[Refactor] Synchronize OLAP external table metadata when loading data (…

…StarRocks#24739)

Currently, StarRocks use StarRocksRepository to synchronize OLAP
external table metadata every 10 seconds, which is
inefficient, because the meta data will never be used when there is no
load job on that external table. Change the synchronization to be
triggered when there is load job on that table.

Since the synchronization is time costly, so we should release the
database lock when synchronizing meta data.

Signed-off-by: gengjun-git <gengjun@starrocks.com>

fe/fe-core/src/main/java/com/starrocks/catalog/Database.java

@@ -213,6 +213,10 @@ public void readUnlock() {
         this.rwLock.readLock().unlock();
     }
 
+    public boolean isReadLockHeldByCurrentThread() {
+        return this.rwLock.getReadHoldCount() > 0;
+    }
+
     public void writeLock() {
         long startMs = TimeUnit.MILLISECONDS.convert(System.nanoTime(), TimeUnit.NANOSECONDS);
         String threadDump = getOwnerInfo(rwLock.getOwner());

fe/fe-core/src/main/java/com/starrocks/catalog/OlapTable.java

@@ -248,6 +248,40 @@ public OlapTable(long id, String tableName, List<Column> baseSchema, KeysType ke
         this.tableProperty = null;
     }
 
+    // Only Copy necessary metadata for query.
+    // We don't do deep copy, because which is very expensive;
+    public void copyOnlyForQuery(OlapTable olapTable) {
+        olapTable.id = this.id;
+        olapTable.name = this.name;
+        olapTable.fullSchema = Lists.newArrayList(this.fullSchema);
+        olapTable.nameToColumn = Maps.newHashMap(this.nameToColumn);
+        olapTable.relatedMaterializedViews = Sets.newHashSet(this.relatedMaterializedViews);
+        olapTable.state = this.state;
+        olapTable.indexNameToId = Maps.newHashMap(this.indexNameToId);
+        olapTable.indexIdToMeta = Maps.newHashMap(this.indexIdToMeta);
+        olapTable.keysType = this.keysType;
+        olapTable.partitionInfo = new PartitionInfo();
+        if (this.partitionInfo instanceof RangePartitionInfo) {
+            olapTable.partitionInfo = new RangePartitionInfo((RangePartitionInfo) this.partitionInfo);
+        } else if (this.partitionInfo instanceof SinglePartitionInfo) {
+            olapTable.partitionInfo = this.partitionInfo;
+        }
+        olapTable.defaultDistributionInfo = this.defaultDistributionInfo;
+        Map<Long, Partition> idToPartitions = new HashMap<>(this.idToPartition.size());
+        Map<String, Partition> nameToPartitions = Maps.newTreeMap(String.CASE_INSENSITIVE_ORDER);
+        for (Map.Entry<Long, Partition> kv : this.idToPartition.entrySet()) {
+            Partition copiedPartition = kv.getValue().shallowCopy();
+            idToPartitions.put(kv.getKey(), copiedPartition);
+            nameToPartitions.put(kv.getValue().getName(), copiedPartition);
+        }
+        olapTable.idToPartition = idToPartitions;
+        olapTable.nameToPartition = nameToPartitions;
+        olapTable.baseIndexId = this.baseIndexId;
+        if (this.tableProperty != null) {
+            olapTable.tableProperty = this.tableProperty.copy();
+        }
+    }
+
     public void setTableProperty(TableProperty tableProperty) {
         this.tableProperty = tableProperty;
     }

fe/fe-core/src/main/java/com/starrocks/catalog/Partition.java

@@ -117,6 +117,21 @@ public Partition(long id, String name,
         this.distributionInfo = distributionInfo;
     }
 
+    public Partition shallowCopy() {
+        Partition partition = new Partition();
+        partition.id = this.id;
+        partition.name = this.name;
+        partition.state = this.state;
+        partition.baseIndex = this.baseIndex;
+        partition.idToVisibleRollupIndex = Maps.newHashMap(this.idToVisibleRollupIndex);
+        partition.idToShadowIndex = Maps.newHashMap(this.idToShadowIndex);
+        partition.visibleVersion = this.visibleVersion;
+        partition.visibleVersionTime = this.visibleVersionTime;
+        partition.nextVersion = this.nextVersion;
+        partition.distributionInfo = this.distributionInfo;
+        return partition;
+    }
+
     public void setIdForRestore(long id) {
         this.id = id;
     }

fe/fe-core/src/main/java/com/starrocks/catalog/RangePartitionInfo.java

@@ -87,6 +87,14 @@ public RangePartitionInfo(List<Column> partitionColumns) {
         this.isMultiColumnPartition = partitionColumns.size() > 1;
     }
 
+    public RangePartitionInfo(RangePartitionInfo other) {
+        super(other.type);
+        this.partitionColumns = Lists.newArrayList(other.partitionColumns);
+        this.idToRange = Maps.newHashMap(other.idToRange);
+        this.idToTempRange = Maps.newHashMap(other.idToTempRange);
+        this.isMultiColumnPartition = partitionColumns.size() > 1;
+    }
+
     @Override
     public List<Column> getPartitionColumns() {
         return partitionColumns;

fe/fe-core/src/main/java/com/starrocks/catalog/Table.java

@@ -126,7 +126,7 @@ public enum TableType {
 
     // not serialized field
     // record all materialized views based on this Table
-    private Set<MvId> relatedMaterializedViews;
+    protected Set<MvId> relatedMaterializedViews;
 
     // unique constraints for mv rewrite
     // a table may have multi unique constraints

fe/fe-core/src/main/java/com/starrocks/catalog/TableProperty.java

@@ -22,6 +22,7 @@
 package com.starrocks.catalog;
 
 import com.clearspring.analytics.util.Lists;
+import com.google.common.base.Preconditions;
 import com.google.common.base.Splitter;
 import com.google.common.collect.Maps;
 import com.google.gson.annotations.SerializedName;
@@ -134,6 +135,18 @@ public TableProperty(Map<String, String> properties) {
         this.properties = properties;
     }
 
+    public TableProperty copy() {
+        TableProperty newTableProperty = new TableProperty(Maps.newHashMap(this.properties));
+        try {
+            newTableProperty.gsonPostProcess();
+        } catch (IOException e) {
+            Preconditions.checkState(false, "gsonPostProcess shouldn't fail");
+        }
+        newTableProperty.hasDelete = this.hasDelete;
+        newTableProperty.hasForbitGlobalDict = this.hasDelete;
+        return newTableProperty;
+    }
+
     public static boolean isSamePrefixProperties(Map<String, String> properties, String prefix) {
         for (String value : properties.keySet()) {
             if (!value.startsWith(prefix)) {

fe/fe-core/src/main/java/com/starrocks/external/starrocks/TableMetaSyncer.java

@@ -5,6 +5,7 @@
 import com.starrocks.catalog.ExternalOlapTable;
 import com.starrocks.common.Config;
 import com.starrocks.rpc.FrontendServiceProxy;
+import com.starrocks.sql.common.MetaNotFoundException;
 import com.starrocks.thrift.TGetTableMetaRequest;
 import com.starrocks.thrift.TGetTableMetaResponse;
 import com.starrocks.thrift.TNetworkAddress;
@@ -17,7 +18,7 @@
 public class TableMetaSyncer {
     private static final Logger LOG = LogManager.getLogger(TableMetaSyncer.class);
 
-    public void syncTable(ExternalOlapTable table) {
+    public void syncTable(ExternalOlapTable table) throws MetaNotFoundException {
         String host = table.getSourceTableHost();
         int port = table.getSourceTablePort();
         TNetworkAddress addr = new TNetworkAddress(host, port);
@@ -37,11 +38,13 @@ public void syncTable(ExternalOlapTable table) {
                     errMsg = "";
                 }
                 LOG.warn("get TableMeta failed: {}", errMsg);
+                throw new MetaNotFoundException(errMsg);
             } else {
                 table.updateMeta(request.getDb_name(), response.getTable_meta(), response.getBackends());
             }
         } catch (Exception e) {
             LOG.warn("call fe {} refreshTable rpc method failed", addr, e);
+            throw new MetaNotFoundException("get TableMeta failed from " + addr + ", error: " + e.getMessage());
         }
     }
-};
+}

fe/fe-core/src/main/java/com/starrocks/qe/StmtExecutor.java

@@ -1279,7 +1279,12 @@ private void handleDMLStmtImpl(ExecPlan execPlan, DmlStmt stmt) throws Exception
 
         MetaUtils.normalizationTableName(context, stmt.getTableName());
         Database database = MetaUtils.getDatabase(context, stmt.getTableName());
-        Table targetTable = MetaUtils.getTable(context, stmt.getTableName());
+        Table targetTable;
+        if (stmt instanceof InsertStmt && ((InsertStmt) stmt).getTargetTable() != null) {
+            targetTable = ((InsertStmt) stmt).getTargetTable();
+        } else {
+            targetTable = MetaUtils.getTable(context, stmt.getTableName());
+        }
 
         String label = DebugUtil.printId(context.getExecutionId());
         if (stmt instanceof InsertStmt) {

fe/fe-core/src/main/java/com/starrocks/server/GlobalStateMgr.java

@@ -124,7 +124,6 @@
 import com.starrocks.consistency.ConsistencyChecker;
 import com.starrocks.credential.CloudCredentialUtil;
 import com.starrocks.external.elasticsearch.EsRepository;
-import com.starrocks.external.starrocks.StarRocksRepository;
 import com.starrocks.ha.FrontendNodeType;
 import com.starrocks.ha.HAProtocol;
 import com.starrocks.ha.LeaderInfo;
@@ -338,7 +337,6 @@ public class GlobalStateMgr {
     private Daemon replayer;
     private Daemon timePrinter;
     private EsRepository esRepository;  // it is a daemon, so add it here
-    private StarRocksRepository starRocksRepository;
     private IcebergRepository icebergRepository;
     private MetastoreEventsProcessor metastoreEventsProcessor;
     private ConnectorTableMetadataProcessor connectorTableMetadataProcessor;
@@ -592,7 +590,6 @@ private GlobalStateMgr(boolean isCheckpointCatalog) {
         this.resourceGroupMgr = new ResourceGroupMgr(this);
 
         this.esRepository = new EsRepository();
-        this.starRocksRepository = new StarRocksRepository();
         this.icebergRepository = new IcebergRepository();
         this.metastoreEventsProcessor = new MetastoreEventsProcessor();
         this.connectorTableMetadataProcessor = new ConnectorTableMetadataProcessor();
@@ -1217,7 +1214,6 @@ private void startNonLeaderDaemonThreads() {
         labelCleaner.start();
         // ES state store
         esRepository.start();
-        starRocksRepository.start();
 
         if (Config.enable_hms_events_incremental_sync) {
             metastoreEventsProcessor.start();
@@ -1291,7 +1287,6 @@ public void loadImage(String imageDir) throws IOException, DdlException {
             localMetastore.recreateTabletInvertIndex();
             // rebuild es state state
             esRepository.loadTableFromCatalog();
-            starRocksRepository.loadTableFromCatalog();
 
             checksum = load.loadLoadJob(dis, checksum);
             checksum = loadAlterJob(dis, checksum);
@@ -2821,15 +2816,10 @@ public EsRepository getEsRepository() {
         return this.esRepository;
     }
 
-    public StarRocksRepository getStarRocksRepository() {
-        return this.starRocksRepository;
-    }
-
     public IcebergRepository getIcebergRepository() {
         return this.icebergRepository;
     }
 
-
     public MetastoreEventsProcessor getMetastoreEventsProcessor() {
         return this.metastoreEventsProcessor;
     }

fe/fe-core/src/main/java/com/starrocks/sql/analyzer/InsertAnalyzer.java

@@ -7,11 +7,14 @@
 import com.starrocks.analysis.Expr;
 import com.starrocks.catalog.Column;
 import com.starrocks.catalog.Database;
+import com.starrocks.catalog.ExternalOlapTable;
 import com.starrocks.catalog.MaterializedView;
 import com.starrocks.catalog.MysqlTable;
 import com.starrocks.catalog.OlapTable;
 import com.starrocks.catalog.Partition;
 import com.starrocks.catalog.Table;
+import com.starrocks.common.AnalysisException;
+import com.starrocks.external.starrocks.TableMetaSyncer;
 import com.starrocks.qe.ConnectContext;
 import com.starrocks.sql.ast.DefaultValueExpr;
 import com.starrocks.sql.ast.InsertStmt;
@@ -40,6 +43,10 @@ public static void analyze(InsertStmt insertStmt, ConnectContext session) {
         Database database = MetaUtils.getDatabase(session, insertStmt.getTableName());
         Table table = MetaUtils.getTable(session, insertStmt.getTableName());
 
+        if (table instanceof ExternalOlapTable) {
+            table = getOLAPExternalTableMeta(database, (ExternalOlapTable) table);
+        }
+
         if (table instanceof MaterializedView && !insertStmt.isSystem()) {
             throw new SemanticException(
                     "The data of '%s' cannot be inserted because '%s' is a materialized view," +
@@ -151,4 +158,23 @@ public static void analyze(InsertStmt insertStmt, ConnectContext session) {
         insertStmt.setTargetColumns(targetColumns);
         session.getDumpInfo().addTable(database.getFullName(), table);
     }
+
+    private static ExternalOlapTable getOLAPExternalTableMeta(Database database, ExternalOlapTable externalOlapTable) {
+        // copy the table, and release database lock when synchronize table meta
+        ExternalOlapTable copiedTable = new ExternalOlapTable();
+        externalOlapTable.copyOnlyForQuery(copiedTable);
+        int lockTimes = 0;
+        while (database.isReadLockHeldByCurrentThread()) {
+            database.readUnlock();
+            lockTimes++;
+        }
+        try {
+            new TableMetaSyncer().syncTable(copiedTable);
+        } finally {
+            while (lockTimes-- > 0) {
+                database.readLock();
+            }
+        }
+        return copiedTable;
+    }
 }