/src/main/java/org/apache/doris/catalog/TabletStatMgr.java
@@ -94,6 +94,9 @@ protected void runAfterCatalogReady() {
                     continue;
                 }
                 OlapTable olapTable = (OlapTable) table;
+                long tableDataSize = 0L;
+                long tableTotalReplicaDataSize = 0L;
+                long tableReplicaCount = 0L;
                 if (!table.writeLockIfExist()) {
                     continue;
                 }
@@ -103,18 +106,29 @@ protected void runAfterCatalogReady() {
                         for (MaterializedIndex index : partition.getMaterializedIndices(IndexExtState.VISIBLE)) {
                             long indexRowCount = 0L;
                             for (Tablet tablet : index.getTablets()) {
+                                long tabletDataSize = 0L;
                                 long tabletRowCount = 0L;
                                 for (Replica replica : tablet.getReplicas()) {
                                     if (replica.checkVersionCatchUp(version, false)
                                             && replica.getRowCount() > tabletRowCount) {
                                         tabletRowCount = replica.getRowCount();
                                     }
+
+                                    if (replica.getDataSize() > tabletDataSize) {
+                                        tabletDataSize = replica.getDataSize();
+                                    }
+                                    tableTotalReplicaDataSize += replica.getDataSize();
+                                    tableReplicaCount++;
                                 }
+                                tableDataSize += tabletDataSize;
                                 indexRowCount += tabletRowCount;
                             } // end for tablets
                             index.setRowCount(indexRowCount);
                         } // end for indices
                     } // end for partitions
+
+                    olapTable.setStatistics(new OlapTable.Statistics(db.getFullName(), table.getName(),
+                        tableDataSize, tableTotalReplicaDataSize, tableReplicaCount));
                     LOG.debug("finished to set row num for table: {} in database: {}",
                              table.getName(), db.getFullName());
                 } finally {