feature: Optimize partial refresh and add retry for executeMutation/query (#213)

* remove lock in refreshTableLocationByTabletId

* fix refresh sql

* fix refresh interval

* add retry logic for common query

* add retry logic for executeMutation

* add retry logic for executeMutation

* add retry logic for ObTableQueryAndMutateRequest

* fix: correct ineffective retry logic

* retry logic for ObTableClient execute

* fix

* fix

Author: maochongxin

src/main/java/com/alipay/oceanbase/rpc/ObTableClient.java

@@ -678,6 +678,7 @@ private <T> T execute(String tableName, TableExecuteCallback<T> callback, ObServ
                                     tryTimes);
                             if (ex instanceof ObTableNeedFetchAllException) {
                                 needFetchAllRouteInfo = true;
+                                getOrRefreshTableEntry(tableName, true, true, true);
                                 // reset failure count while fetch all route info
                                 this.resetExecuteContinuousFailureCount(tableName);
                             }
@@ -768,7 +769,6 @@ private <T> T executeMutation(String tableName, MutationExecuteCallback<T> callb
             throw new IllegalArgumentException("table name is null");
         }
         boolean needRefreshTableEntry = false;
-        boolean needFetchAllRouteInfo = false;
         int tryTimes = 0;
         long startExecute = System.currentTimeMillis();
         while (true) {
@@ -788,10 +788,14 @@ private <T> T executeMutation(String tableName, MutationExecuteCallback<T> callb
                     obPair = new ObPair<Long, ObTableParam>(0L, new ObTableParam(odpTable));
                 } else {
                     if (null != callback.getRowKey()) {
+                        // in the case of retry, the location always needs to be refreshed here
+                        if (tryTimes > 1) {
+                            TableEntry entry = getOrRefreshTableEntry(tableName, false, false, false);
+                            Long partId = getPartition(entry, callback.getRowKey());
+                            refreshTableLocationByTabletId(entry, tableName, getTabletIdByPartId(entry, partId));
+                        }
                         // using row key
-                        obPair = getTable(tableName, callback.getRowKey(),
-                            needRefreshTableEntry, tableEntryRefreshIntervalWait,
-                            needFetchAllRouteInfo, route);
+                        obPair = getTable(tableName, callback.getRowKey(), needRefreshTableEntry, tableEntryRefreshIntervalWait, false, route);
                     } else if (null != callback.getKeyRanges()) {
                         // using scan range
                         obPair = getTable(tableName, new ObTableQuery(),
@@ -853,7 +857,7 @@ private <T> T executeMutation(String tableName, MutationExecuteCallback<T> callb
                                     ((ObTableException) ex).getErrorCode(), ex.getMessage(),
                                     tryTimes);
                             if (ex instanceof ObTableNeedFetchAllException) {
-                                needFetchAllRouteInfo = true;
+                                getOrRefreshTableEntry(tableName, needRefreshTableEntry, isTableEntryRefreshIntervalWait(), true);
                                 // reset failure count while fetch all route info
                                 this.resetExecuteContinuousFailureCount(tableName);
                             }
@@ -1337,33 +1341,14 @@ public TableEntry refreshTableLocationByTabletId(TableEntry tableEntry, String t
             }
             long lastRefreshTime = tableEntry.getPartitionEntry().getPartitionInfo(tabletId).getLastUpdateTime();
             long currentTime = System.currentTimeMillis();
-            if (currentTime - lastRefreshTime < tableEntryRefreshLockTimeout) {
+            if (currentTime - lastRefreshTime < tableEntryRefreshIntervalCeiling) {
                 return tableEntry;
             }
-            
-            Lock lock = tableEntry.refreshLockMap.computeIfAbsent(tabletId, k -> new ReentrantLock());
-
-            if (!lock.tryLock(tableEntryRefreshLockTimeout, TimeUnit.MILLISECONDS)) {
-                String errMsg = String.format("Try to lock table-entry refreshing timeout. DataSource: %s, TableName: %s, Timeout: %d.",
-                        dataSourceName, tableName, tableEntryRefreshLockTimeout);
-                RUNTIME.error(errMsg);
-                throw new ObTableEntryRefreshException(errMsg);
-            }
+            tableEntry = loadTableEntryLocationWithPriority(serverRoster, tableEntryKey, tableEntry, tabletId,
+                    tableEntryAcquireConnectTimeout, tableEntryAcquireSocketTimeout,
+                    serverAddressPriorityTimeout, serverAddressCachingTimeout, sysUA);
 
-            try {
-                lastRefreshTime = tableEntry.getPartitionEntry().getPartitionInfo(tabletId).getLastUpdateTime();
-                currentTime = System.currentTimeMillis();
-                if (currentTime - lastRefreshTime < tableEntryRefreshLockTimeout) {
-                    return tableEntry;
-                }
-                tableEntry = loadTableEntryLocationWithPriority(serverRoster, tableEntryKey, tableEntry, tabletId,
-                        tableEntryAcquireConnectTimeout, tableEntryAcquireSocketTimeout,
-                        serverAddressPriorityTimeout, serverAddressCachingTimeout, sysUA);
-
-                tableEntry.prepareForWeakRead(serverRoster.getServerLdcLocation());
-            } finally {
-                lock.unlock();
-            }
+            tableEntry.prepareForWeakRead(serverRoster.getServerLdcLocation());
 
         } catch (ObTableNotExistException | ObTableServerCacheExpiredException e) {
             RUNTIME.error("RefreshTableEntry encountered an exception", e);
@@ -1676,7 +1661,9 @@ private ObPair<Long, ObTableParam> getTable(String tableName, Object[] rowKey,
         }
 
         long partId = getPartition(tableEntry, row); // partition id in 3.x, origin partId in 4.x, logicId
-
+        if (refresh) {
+            refreshTableLocationByTabletId(tableEntry, tableName, getTabletIdByPartId(tableEntry, partId));
+        }
         return getTableInternal(tableName, tableEntry, partId, waitForRefresh, route);
     }