Merge branch '3.0' into refact/query_opt

CMakeLists.txt

@@ -34,7 +34,7 @@ endif(${BUILD_TEST})
 
 add_subdirectory(source)
 add_subdirectory(tools)
-add_subdirectory(tests)
+add_subdirectory(utils)
 add_subdirectory(examples/c)
 
 # docs 

cmake/taostools_CMakeLists.txt.in

@@ -2,7 +2,7 @@
 # taos-tools
 ExternalProject_Add(taos-tools
         GIT_REPOSITORY https://github.com/taosdata/taos-tools.git
-        GIT_TAG f169c0f
+        GIT_TAG a4d9b92
         SOURCE_DIR "${TD_SOURCE_DIR}/tools/taos-tools"
         BINARY_DIR ""
         #BUILD_IN_SOURCE TRUE

cmake/taosws_CMakeLists.txt.in

@@ -2,7 +2,7 @@
 # taosws-rs
 ExternalProject_Add(taosws-rs
         GIT_REPOSITORY https://github.com/taosdata/taos-connector-rust.git
-        GIT_TAG b91b39c
+        GIT_TAG 6fc47d7
         SOURCE_DIR "${TD_SOURCE_DIR}/tools/taosws-rs"
         BINARY_DIR ""
         #BUILD_IN_SOURCE TRUE

docs/en/12-taos-sql/10-function.md

@@ -917,7 +917,7 @@ SELECT MAX(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **Return value type**:Same as the data type of the column being operated upon
 
-**Applicable data types**: Numeric
+**Applicable data types**: Numeric, Timestamp
 
 **Applicable table types**: standard tables and supertables
 
@@ -932,7 +932,7 @@ SELECT MIN(field_name) FROM {tb_name | stb_name} [WHERE clause];
 
 **Return value type**:Same as the data type of the column being operated upon
 
-**Applicable data types**: Numeric
+**Applicable data types**: Numeric, Timestamp
 
 **Applicable table types**: standard tables and supertables
 

docs/zh/07-develop/_sub_java.mdx

 ```java
 {{#include docs/examples/java/src/main/java/com/taos/example/SubscribeDemo.java}}
-{{#include docs/examples/java/src/main/java/com/taos/example/MetersDeserializer.java}}
-{{#include docs/examples/java/src/main/java/com/taos/example/Meters.java}}
 ```
 ```java
 {{#include docs/examples/java/src/main/java/com/taos/example/MetersDeserializer.java}}

include/common/taosdef.h

@@ -65,13 +65,6 @@ typedef enum {
   TSDB_STATIS_NONE = 1,  // statis part not exist
 } ETsdbStatisStatus;
 
-typedef enum {
-  TSDB_SMA_STAT_UNKNOWN = -1,  // unknown
-  TSDB_SMA_STAT_OK = 0,        // ready to provide service
-  TSDB_SMA_STAT_EXPIRED = 1,   // not ready or expired
-  TSDB_SMA_STAT_DROPPED = 2,   // sma dropped
-} ETsdbSmaStat;                // bit operation
-
 typedef enum {
   TSDB_SMA_TYPE_BLOCK = 0,       // Block-wise SMA
   TSDB_SMA_TYPE_TIME_RANGE = 1,  // Time-range-wise SMA

include/common/tcommon.h

@@ -184,7 +184,6 @@ typedef struct SQueryTableDataCond {
   STimeWindow  twindows;
   int64_t      startVersion;
   int64_t      endVersion;
-  int64_t      schemaVersion;
 } SQueryTableDataCond;
 
 int32_t tEncodeDataBlock(void** buf, const SSDataBlock* pBlock);

include/common/tdatablock.h

@@ -184,7 +184,8 @@ static FORCE_INLINE void colDataAppendDouble(SColumnInfoData* pColumnInfoData, u
 int32_t getJsonValueLen(const char* data);
 
 int32_t colDataAppend(SColumnInfoData* pColumnInfoData, uint32_t currentRow, const char* pData, bool isNull);
-int32_t colDataAppendNItems(SColumnInfoData* pColumnInfoData, uint32_t currentRow, const char* pData, uint32_t numOfRows);
+int32_t colDataAppendNItems(SColumnInfoData* pColumnInfoData, uint32_t currentRow, const char* pData,
+                            uint32_t numOfRows);
 int32_t colDataMergeCol(SColumnInfoData* pColumnInfoData, int32_t numOfRow1, int32_t* capacity,
                         const SColumnInfoData* pSource, int32_t numOfRow2);
 int32_t colDataAssign(SColumnInfoData* pColumnInfoData, const SColumnInfoData* pSource, int32_t numOfRows,
@@ -225,15 +226,16 @@ size_t blockDataGetCapacityInRow(const SSDataBlock* pBlock, size_t pageSize);
 int32_t blockDataTrimFirstNRows(SSDataBlock* pBlock, size_t n);
 int32_t blockDataKeepFirstNRows(SSDataBlock* pBlock, size_t n);
 
-int32_t      assignOneDataBlock(SSDataBlock* dst, const SSDataBlock* src);
-int32_t      copyDataBlock(SSDataBlock* dst, const SSDataBlock* src);
+int32_t assignOneDataBlock(SSDataBlock* dst, const SSDataBlock* src);
+int32_t copyDataBlock(SSDataBlock* dst, const SSDataBlock* src);
 
 SSDataBlock* createDataBlock();
 void*        blockDataDestroy(SSDataBlock* pBlock);
 void         blockDataFreeRes(SSDataBlock* pBlock);
 SSDataBlock* createOneDataBlock(const SSDataBlock* pDataBlock, bool copyData);
+SSDataBlock* createSpecialDataBlock(EStreamType type);
 
-int32_t      blockDataAppendColInfo(SSDataBlock* pBlock, SColumnInfoData* pColInfoData);
+int32_t blockDataAppendColInfo(SSDataBlock* pBlock, SColumnInfoData* pColInfoData);
 
 SColumnInfoData  createColumnInfoData(int16_t type, int32_t bytes, int16_t colId);
 SColumnInfoData* bdGetColumnInfoData(const SSDataBlock* pBlock, int32_t index);
@@ -249,7 +251,6 @@ char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** dumpBuf);
 int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SSDataBlock* pDataBlocks, STSchema* pTSchema, int32_t vgId,
                                     tb_uid_t suid);
 
-
 char* buildCtbNameByGroupId(const char* stbName, uint64_t groupId);
 
 static FORCE_INLINE int32_t blockGetEncodeSize(const SSDataBlock* pBlock) {

include/common/tmsg.h

@@ -785,6 +785,8 @@ typedef struct {
   int32_t walRollPeriod;
   int64_t walSegmentSize;
   int32_t sstTrigger;
+  int16_t hashPrefix;
+  int16_t hashSuffix;
 } SCreateDbReq;
 
 int32_t tSerializeSCreateDbReq(void* buf, int32_t bufLen, SCreateDbReq* pReq);
@@ -1068,6 +1070,7 @@ typedef struct {
 typedef struct {
   int32_t vgId;
   int32_t syncState;
+  int64_t cacheUsage;
   int64_t numOfTables;
   int64_t numOfTimeSeries;
   int64_t totalStorage;
@@ -1192,6 +1195,9 @@ typedef struct {
   int64_t  walRetentionSize;
   int32_t  walRollPeriod;
   int64_t  walSegmentSize;
+  int16_t  sstTrigger;
+  int16_t  hashPrefix;
+  int16_t  hashSuffix;
 } SCreateVnodeReq;
 
 int32_t tSerializeSCreateVnodeReq(void* buf, int32_t bufLen, SCreateVnodeReq* pReq);

include/common/ttokendef.h

@@ -103,224 +103,226 @@
 #define TK_WAL_ROLL_PERIOD      85
 #define TK_WAL_SEGMENT_SIZE     86
 #define TK_SST_TRIGGER          87
-#define TK_NK_COLON             88
-#define TK_TABLE                89
-#define TK_NK_LP                90
-#define TK_NK_RP                91
-#define TK_STABLE               92
-#define TK_ADD                  93
-#define TK_COLUMN               94
-#define TK_MODIFY               95
-#define TK_RENAME               96
-#define TK_TAG                  97
-#define TK_SET                  98
-#define TK_NK_EQ                99
-#define TK_USING                100
-#define TK_TAGS                 101
-#define TK_COMMENT              102
-#define TK_BOOL                 103
-#define TK_TINYINT              104
-#define TK_SMALLINT             105
-#define TK_INT                  106
-#define TK_INTEGER              107
-#define TK_BIGINT               108
-#define TK_FLOAT                109
-#define TK_DOUBLE               110
-#define TK_BINARY               111
-#define TK_TIMESTAMP            112
-#define TK_NCHAR                113
-#define TK_UNSIGNED             114
-#define TK_JSON                 115
-#define TK_VARCHAR              116
-#define TK_MEDIUMBLOB           117
-#define TK_BLOB                 118
-#define TK_VARBINARY            119
-#define TK_DECIMAL              120
-#define TK_MAX_DELAY            121
-#define TK_WATERMARK            122
-#define TK_ROLLUP               123
-#define TK_TTL                  124
-#define TK_SMA                  125
-#define TK_FIRST                126
-#define TK_LAST                 127
-#define TK_SHOW                 128
-#define TK_DATABASES            129
-#define TK_TABLES               130
-#define TK_STABLES              131
-#define TK_MNODES               132
-#define TK_MODULES              133
-#define TK_QNODES               134
-#define TK_FUNCTIONS            135
-#define TK_INDEXES              136
-#define TK_ACCOUNTS             137
-#define TK_APPS                 138
-#define TK_CONNECTIONS          139
-#define TK_LICENCES             140
-#define TK_GRANTS               141
-#define TK_QUERIES              142
-#define TK_SCORES               143
-#define TK_TOPICS               144
-#define TK_VARIABLES            145
-#define TK_BNODES               146
-#define TK_SNODES               147
-#define TK_CLUSTER              148
-#define TK_TRANSACTIONS         149
-#define TK_DISTRIBUTED          150
-#define TK_CONSUMERS            151
-#define TK_SUBSCRIPTIONS        152
-#define TK_VNODES               153
-#define TK_LIKE                 154
-#define TK_INDEX                155
-#define TK_FUNCTION             156
-#define TK_INTERVAL             157
-#define TK_TOPIC                158
-#define TK_AS                   159
-#define TK_WITH                 160
-#define TK_META                 161
-#define TK_CONSUMER             162
-#define TK_GROUP                163
-#define TK_DESC                 164
-#define TK_DESCRIBE             165
-#define TK_RESET                166
-#define TK_QUERY                167
-#define TK_CACHE                168
-#define TK_EXPLAIN              169
-#define TK_ANALYZE              170
-#define TK_VERBOSE              171
-#define TK_NK_BOOL              172
-#define TK_RATIO                173
-#define TK_NK_FLOAT             174
-#define TK_OUTPUTTYPE           175
-#define TK_AGGREGATE            176
-#define TK_BUFSIZE              177
-#define TK_STREAM               178
-#define TK_INTO                 179
-#define TK_TRIGGER              180
-#define TK_AT_ONCE              181
-#define TK_WINDOW_CLOSE         182
-#define TK_IGNORE               183
-#define TK_EXPIRED              184
-#define TK_KILL                 185
-#define TK_CONNECTION           186
-#define TK_TRANSACTION          187
-#define TK_BALANCE              188
-#define TK_VGROUP               189
-#define TK_MERGE                190
-#define TK_REDISTRIBUTE         191
-#define TK_SPLIT                192
-#define TK_DELETE               193
-#define TK_INSERT               194
-#define TK_NULL                 195
-#define TK_NK_QUESTION          196
-#define TK_NK_ARROW             197
-#define TK_ROWTS                198
-#define TK_TBNAME               199
-#define TK_QSTART               200
-#define TK_QEND                 201
-#define TK_QDURATION            202
-#define TK_WSTART               203
-#define TK_WEND                 204
-#define TK_WDURATION            205
-#define TK_CAST                 206
-#define TK_NOW                  207
-#define TK_TODAY                208
-#define TK_TIMEZONE             209
-#define TK_CLIENT_VERSION       210
-#define TK_SERVER_VERSION       211
-#define TK_SERVER_STATUS        212
-#define TK_CURRENT_USER         213
-#define TK_COUNT                214
-#define TK_LAST_ROW             215
-#define TK_BETWEEN              216
-#define TK_IS                   217
-#define TK_NK_LT                218
-#define TK_NK_GT                219
-#define TK_NK_LE                220
-#define TK_NK_GE                221
-#define TK_NK_NE                222
-#define TK_MATCH                223
-#define TK_NMATCH               224
-#define TK_CONTAINS             225
-#define TK_IN                   226
-#define TK_JOIN                 227
-#define TK_INNER                228
-#define TK_SELECT               229
-#define TK_DISTINCT             230
-#define TK_WHERE                231
-#define TK_PARTITION            232
-#define TK_BY                   233
-#define TK_SESSION              234
-#define TK_STATE_WINDOW         235
-#define TK_SLIDING              236
-#define TK_FILL                 237
-#define TK_VALUE                238
-#define TK_NONE                 239
-#define TK_PREV                 240
-#define TK_LINEAR               241
-#define TK_NEXT                 242
-#define TK_HAVING               243
-#define TK_RANGE                244
-#define TK_EVERY                245
-#define TK_ORDER                246
-#define TK_SLIMIT               247
-#define TK_SOFFSET              248
-#define TK_LIMIT                249
-#define TK_OFFSET               250
-#define TK_ASC                  251
-#define TK_NULLS                252
-#define TK_ABORT                253
-#define TK_AFTER                254
-#define TK_ATTACH               255
-#define TK_BEFORE               256
-#define TK_BEGIN                257
-#define TK_BITAND               258
-#define TK_BITNOT               259
-#define TK_BITOR                260
-#define TK_BLOCKS               261
-#define TK_CHANGE               262
-#define TK_COMMA                263
-#define TK_COMPACT              264
-#define TK_CONCAT               265
-#define TK_CONFLICT             266
-#define TK_COPY                 267
-#define TK_DEFERRED             268
-#define TK_DELIMITERS           269
-#define TK_DETACH               270
-#define TK_DIVIDE               271
-#define TK_DOT                  272
-#define TK_EACH                 273
-#define TK_END                  274
-#define TK_FAIL                 275
-#define TK_FILE                 276
-#define TK_FOR                  277
-#define TK_GLOB                 278
-#define TK_ID                   279
-#define TK_IMMEDIATE            280
-#define TK_IMPORT               281
-#define TK_INITIALLY            282
-#define TK_INSTEAD              283
-#define TK_ISNULL               284
-#define TK_KEY                  285
-#define TK_NK_BITNOT            286
-#define TK_NK_SEMI              287
-#define TK_NOTNULL              288
-#define TK_OF                   289
-#define TK_PLUS                 290
-#define TK_PRIVILEGE            291
-#define TK_RAISE                292
-#define TK_REPLACE              293
-#define TK_RESTRICT             294
-#define TK_ROW                  295
-#define TK_SEMI                 296
-#define TK_STAR                 297
-#define TK_STATEMENT            298
-#define TK_STRING               299
-#define TK_TIMES                300
-#define TK_UPDATE               301
-#define TK_VALUES               302
-#define TK_VARIABLE             303
-#define TK_VIEW                 304
-#define TK_WAL                  305
+#define TK_TABLE_PREFIX         88
+#define TK_TABLE_SUFFIX         89
+#define TK_NK_COLON             90
+#define TK_TABLE                91
+#define TK_NK_LP                92
+#define TK_NK_RP                93
+#define TK_STABLE               94
+#define TK_ADD                  95
+#define TK_COLUMN               96
+#define TK_MODIFY               97
+#define TK_RENAME               98
+#define TK_TAG                  99
+#define TK_SET                  100
+#define TK_NK_EQ                101
+#define TK_USING                102
+#define TK_TAGS                 103
+#define TK_COMMENT              104
+#define TK_BOOL                 105
+#define TK_TINYINT              106
+#define TK_SMALLINT             107
+#define TK_INT                  108
+#define TK_INTEGER              109
+#define TK_BIGINT               110
+#define TK_FLOAT                111
+#define TK_DOUBLE               112
+#define TK_BINARY               113
+#define TK_TIMESTAMP            114
+#define TK_NCHAR                115
+#define TK_UNSIGNED             116
+#define TK_JSON                 117
+#define TK_VARCHAR              118
+#define TK_MEDIUMBLOB           119
+#define TK_BLOB                 120
+#define TK_VARBINARY            121
+#define TK_DECIMAL              122
+#define TK_MAX_DELAY            123
+#define TK_WATERMARK            124
+#define TK_ROLLUP               125
+#define TK_TTL                  126
+#define TK_SMA                  127
+#define TK_FIRST                128
+#define TK_LAST                 129
+#define TK_SHOW                 130
+#define TK_DATABASES            131
+#define TK_TABLES               132
+#define TK_STABLES              133
+#define TK_MNODES               134
+#define TK_MODULES              135
+#define TK_QNODES               136
+#define TK_FUNCTIONS            137
+#define TK_INDEXES              138
+#define TK_ACCOUNTS             139
+#define TK_APPS                 140
+#define TK_CONNECTIONS          141
+#define TK_LICENCES             142
+#define TK_GRANTS               143
+#define TK_QUERIES              144
+#define TK_SCORES               145
+#define TK_TOPICS               146
+#define TK_VARIABLES            147
+#define TK_BNODES               148
+#define TK_SNODES               149
+#define TK_CLUSTER              150
+#define TK_TRANSACTIONS         151
+#define TK_DISTRIBUTED          152
+#define TK_CONSUMERS            153
+#define TK_SUBSCRIPTIONS        154
+#define TK_VNODES               155
+#define TK_LIKE                 156
+#define TK_INDEX                157
+#define TK_FUNCTION             158
+#define TK_INTERVAL             159
+#define TK_TOPIC                160
+#define TK_AS                   161
+#define TK_WITH                 162
+#define TK_META                 163
+#define TK_CONSUMER             164
+#define TK_GROUP                165
+#define TK_DESC                 166
+#define TK_DESCRIBE             167
+#define TK_RESET                168
+#define TK_QUERY                169
+#define TK_CACHE                170
+#define TK_EXPLAIN              171
+#define TK_ANALYZE              172
+#define TK_VERBOSE              173
+#define TK_NK_BOOL              174
+#define TK_RATIO                175
+#define TK_NK_FLOAT             176
+#define TK_OUTPUTTYPE           177
+#define TK_AGGREGATE            178
+#define TK_BUFSIZE              179
+#define TK_STREAM               180
+#define TK_INTO                 181
+#define TK_TRIGGER              182
+#define TK_AT_ONCE              183
+#define TK_WINDOW_CLOSE         184
+#define TK_IGNORE               185
+#define TK_EXPIRED              186
+#define TK_KILL                 187
+#define TK_CONNECTION           188
+#define TK_TRANSACTION          189
+#define TK_BALANCE              190
+#define TK_VGROUP               191
+#define TK_MERGE                192
+#define TK_REDISTRIBUTE         193
+#define TK_SPLIT                194
+#define TK_DELETE               195
+#define TK_INSERT               196
+#define TK_NULL                 197
+#define TK_NK_QUESTION          198
+#define TK_NK_ARROW             199
+#define TK_ROWTS                200
+#define TK_TBNAME               201
+#define TK_QSTART               202
+#define TK_QEND                 203
+#define TK_QDURATION            204
+#define TK_WSTART               205
+#define TK_WEND                 206
+#define TK_WDURATION            207
+#define TK_CAST                 208
+#define TK_NOW                  209
+#define TK_TODAY                210
+#define TK_TIMEZONE             211
+#define TK_CLIENT_VERSION       212
+#define TK_SERVER_VERSION       213
+#define TK_SERVER_STATUS        214
+#define TK_CURRENT_USER         215
+#define TK_COUNT                216
+#define TK_LAST_ROW             217
+#define TK_BETWEEN              218
+#define TK_IS                   219
+#define TK_NK_LT                220
+#define TK_NK_GT                221
+#define TK_NK_LE                222
+#define TK_NK_GE                223
+#define TK_NK_NE                224
+#define TK_MATCH                225
+#define TK_NMATCH               226
+#define TK_CONTAINS             227
+#define TK_IN                   228
+#define TK_JOIN                 229
+#define TK_INNER                230
+#define TK_SELECT               231
+#define TK_DISTINCT             232
+#define TK_WHERE                233
+#define TK_PARTITION            234
+#define TK_BY                   235
+#define TK_SESSION              236
+#define TK_STATE_WINDOW         237
+#define TK_SLIDING              238
+#define TK_FILL                 239
+#define TK_VALUE                240
+#define TK_NONE                 241
+#define TK_PREV                 242
+#define TK_LINEAR               243
+#define TK_NEXT                 244
+#define TK_HAVING               245
+#define TK_RANGE                246
+#define TK_EVERY                247
+#define TK_ORDER                248
+#define TK_SLIMIT               249
+#define TK_SOFFSET              250
+#define TK_LIMIT                251
+#define TK_OFFSET               252
+#define TK_ASC                  253
+#define TK_NULLS                254
+#define TK_ABORT                255
+#define TK_AFTER                256
+#define TK_ATTACH               257
+#define TK_BEFORE               258
+#define TK_BEGIN                259
+#define TK_BITAND               260
+#define TK_BITNOT               261
+#define TK_BITOR                262
+#define TK_BLOCKS               263
+#define TK_CHANGE               264
+#define TK_COMMA                265
+#define TK_COMPACT              266
+#define TK_CONCAT               267
+#define TK_CONFLICT             268
+#define TK_COPY                 269
+#define TK_DEFERRED             270
+#define TK_DELIMITERS           271
+#define TK_DETACH               272
+#define TK_DIVIDE               273
+#define TK_DOT                  274
+#define TK_EACH                 275
+#define TK_END                  276
+#define TK_FAIL                 277
+#define TK_FILE                 278
+#define TK_FOR                  279
+#define TK_GLOB                 280
+#define TK_ID                   281
+#define TK_IMMEDIATE            282
+#define TK_IMPORT               283
+#define TK_INITIALLY            284
+#define TK_INSTEAD              285
+#define TK_ISNULL               286
+#define TK_KEY                  287
+#define TK_NK_BITNOT            288
+#define TK_NK_SEMI              289
+#define TK_NOTNULL              290
+#define TK_OF                   291
+#define TK_PLUS                 292
+#define TK_PRIVILEGE            293
+#define TK_RAISE                294
+#define TK_REPLACE              295
+#define TK_RESTRICT             296
+#define TK_ROW                  297
+#define TK_SEMI                 298
+#define TK_STAR                 299
+#define TK_STATEMENT            300
+#define TK_STRING               301
+#define TK_TIMES                302
+#define TK_UPDATE               303
+#define TK_VALUES               304
+#define TK_VARIABLE             305
+#define TK_VIEW                 306
+#define TK_WAL                  307
 
 #define TK_NK_SPACE   300
 #define TK_NK_COMMENT 301

include/libs/nodes/cmdnodes.h

@@ -82,6 +82,8 @@ typedef struct SDatabaseOptions {
   bool        walRetentionSizeIsSet;
   bool        walRollPeriodIsSet;
   int32_t     sstTrigger;
+  int32_t     tablePrefix;
+  int32_t     tableSuffix;
 } SDatabaseOptions;
 
 typedef struct SCreateDatabaseStmt {

include/util/taoserror.h

@@ -616,6 +616,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_RSMA_FETCH_MSG_MSSED_UP        TAOS_DEF_ERROR_CODE(0, 0x3155)
 #define TSDB_CODE_RSMA_EMPTY_INFO                TAOS_DEF_ERROR_CODE(0, 0x3156)
 #define TSDB_CODE_RSMA_INVALID_SCHEMA            TAOS_DEF_ERROR_CODE(0, 0x3157)
+#define TSDB_CODE_RSMA_REGEX_MATCH               TAOS_DEF_ERROR_CODE(0, 0x3158)
 
 //index
 #define TSDB_CODE_INDEX_REBUILDING               TAOS_DEF_ERROR_CODE(0, 0x3200)

include/util/tdef.h

@@ -362,6 +362,12 @@ typedef enum ELogicConditionType {
 #define TSDB_MIN_SST_TRIGGER            1
 #define TSDB_MAX_SST_TRIGGER            128
 #define TSDB_DEFAULT_SST_TRIGGER        8
+#define TSDB_MIN_HASH_PREFIX            0
+#define TSDB_MAX_HASH_PREFIX            128
+#define TSDB_DEFAULT_HASH_PREFIX        0
+#define TSDB_MIN_HASH_SUFFIX            0
+#define TSDB_MAX_HASH_SUFFIX            128
+#define TSDB_DEFAULT_HASH_SUFFIX        0
 
 #define TSDB_DB_MIN_WAL_RETENTION_PERIOD -1
 #define TSDB_REP_DEF_DB_WAL_RET_PERIOD   0

include/util/trbtree.h

+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef _TD_UTIL_RBTREE_H_
+#define _TD_UTIL_RBTREE_H_
+
+#include "os.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct SRBTree     SRBTree;
+typedef struct SRBTreeNode SRBTreeNode;
+typedef struct SRBTreeIter SRBTreeIter;
+
+typedef int32_t (*tRBTreeCmprFn)(const void *, const void *);
+
+// SRBTree =============================================
+#define tRBTreeMin(T) ((T)->min == ((T)->NIL) ? NULL : (T)->min)
+#define tRBTreeMax(T) ((T)->max == ((T)->NIL) ? NULL : (T)->max)
+
+void         tRBTreeCreate(SRBTree *pTree, tRBTreeCmprFn cmprFn);
+SRBTreeNode *tRBTreePut(SRBTree *pTree, SRBTreeNode *z);
+void         tRBTreeDrop(SRBTree *pTree, SRBTreeNode *z);
+SRBTreeNode *tRBTreeDropByKey(SRBTree *pTree, void *pKey);
+SRBTreeNode *tRBTreeGet(SRBTree *pTree, void *pKey);
+
+// SRBTreeIter =============================================
+#define tRBTreeIterCreate(tree, ascend) \
+  (SRBTreeIter) { .asc = (ascend), .pTree = (tree), .pNode = (ascend) ? (tree)->min : (tree)->max }
+
+SRBTreeNode *tRBTreeIterNext(SRBTreeIter *pIter);
+
+// STRUCT =============================================
+typedef enum { RED, BLACK } ECOLOR;
+struct SRBTreeNode {
+  ECOLOR       color;
+  SRBTreeNode *parent;
+  SRBTreeNode *left;
+  SRBTreeNode *right;
+};
+
+#define RBTREE_NODE_PAYLOAD(N) ((const void *)&(N)[1])
+
+struct SRBTree {
+  tRBTreeCmprFn cmprFn;
+  int64_t       n;
+  SRBTreeNode  *root;
+  SRBTreeNode  *min;
+  SRBTreeNode  *max;
+  SRBTreeNode  *NIL;
+  SRBTreeNode   NILNODE;
+};
+
+struct SRBTreeIter {
+  int8_t       asc;
+  SRBTree     *pTree;
+  SRBTreeNode *pNode;
+};
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /*_TD_UTIL_RBTREE_H_*/
\ No newline at end of file

include/util/tutil.h

@@ -20,6 +20,7 @@
 #include "tcrc32c.h"
 #include "tdef.h"
 #include "tmd5.h"
+#include "thash.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -68,6 +69,8 @@ static FORCE_INLINE void taosEncryptPass_c(uint8_t *inBuf, size_t len, char *tar
   memcpy(target, buf, TSDB_PASSWORD_LEN);
 }
 
+#define taosGetTbHashVal(tbname, tblen, method, prefix, suffix) MurmurHash3_32((tbname), (tblen))
+
 #ifdef __cplusplus
 }
 #endif

source/client/src/clientImpl.c

@@ -1672,7 +1672,12 @@ static int32_t doConvertJson(SReqResultInfo* pResultInfo, int32_t numOfCols, int
       break;
     }
   }
-  if (!needConvert) return TSDB_CODE_SUCCESS;
+
+  if (!needConvert) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  tscDebug("start to convert form json format string");
 
   char*   p = (char*)pResultInfo->pData;
   int32_t dataLen = estimateJsonLen(pResultInfo, numOfCols, numOfRows);

source/client/src/taosx.c → source/client/src/clientRawBlockWrite.c

@@ -765,23 +765,25 @@ static int32_t taosCreateTable(TAOS* taos, void* meta, int32_t metaLen) {
     }
     taosArrayPush(pRequest->tableList, &pName);
 
+    pCreateReq->flags |= TD_CREATE_IF_NOT_EXISTS;
     // change tag cid to new cid
-    if(pCreateReq->type == TSDB_CHILD_TABLE){
+    if (pCreateReq->type == TSDB_CHILD_TABLE) {
       STableMeta* pTableMeta = NULL;
       SName       sName = {0};
       toName(pTscObj->acctId, pRequest->pDb, pCreateReq->ctb.name, &sName);
       code = catalogGetTableMeta(pCatalog, &conn, &sName, &pTableMeta);
-      if(code != TSDB_CODE_SUCCESS){
+      if (code != TSDB_CODE_SUCCESS) {
         uError("taosCreateTable:catalogGetTableMeta failed. table name: %s", pCreateReq->ctb.name);
         goto end;
       }
 
-      for(int32_t i = 0; i < taosArrayGetSize(pCreateReq->ctb.tagName); i++){
+      for (int32_t i = 0; i < taosArrayGetSize(pCreateReq->ctb.tagName); i++) {
         char* tName = taosArrayGet(pCreateReq->ctb.tagName, i);
-        for(int32_t j = pTableMeta->tableInfo.numOfColumns; j < pTableMeta->tableInfo.numOfColumns + pTableMeta->tableInfo.numOfTags; j++){
-          SSchema *tag = &pTableMeta->schema[j];
-          if(strcmp(tag->name, tName) == 0 && tag->type != TSDB_DATA_TYPE_JSON){
-            tTagSetCid((STag *)pCreateReq->ctb.pTag, i, tag->colId);
+        for (int32_t j = pTableMeta->tableInfo.numOfColumns;
+             j < pTableMeta->tableInfo.numOfColumns + pTableMeta->tableInfo.numOfTags; j++) {
+          SSchema* tag = &pTableMeta->schema[j];
+          if (strcmp(tag->name, tName) == 0 && tag->type != TSDB_DATA_TYPE_JSON) {
+            tTagSetCid((STag*)pCreateReq->ctb.pTag, i, tag->colId);
           }
         }
       }
@@ -1322,12 +1324,12 @@ end:
   return code;
 }
 
-static int32_t tmqWriteRaw(TAOS* taos, void* data, int32_t dataLen) {
-  int32_t   code = TSDB_CODE_SUCCESS;
-  SHashObj* pVgHash = NULL;
-  SQuery*   pQuery = NULL;
-  SMqRspObj rspObj = {0};
-  SDecoder  decoder = {0};
+static int32_t tmqWriteRawDataImpl(TAOS* taos, void* data, int32_t dataLen) {
+  int32_t     code = TSDB_CODE_SUCCESS;
+  SHashObj*   pVgHash = NULL;
+  SQuery*     pQuery = NULL;
+  SMqRspObj   rspObj = {0};
+  SDecoder    decoder = {0};
   STableMeta* pTableMeta = NULL;
 
   terrno = TSDB_CODE_SUCCESS;
@@ -1405,7 +1407,7 @@ static int32_t tmqWriteRaw(TAOS* taos, void* data, int32_t dataLen) {
     }
 
     code = catalogGetTableMeta(pCatalog, &conn, &pName, &pTableMeta);
-    if (code == TSDB_CODE_PAR_TABLE_NOT_EXIST){
+    if (code == TSDB_CODE_PAR_TABLE_NOT_EXIST) {
       uError("WriteRaw:catalogGetTableMeta table not exist. table name: %s", tbName);
       code = TSDB_CODE_SUCCESS;
       continue;
@@ -1466,7 +1468,7 @@ static int32_t tmqWriteRaw(TAOS* taos, void* data, int32_t dataLen) {
     }
 
     // pSW->pSchema should be same as pTableMeta->schema
-//    ASSERT(pSW->nCols == pTableMeta->tableInfo.numOfColumns);
+    //    ASSERT(pSW->nCols == pTableMeta->tableInfo.numOfColumns);
     uint64_t suid = (TSDB_NORMAL_TABLE == pTableMeta->tableType ? 0 : pTableMeta->suid);
     uint64_t uid = pTableMeta->uid;
     int16_t  sver = pTableMeta->sversion;
@@ -1494,10 +1496,10 @@ static int32_t tmqWriteRaw(TAOS* taos, void* data, int32_t dataLen) {
       int32_t offset = 0;
       for (int32_t k = 0; k < pTableMeta->tableInfo.numOfColumns; k++) {
         const SSchema* pColumn = &pTableMeta->schema[k];
-        int32_t* index = taosHashGet(schemaHash, pColumn->name, strlen(pColumn->name));
-        if(!index){
+        int32_t*       index = taosHashGet(schemaHash, pColumn->name, strlen(pColumn->name));
+        if (!index) {
           tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NULL, NULL, false, offset, k);
-        }else{
+        } else {
           char* colData = rspObj.resInfo.row[*index];
           if (!colData) {
             tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NULL, NULL, false, offset, k);
@@ -1668,7 +1670,7 @@ int32_t tmq_write_raw(TAOS* taos, tmq_raw_data raw) {
   } else if (raw.raw_type == TDMT_VND_DELETE) {
     return taosDeleteData(taos, raw.raw, raw.raw_len);
   } else if (raw.raw_type == RES_TYPE__TMQ) {
-    return tmqWriteRaw(taos, raw.raw, raw.raw_len);
+    return tmqWriteRawDataImpl(taos, raw.raw, raw.raw_len);
   }
   return TSDB_CODE_INVALID_PARA;
 }

source/client/src/clientSml.c

@@ -547,6 +547,8 @@ static int32_t smlModifyDBSchemas(SSmlHandle *info) {
         goto end;
       }
       needCheckMeta = true;
+      taosHashCleanup(hashTmp);
+      hashTmp = NULL;
     } else {
       uError("SML:0x%" PRIx64 " load table meta error: %s", info->id, tstrerror(code));
       goto end;
@@ -576,7 +578,6 @@ static int32_t smlModifyDBSchemas(SSmlHandle *info) {
     sTableData->tableMeta = pTableMeta;
 
     tableMetaSml = (SSmlSTableMeta **)taosHashIterate(info->superTables, tableMetaSml);
-    taosHashCleanup(hashTmp);
   }
   return 0;
 

source/common/src/systable.c

@@ -90,7 +90,7 @@ static const SSysDbTableSchema userDBSchema[] = {
     {.name = "minrows", .bytes = 4, .type = TSDB_DATA_TYPE_INT, .sysInfo = true},
     {.name = "maxrows", .bytes = 4, .type = TSDB_DATA_TYPE_INT, .sysInfo = true},
     {.name = "comp", .bytes = 1, .type = TSDB_DATA_TYPE_TINYINT, .sysInfo = true},
-    {.name = "precision", .bytes = 2 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR, .sysInfo = true},
+    {.name = "precision", .bytes = 2 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR, .sysInfo = false},
     {.name = "status", .bytes = 10 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR, .sysInfo = false},
     {.name = "retentions", .bytes = 60 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR, .sysInfo = true},
     {.name = "single_stable", .bytes = 1, .type = TSDB_DATA_TYPE_BOOL, .sysInfo = true},
@@ -102,6 +102,7 @@ static const SSysDbTableSchema userDBSchema[] = {
     {.name = "wal_retention_size", .bytes = 8, .type = TSDB_DATA_TYPE_BIGINT, .sysInfo = true},
     {.name = "wal_roll_period", .bytes = 4, .type = TSDB_DATA_TYPE_INT, .sysInfo = true},
     {.name = "wal_segment_size", .bytes = 8, .type = TSDB_DATA_TYPE_BIGINT, .sysInfo = true},
+    {.name = "sst_trigger", .bytes = 2, .type = TSDB_DATA_TYPE_SMALLINT, .sysInfo = true},
 };
 
 static const SSysDbTableSchema userFuncSchema[] = {
@@ -206,6 +207,7 @@ static const SSysDbTableSchema vgroupsSchema[] = {
     {.name = "v3_dnode", .bytes = 4, .type = TSDB_DATA_TYPE_INT, .sysInfo = true},
     {.name = "v3_status", .bytes = 10 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR, .sysInfo = true},
     {.name = "status", .bytes = 12 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR, .sysInfo = true},
+    {.name = "cacheload", .bytes = 4, .type = TSDB_DATA_TYPE_INT, .sysInfo = true},
     {.name = "nfiles", .bytes = 4, .type = TSDB_DATA_TYPE_INT, .sysInfo = true},
     {.name = "file_size", .bytes = 4, .type = TSDB_DATA_TYPE_INT, .sysInfo = true},
     {.name = "tsma", .bytes = 1, .type = TSDB_DATA_TYPE_TINYINT, .sysInfo = true},

source/common/src/tdatablock.c

@@ -140,7 +140,8 @@ int32_t colDataReserve(SColumnInfoData* pColumnInfoData, size_t newSize) {
   return TSDB_CODE_SUCCESS;
 }
 
-static void doCopyNItems(struct SColumnInfoData* pColumnInfoData, int32_t currentRow, const char* pData, int32_t itemLen, int32_t numOfRows) {
+static void doCopyNItems(struct SColumnInfoData* pColumnInfoData, int32_t currentRow, const char* pData,
+                         int32_t itemLen, int32_t numOfRows) {
   ASSERT(pColumnInfoData->info.bytes >= itemLen);
   size_t start = 1;
 
@@ -148,21 +149,23 @@ static void doCopyNItems(struct SColumnInfoData* pColumnInfoData, int32_t curren
   memcpy(pColumnInfoData->pData, pData, itemLen);
 
   int32_t t = 0;
-  int32_t count = log(numOfRows)/log(2);
-  while(t < count) {
+  int32_t count = log(numOfRows) / log(2);
+  while (t < count) {
     int32_t xlen = 1 << t;
-    memcpy(pColumnInfoData->pData + start * itemLen + pColumnInfoData->varmeta.length, pColumnInfoData->pData, xlen * itemLen);
+    memcpy(pColumnInfoData->pData + start * itemLen + pColumnInfoData->varmeta.length, pColumnInfoData->pData,
+           xlen * itemLen);
     t += 1;
     start += xlen;
   }
 
   // the tail part
   if (numOfRows > start) {
-    memcpy(pColumnInfoData->pData + start * itemLen + currentRow * itemLen, pColumnInfoData->pData, (numOfRows - start) * itemLen);
+    memcpy(pColumnInfoData->pData + start * itemLen + currentRow * itemLen, pColumnInfoData->pData,
+           (numOfRows - start) * itemLen);
   }
 
   if (IS_VAR_DATA_TYPE(pColumnInfoData->info.type)) {
-    for(int32_t i = 0; i < numOfRows; ++i) {
+    for (int32_t i = 0; i < numOfRows; ++i) {
       pColumnInfoData->varmeta.offset[i + currentRow] = pColumnInfoData->varmeta.length + i * itemLen;
     }
 
@@ -170,7 +173,8 @@ static void doCopyNItems(struct SColumnInfoData* pColumnInfoData, int32_t curren
   }
 }
 
-int32_t colDataAppendNItems(SColumnInfoData* pColumnInfoData, uint32_t currentRow, const char* pData, uint32_t numOfRows) {
+int32_t colDataAppendNItems(SColumnInfoData* pColumnInfoData, uint32_t currentRow, const char* pData,
+                            uint32_t numOfRows) {
   ASSERT(pData != NULL && pColumnInfoData != NULL);
 
   int32_t len = pColumnInfoData->info.bytes;
@@ -278,7 +282,7 @@ int32_t colDataMergeCol(SColumnInfoData* pColumnInfoData, int32_t numOfRow1, int
   } else {
     if (finalNumOfRows > *capacity || (numOfRow1 == 0 && pColumnInfoData->info.bytes != 0)) {
       // all data may be null, when the pColumnInfoData->info.type == 0, bytes == 0;
-//      ASSERT(finalNumOfRows * pColumnInfoData->info.bytes);
+      //      ASSERT(finalNumOfRows * pColumnInfoData->info.bytes);
       char* tmp = taosMemoryRealloc(pColumnInfoData->pData, finalNumOfRows * pColumnInfoData->info.bytes);
       if (tmp == NULL) {
         return TSDB_CODE_VND_OUT_OF_MEMORY;
@@ -557,7 +561,7 @@ int32_t blockDataToBuf(char* buf, const SSDataBlock* pBlock) {
 }
 
 int32_t blockDataFromBuf(SSDataBlock* pBlock, const char* buf) {
-  int32_t numOfRows = *(int32_t*) buf;
+  int32_t numOfRows = *(int32_t*)buf;
   blockDataEnsureCapacity(pBlock, numOfRows);
 
   pBlock->info.rows = numOfRows;
@@ -676,7 +680,8 @@ size_t blockDataGetRowSize(SSDataBlock* pBlock) {
  * @return
  */
 size_t blockDataGetSerialMetaSize(uint32_t numOfCols) {
-  // | version | total length | total rows | total columns | flag seg| block group id | column schema | each column length |
+  // | version | total length | total rows | total columns | flag seg| block group id | column schema | each column
+  // length |
   return sizeof(int32_t) + sizeof(int32_t) + sizeof(int32_t) + sizeof(int32_t) + sizeof(int32_t) + sizeof(uint64_t) +
          numOfCols * (sizeof(int8_t) + sizeof(int32_t)) + numOfCols * sizeof(int32_t);
 }
@@ -1302,6 +1307,40 @@ int32_t copyDataBlock(SSDataBlock* dst, const SSDataBlock* src) {
   return TSDB_CODE_SUCCESS;
 }
 
+SSDataBlock* createSpecialDataBlock(EStreamType type) {
+  SSDataBlock* pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));
+  pBlock->info.hasVarCol = false;
+  pBlock->info.groupId = 0;
+  pBlock->info.rows = 0;
+  pBlock->info.type = type;
+  pBlock->info.rowSize =
+      sizeof(TSKEY) + sizeof(TSKEY) + sizeof(uint64_t) + sizeof(uint64_t) + sizeof(TSKEY) + sizeof(TSKEY);
+  pBlock->info.watermark = INT64_MIN;
+
+  pBlock->pDataBlock = taosArrayInit(6, sizeof(SColumnInfoData));
+  SColumnInfoData infoData = {0};
+  infoData.info.type = TSDB_DATA_TYPE_TIMESTAMP;
+  infoData.info.bytes = sizeof(TSKEY);
+  // window start ts
+  taosArrayPush(pBlock->pDataBlock, &infoData);
+  // window end ts
+  taosArrayPush(pBlock->pDataBlock, &infoData);
+
+  infoData.info.type = TSDB_DATA_TYPE_UBIGINT;
+  infoData.info.bytes = sizeof(uint64_t);
+  // uid
+  taosArrayPush(pBlock->pDataBlock, &infoData);
+  // group id
+  taosArrayPush(pBlock->pDataBlock, &infoData);
+
+  // calculate start ts
+  taosArrayPush(pBlock->pDataBlock, &infoData);
+  // calculate end ts
+  taosArrayPush(pBlock->pDataBlock, &infoData);
+
+  return pBlock;
+}
+
 SSDataBlock* createOneDataBlock(const SSDataBlock* pDataBlock, bool copyData) {
   if (pDataBlock == NULL) {
     return NULL;
@@ -1426,7 +1465,7 @@ size_t blockDataGetCapacityInRow(const SSDataBlock* pBlock, size_t pageSize) {
 }
 
 void colDataDestroy(SColumnInfoData* pColData) {
-  if(!pColData) return;
+  if (!pColData) return;
   if (IS_VAR_DATA_TYPE(pColData->info.type)) {
     taosMemoryFreeClear(pColData->varmeta.offset);
   } else {
@@ -1693,7 +1732,7 @@ static char* formatTimestamp(char* buf, int64_t val, int precision) {
   }
   struct tm ptm = {0};
   taosLocalTime(&tt, &ptm);
-  size_t     pos = strftime(buf, 35, "%Y-%m-%d %H:%M:%S", &ptm);
+  size_t pos = strftime(buf, 35, "%Y-%m-%d %H:%M:%S", &ptm);
 
   if (precision == TSDB_TIME_PRECISION_NANO) {
     sprintf(buf + pos, ".%09d", ms);
@@ -1847,20 +1886,20 @@ char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** pDataBuf)
           break;
         case TSDB_DATA_TYPE_VARCHAR: {
           memset(pBuf, 0, sizeof(pBuf));
-          char* pData = colDataGetVarData(pColInfoData, j);
+          char*   pData = colDataGetVarData(pColInfoData, j);
           int32_t dataSize = TMIN(sizeof(pBuf), varDataLen(pData));
           memcpy(pBuf, varDataVal(pData), dataSize);
           len += snprintf(dumpBuf + len, size - len, " %15s |", pBuf);
           if (len >= size - 1) return dumpBuf;
-          } break;
+        } break;
         case TSDB_DATA_TYPE_NCHAR: {
-          char* pData = colDataGetVarData(pColInfoData, j);
+          char*   pData = colDataGetVarData(pColInfoData, j);
           int32_t dataSize = TMIN(sizeof(pBuf), varDataLen(pData));
           memset(pBuf, 0, sizeof(pBuf));
-          taosUcs4ToMbs((TdUcs4 *)varDataVal(pData), dataSize, pBuf);
+          taosUcs4ToMbs((TdUcs4*)varDataVal(pData), dataSize, pBuf);
           len += snprintf(dumpBuf + len, size - len, " %15s |", pBuf);
           if (len >= size - 1) return dumpBuf;
-          } break;
+        } break;
       }
     }
     len += snprintf(dumpBuf + len, size - len, "\n");
@@ -1877,7 +1916,7 @@ char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** pDataBuf)
  * @param pDataBlocks
  * @param vgId
  * @param suid
- * 
+ *
  */
 int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SSDataBlock* pDataBlock, STSchema* pTSchema, int32_t vgId,
                                     tb_uid_t suid) {
@@ -1904,8 +1943,8 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SSDataBlock* pDataB
   tdSRowInit(&rb, pTSchema->version);
 
   for (int32_t i = 0; i < sz; ++i) {
-    int32_t      colNum = taosArrayGetSize(pDataBlock->pDataBlock);
-    int32_t      rows = pDataBlock->info.rows;
+    int32_t colNum = taosArrayGetSize(pDataBlock->pDataBlock);
+    int32_t rows = pDataBlock->info.rows;
     //    int32_t      rowSize = pDataBlock->info.rowSize;
     //    int64_t      groupId = pDataBlock->info.groupId;
 
@@ -1926,7 +1965,7 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SSDataBlock* pDataB
 
     msgLen += sizeof(SSubmitBlk);
     int32_t dataLen = 0;
-    for (int32_t j = 0; j < rows; ++j) {                     // iterate by row
+    for (int32_t j = 0; j < rows; ++j) {                               // iterate by row
       tdSRowResetBuf(&rb, POINTER_SHIFT(pDataBuf, msgLen + dataLen));  // set row buf
       bool    isStartKey = false;
       int32_t offset = 0;
@@ -2081,6 +2120,7 @@ void blockEncode(const SSDataBlock* pBlock, char* data, int32_t* dataLen, int32_
   int32_t* rows = (int32_t*)data;
   *rows = pBlock->info.rows;
   data += sizeof(int32_t);
+  ASSERT(*rows > 0);
 
   int32_t* cols = (int32_t*)data;
   *cols = numOfCols;
@@ -2089,7 +2129,7 @@ void blockEncode(const SSDataBlock* pBlock, char* data, int32_t* dataLen, int32_
   // flag segment.
   // the inital bit is for column info
   int32_t* flagSegment = (int32_t*)data;
-  *flagSegment = (1<<31);
+  *flagSegment = (1 << 31);
 
   data += sizeof(int32_t);
 
@@ -2144,12 +2184,14 @@ void blockEncode(const SSDataBlock* pBlock, char* data, int32_t* dataLen, int32_
 
   *actualLen = *dataLen;
   *groupId = pBlock->info.groupId;
+  ASSERT(*dataLen > 0);
+  uDebug("build data block, actualLen:%d, rows:%d, cols:%d", *dataLen, *rows, *cols);
 }
 
 const char* blockDecode(SSDataBlock* pBlock, const char* pData) {
   const char* pStart = pData;
 
-  int32_t version = *(int32_t*) pStart;
+  int32_t version = *(int32_t*)pStart;
   pStart += sizeof(int32_t);
   ASSERT(version == 1);
 
@@ -2158,7 +2200,7 @@ const char* blockDecode(SSDataBlock* pBlock, const char* pData) {
   pStart += sizeof(int32_t);
 
   // total rows sizeof(int32_t)
-  int32_t numOfRows =  *(int32_t*)pStart;
+  int32_t numOfRows = *(int32_t*)pStart;
   pStart += sizeof(int32_t);
 
   // total columns sizeof(int32_t)

source/common/src/tglobal.c

@@ -129,10 +129,6 @@ int32_t tsMinIntervalTime = 1;
 int32_t tsQueryBufferSize = -1;
 int64_t tsQueryBufferSizeBytes = -1;
 
-// tsdb config
-// For backward compatibility
-bool tsdbForceKeepFile = false;
-
 int32_t  tsDiskCfgNum = 0;
 SDiskCfg tsDiskCfg[TFS_MAX_DISKS] = {0};
 

source/common/src/tmsg.c

@@ -994,6 +994,7 @@ int32_t tSerializeSStatusReq(void *buf, int32_t bufLen, SStatusReq *pReq) {
     SVnodeLoad *pload = taosArrayGet(pReq->pVloads, i);
     if (tEncodeI32(&encoder, pload->vgId) < 0) return -1;
     if (tEncodeI32(&encoder, pload->syncState) < 0) return -1;
+    if (tEncodeI64(&encoder, pload->cacheUsage) < 0) return -1;
     if (tEncodeI64(&encoder, pload->numOfTables) < 0) return -1;
     if (tEncodeI64(&encoder, pload->numOfTimeSeries) < 0) return -1;
     if (tEncodeI64(&encoder, pload->totalStorage) < 0) return -1;
@@ -1063,6 +1064,7 @@ int32_t tDeserializeSStatusReq(void *buf, int32_t bufLen, SStatusReq *pReq) {
     SVnodeLoad vload = {0};
     if (tDecodeI32(&decoder, &vload.vgId) < 0) return -1;
     if (tDecodeI32(&decoder, &vload.syncState) < 0) return -1;
+    if (tDecodeI64(&decoder, &vload.cacheUsage) < 0) return -1;
     if (tDecodeI64(&decoder, &vload.numOfTables) < 0) return -1;
     if (tDecodeI64(&decoder, &vload.numOfTimeSeries) < 0) return -1;
     if (tDecodeI64(&decoder, &vload.totalStorage) < 0) return -1;
@@ -2025,6 +2027,8 @@ int32_t tSerializeSCreateDbReq(void *buf, int32_t bufLen, SCreateDbReq *pReq) {
   if (tEncodeI32(&encoder, pReq->walRollPeriod) < 0) return -1;
   if (tEncodeI64(&encoder, pReq->walSegmentSize) < 0) return -1;
   if (tEncodeI32(&encoder, pReq->sstTrigger) < 0) return -1;
+  if (tEncodeI16(&encoder, pReq->hashPrefix) < 0) return -1;
+  if (tEncodeI16(&encoder, pReq->hashSuffix) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->ignoreExist) < 0) return -1;
   if (tEncodeI32(&encoder, pReq->numOfRetensions) < 0) return -1;
   for (int32_t i = 0; i < pReq->numOfRetensions; ++i) {
@@ -2072,6 +2076,8 @@ int32_t tDeserializeSCreateDbReq(void *buf, int32_t bufLen, SCreateDbReq *pReq)
   if (tDecodeI32(&decoder, &pReq->walRollPeriod) < 0) return -1;
   if (tDecodeI64(&decoder, &pReq->walSegmentSize) < 0) return -1;
   if (tDecodeI32(&decoder, &pReq->sstTrigger) < 0) return -1;
+  if (tDecodeI16(&decoder, &pReq->hashPrefix) < 0) return -1;
+  if (tDecodeI16(&decoder, &pReq->hashSuffix) < 0) return -1;
   if (tDecodeI8(&decoder, &pReq->ignoreExist) < 0) return -1;
   if (tDecodeI32(&decoder, &pReq->numOfRetensions) < 0) return -1;
   pReq->pRetensions = taosArrayInit(pReq->numOfRetensions, sizeof(SRetention));
@@ -3766,6 +3772,9 @@ int32_t tSerializeSCreateVnodeReq(void *buf, int32_t bufLen, SCreateVnodeReq *pR
   if (tEncodeI64(&encoder, pReq->walRetentionSize) < 0) return -1;
   if (tEncodeI32(&encoder, pReq->walRollPeriod) < 0) return -1;
   if (tEncodeI64(&encoder, pReq->walSegmentSize) < 0) return -1;
+  if (tEncodeI16(&encoder, pReq->sstTrigger) < 0) return -1;
+  if (tEncodeI16(&encoder, pReq->hashPrefix) < 0) return -1;
+  if (tEncodeI16(&encoder, pReq->hashSuffix) < 0) return -1;
 
   tEndEncode(&encoder);
 
@@ -3838,6 +3847,9 @@ int32_t tDeserializeSCreateVnodeReq(void *buf, int32_t bufLen, SCreateVnodeReq *
   if (tDecodeI64(&decoder, &pReq->walRetentionSize) < 0) return -1;
   if (tDecodeI32(&decoder, &pReq->walRollPeriod) < 0) return -1;
   if (tDecodeI64(&decoder, &pReq->walSegmentSize) < 0) return -1;
+  if (tDecodeI16(&decoder, &pReq->sstTrigger) < 0) return -1;
+  if (tDecodeI16(&decoder, &pReq->hashPrefix) < 0) return -1;
+  if (tDecodeI16(&decoder, &pReq->hashSuffix) < 0) return -1;
 
   tEndDecode(&decoder);
   tDecoderClear(&decoder);

source/common/src/trow.c

@@ -538,12 +538,12 @@ bool tdSTSRowIterGetTpVal(STSRowIter *pIter, col_type_t colType, int32_t offset,
     } else {
       pVal->val = POINTER_SHIFT(TD_ROW_DATA(pRow), offset);
     }
-    return TSDB_CODE_SUCCESS;
+    return true;
   }
 
   if (tdGetBitmapValType(pIter->pBitmap, pIter->colIdx - 1, &pVal->valType, 0) != TSDB_CODE_SUCCESS) {
     pVal->valType = TD_VTYPE_NONE;
-    return terrno;
+    return true;
   }
 
   if (pVal->valType == TD_VTYPE_NORM) {

source/dnode/mgmt/mgmt_vnode/src/vmHandle.c

@@ -167,9 +167,12 @@ static void vmGenerateVnodeCfg(SCreateVnodeReq *pCreate, SVnodeCfg *pCfg) {
   pCfg->walCfg.segSize = pCreate->walSegmentSize;
   pCfg->walCfg.level = pCreate->walLevel;
 
+  pCfg->sstTrigger = pCreate->sstTrigger;
   pCfg->hashBegin = pCreate->hashBegin;
   pCfg->hashEnd = pCreate->hashEnd;
   pCfg->hashMethod = pCreate->hashMethod;
+  pCfg->hashPrefix = pCreate->hashPrefix;
+  pCfg->hashSuffix = pCreate->hashSuffix;
 
   pCfg->standby = pCfg->standby;
   pCfg->syncCfg.myIndex = pCreate->selfIndex;
@@ -219,8 +222,11 @@ int32_t vmProcessCreateVnodeReq(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) {
     return -1;
   }
 
-  dDebug("vgId:%d, start to create vnode, tsma:%d standby:%d cacheLast:%d cacheLastSize:%d", createReq.vgId,
-         createReq.isTsma, createReq.standby, createReq.cacheLast, createReq.cacheLastSize);
+  dInfo("vgId:%d, start to create vnode, tsma:%d standby:%d cacheLast:%d cacheLastSize:%d sstTrigger:%d",
+         createReq.vgId, createReq.isTsma, createReq.standby, createReq.cacheLast, createReq.cacheLastSize,
+         createReq.sstTrigger);
+  dInfo("vgId:%d, hashMethod:%d begin:%u end:%u prefix:%d surfix:%d", createReq.vgId, createReq.hashMethod,
+         createReq.hashBegin, createReq.hashEnd, createReq.hashPrefix, createReq.hashSuffix);
   vmGenerateVnodeCfg(&createReq, &vnodeCfg);
 
   if (vmTsmaAdjustDays(&vnodeCfg, &createReq) < 0) {

source/dnode/mnode/impl/inc/mndDef.h

@@ -305,11 +305,14 @@ typedef struct {
   int8_t  hashMethod;  // default is 1
   int8_t  cacheLast;
   int8_t  schemaless;
+  int16_t hashPrefix;
+  int16_t hashSuffix;
+  int16_t sstTrigger;
   int32_t numOfRetensions;
   SArray* pRetensions;
   int32_t walRetentionPeriod;
-  int64_t walRetentionSize;
   int32_t walRollPeriod;
+  int64_t walRetentionSize;
   int64_t walSegmentSize;
 } SDbCfg;
 
@@ -340,6 +343,7 @@ typedef struct {
   uint32_t  hashEnd;
   char      dbName[TSDB_DB_FNAME_LEN];
   int64_t   dbUid;
+  int64_t   cacheUsage;
   int64_t   numOfTables;
   int64_t   numOfTimeSeries;
   int64_t   totalStorage;

source/dnode/mnode/impl/src/mndDb.c

@@ -30,7 +30,7 @@
 #include "systable.h"
 
 #define DB_VER_NUMBER   1
-#define DB_RESERVE_SIZE 64
+#define DB_RESERVE_SIZE 58
 
 static SSdbRaw *mndDbActionEncode(SDbObj *pDb);
 static SSdbRow *mndDbActionDecode(SSdbRaw *pRaw);
@@ -124,6 +124,9 @@ static SSdbRaw *mndDbActionEncode(SDbObj *pDb) {
   SDB_SET_INT64(pRaw, dataPos, pDb->cfg.walRetentionSize, _OVER)
   SDB_SET_INT32(pRaw, dataPos, pDb->cfg.walRollPeriod, _OVER)
   SDB_SET_INT64(pRaw, dataPos, pDb->cfg.walSegmentSize, _OVER)
+  SDB_SET_INT16(pRaw, dataPos, pDb->cfg.sstTrigger, _OVER)
+  SDB_SET_INT16(pRaw, dataPos, pDb->cfg.hashPrefix, _OVER)
+  SDB_SET_INT16(pRaw, dataPos, pDb->cfg.hashSuffix, _OVER)
 
   SDB_SET_RESERVE(pRaw, dataPos, DB_RESERVE_SIZE, _OVER)
   SDB_SET_DATALEN(pRaw, dataPos, _OVER)
@@ -207,6 +210,9 @@ static SSdbRow *mndDbActionDecode(SSdbRaw *pRaw) {
   SDB_GET_INT64(pRaw, dataPos, &pDb->cfg.walRetentionSize, _OVER)
   SDB_GET_INT32(pRaw, dataPos, &pDb->cfg.walRollPeriod, _OVER)
   SDB_GET_INT64(pRaw, dataPos, &pDb->cfg.walSegmentSize, _OVER)
+  SDB_GET_INT16(pRaw, dataPos, &pDb->cfg.sstTrigger, _OVER)
+  SDB_GET_INT16(pRaw, dataPos, &pDb->cfg.hashPrefix, _OVER)
+  SDB_GET_INT16(pRaw, dataPos, &pDb->cfg.hashSuffix, _OVER)
 
   SDB_GET_RESERVE(pRaw, dataPos, DB_RESERVE_SIZE, _OVER)
   taosInitRWLatch(&pDb->lock);
@@ -254,6 +260,7 @@ static int32_t mndDbActionUpdate(SSdb *pSdb, SDbObj *pOld, SDbObj *pNew) {
   pOld->cfg.strict = pNew->cfg.strict;
   pOld->cfg.cacheLast = pNew->cfg.cacheLast;
   pOld->cfg.replications = pNew->cfg.replications;
+  pOld->cfg.sstTrigger = pNew->cfg.sstTrigger;
   taosWUnLockLatch(&pOld->lock);
   return 0;
 }
@@ -330,6 +337,9 @@ static int32_t mndCheckDbCfg(SMnode *pMnode, SDbCfg *pCfg) {
   if (pCfg->walRetentionSize < TSDB_DB_MIN_WAL_RETENTION_SIZE) return -1;
   if (pCfg->walRollPeriod < TSDB_DB_MIN_WAL_ROLL_PERIOD) return -1;
   if (pCfg->walSegmentSize < TSDB_DB_MIN_WAL_SEGMENT_SIZE) return -1;
+  if (pCfg->sstTrigger < TSDB_MIN_SST_TRIGGER || pCfg->sstTrigger > TSDB_MAX_SST_TRIGGER) return -1;
+  if (pCfg->hashPrefix < TSDB_MIN_HASH_PREFIX || pCfg->hashPrefix > TSDB_MAX_HASH_PREFIX) return -1;
+  if (pCfg->hashSuffix < TSDB_MIN_HASH_SUFFIX || pCfg->hashSuffix > TSDB_MAX_HASH_SUFFIX) return -1;
 
   terrno = 0;
   return terrno;
@@ -363,6 +373,9 @@ static void mndSetDefaultDbCfg(SDbCfg *pCfg) {
     pCfg->walRetentionSize = TSDB_REPS_DEF_DB_WAL_RET_SIZE;
   if (pCfg->walRollPeriod < 0) pCfg->walRollPeriod = TSDB_REPS_DEF_DB_WAL_ROLL_PERIOD;
   if (pCfg->walSegmentSize < 0) pCfg->walSegmentSize = TSDB_DEFAULT_DB_WAL_SEGMENT_SIZE;
+  if (pCfg->sstTrigger <= 0) pCfg->sstTrigger = TSDB_DEFAULT_SST_TRIGGER;
+  if (pCfg->hashPrefix < 0) pCfg->hashPrefix = TSDB_DEFAULT_HASH_PREFIX;
+  if (pCfg->hashSuffix < 0) pCfg->hashSuffix = TSDB_DEFAULT_HASH_SUFFIX;
 }
 
 static int32_t mndSetCreateDbRedoLogs(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroups) {
@@ -479,6 +492,9 @@ static int32_t mndCreateDb(SMnode *pMnode, SRpcMsg *pReq, SCreateDbReq *pCreate,
       .walRetentionSize = pCreate->walRetentionSize,
       .walRollPeriod = pCreate->walRollPeriod,
       .walSegmentSize = pCreate->walSegmentSize,
+      .sstTrigger = pCreate->sstTrigger,
+      .hashPrefix = pCreate->hashPrefix,
+      .hashSuffix = pCreate->hashSuffix,
   };
 
   dbObj.cfg.numOfRetensions = pCreate->numOfRetensions;
@@ -1536,6 +1552,24 @@ static void mndDumpDbInfoData(SMnode *pMnode, SSDataBlock *pBlock, SDbObj *pDb,
     STR_WITH_MAXSIZE_TO_VARSTR(buf, "NULL", bytes);
   }
 
+  const char *precStr = NULL;
+  switch (pDb->cfg.precision) {
+    case TSDB_TIME_PRECISION_MILLI:
+      precStr = TSDB_TIME_PRECISION_MILLI_STR;
+      break;
+    case TSDB_TIME_PRECISION_MICRO:
+      precStr = TSDB_TIME_PRECISION_MICRO_STR;
+      break;
+    case TSDB_TIME_PRECISION_NANO:
+      precStr = TSDB_TIME_PRECISION_NANO_STR;
+      break;
+    default:
+      precStr = "none";
+      break;
+  }
+  char precVstr[10] = {0};
+  STR_WITH_SIZE_TO_VARSTR(precVstr, precStr, 2);
+
   char *statusStr = "ready";
   if (objStatus == SDB_STATUS_CREATING) {
     statusStr = "creating";
@@ -1546,7 +1580,6 @@ static void mndDumpDbInfoData(SMnode *pMnode, SSDataBlock *pBlock, SDbObj *pDb,
       statusStr = "unsynced";
     }
   }
-
   char statusVstr[24] = {0};
   STR_WITH_SIZE_TO_VARSTR(statusVstr, statusStr, strlen(statusStr));
 
@@ -1557,6 +1590,8 @@ static void mndDumpDbInfoData(SMnode *pMnode, SSDataBlock *pBlock, SDbObj *pDb,
         colDataAppend(pColInfo, rows, buf, false);
       } else if (i == 3) {
         colDataAppend(pColInfo, rows, (const char *)&numOfTables, false);
+      } else if (i == 14) {
+        colDataAppend(pColInfo, rows, precVstr, false);
       } else if (i == 15) {
         colDataAppend(pColInfo, rows, statusVstr, false);
       } else {
@@ -1621,23 +1656,6 @@ static void mndDumpDbInfoData(SMnode *pMnode, SSDataBlock *pBlock, SDbObj *pDb,
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, rows, (const char *)&pDb->cfg.compression, false);
 
-    const char *precStr = NULL;
-    switch (pDb->cfg.precision) {
-      case TSDB_TIME_PRECISION_MILLI:
-        precStr = TSDB_TIME_PRECISION_MILLI_STR;
-        break;
-      case TSDB_TIME_PRECISION_MICRO:
-        precStr = TSDB_TIME_PRECISION_MICRO_STR;
-        break;
-      case TSDB_TIME_PRECISION_NANO:
-        precStr = TSDB_TIME_PRECISION_NANO_STR;
-        break;
-      default:
-        precStr = "none";
-        break;
-    }
-    char precVstr[10] = {0};
-    STR_WITH_SIZE_TO_VARSTR(precVstr, precStr, 2);
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, rows, (const char *)precVstr, false);
 
@@ -1682,6 +1700,9 @@ static void mndDumpDbInfoData(SMnode *pMnode, SSDataBlock *pBlock, SDbObj *pDb,
 
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, rows, (const char *)&pDb->cfg.walSegmentSize, false);
+
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, rows, (const char *)&pDb->cfg.sstTrigger, false);
   }
 
   taosMemoryFree(buf);

source/dnode/mnode/impl/src/mndDnode.c

@@ -347,6 +347,7 @@ static int32_t mndProcessStatusReq(SRpcMsg *pReq) {
     SVgObj *pVgroup = mndAcquireVgroup(pMnode, pVload->vgId);
     if (pVgroup != NULL) {
       if (pVload->syncState == TAOS_SYNC_STATE_LEADER) {
+        pVgroup->cacheUsage = pVload->cacheUsage;
         pVgroup->numOfTables = pVload->numOfTables;
         pVgroup->numOfTimeSeries = pVload->numOfTimeSeries;
         pVgroup->totalStorage = pVload->totalStorage;
@@ -853,8 +854,8 @@ static int32_t mndProcessConfigDnodeReq(SRpcMsg *pReq) {
   }
 
   int32_t code = -1;
-  SSdb *pSdb = pMnode->pSdb;
-  void *pIter = NULL;
+  SSdb   *pSdb = pMnode->pSdb;
+  void   *pIter = NULL;
   while (1) {
     SDnodeObj *pDnode = NULL;
     pIter = sdbFetch(pSdb, SDB_DNODE, pIter, (void **)&pDnode);
@@ -877,7 +878,7 @@ static int32_t mndProcessConfigDnodeReq(SRpcMsg *pReq) {
 
     sdbRelease(pSdb, pDnode);
   }
-  
+
   if (code == -1) {
     terrno = TSDB_CODE_MND_DNODE_NOT_EXIST;
   }

source/dnode/mnode/impl/src/mndVgroup.c

@@ -234,6 +234,9 @@ void *mndBuildCreateVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVg
   createReq.walRetentionSize = pDb->cfg.walRetentionSize;
   createReq.walRollPeriod = pDb->cfg.walRollPeriod;
   createReq.walSegmentSize = pDb->cfg.walSegmentSize;
+  createReq.sstTrigger = pDb->cfg.sstTrigger;
+  createReq.hashPrefix = pDb->cfg.hashPrefix;
+  createReq.hashSuffix = pDb->cfg.hashSuffix;
 
   for (int32_t v = 0; v < pVgroup->replica; ++v) {
     SReplica  *pReplica = &createReq.replicas[v];
@@ -693,6 +696,9 @@ static int32_t mndRetrieveVgroups(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *p
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppendNULL(pColInfo, numOfRows);
 
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)&pVgroup->cacheUsage, false);
+
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppendNULL(pColInfo, numOfRows);
 

source/dnode/vnode/CMakeLists.txt

@@ -29,6 +29,7 @@ target_sources(
     # sma
     "src/sma/smaEnv.c"
     "src/sma/smaUtil.c"
+    "src/sma/smaFS.c"
     "src/sma/smaOpen.c"
     "src/sma/smaCommit.c"
     "src/sma/smaRollup.c"
@@ -49,6 +50,10 @@ target_sources(
     "src/tsdb/tsdbSnapshot.c"
     "src/tsdb/tsdbCacheRead.c"
     "src/tsdb/tsdbRetention.c"
+    "src/tsdb/tsdbDiskData.c"
+    "src/tsdb/tsdbCompress.c"
+    "src/tsdb/tsdbCompact.c"
+    "src/tsdb/tsdbMergeTree.c"
 
     # tq
     "src/tq/tq.c"

source/dnode/vnode/inc/vnode.h

@@ -155,6 +155,7 @@ int32_t tsdbGetTableSchema(SVnode *pVnode, int64_t uid, STSchema **pSchema, int6
 
 void   tsdbCacheSetCapacity(SVnode *pVnode, size_t capacity);
 size_t tsdbCacheGetCapacity(SVnode *pVnode);
+size_t tsdbCacheGetUsage(SVnode *pVnode);
 
 // tq
 typedef struct SMetaTableInfo {
@@ -287,6 +288,9 @@ struct SVnodeCfg {
   SVnodeStats vndStats;
   uint32_t    hashBegin;
   uint32_t    hashEnd;
+  int16_t     sstTrigger;
+  int16_t     hashPrefix;
+  int16_t     hashSuffix;
 };
 
 typedef struct {

source/dnode/vnode/src/inc/sma.h

@@ -38,9 +38,10 @@ typedef struct SSmaEnv       SSmaEnv;
 typedef struct SSmaStat      SSmaStat;
 typedef struct STSmaStat     STSmaStat;
 typedef struct SRSmaStat     SRSmaStat;
-typedef struct SSmaKey       SSmaKey;
+typedef struct SRSmaRef      SRSmaRef;
 typedef struct SRSmaInfo     SRSmaInfo;
 typedef struct SRSmaInfoItem SRSmaInfoItem;
+typedef struct SRSmaFS       SRSmaFS;
 typedef struct SQTaskFile    SQTaskFile;
 typedef struct SQTaskFReader SQTaskFReader;
 typedef struct SQTaskFWriter SQTaskFWriter;
@@ -54,10 +55,21 @@ struct SSmaEnv {
 
 #define SMA_ENV_FLG_CLOSE ((int8_t)0x1)
 
+struct SRSmaRef {
+  int64_t refId;  // for SRSmaStat
+  int64_t suid;
+};
+
 typedef struct {
   int8_t  inited;
   int32_t rsetId;
   void   *tmrHandle;  // shared by all fetch tasks
+  /**
+   * @brief key: void* of SRSmaInfoItem, value: SRSmaRef
+   *  N.B. Although there is a very small possibility that "void*" point to different objects while with the same
+   * address after release/renew, the functionality is not affected as it just used to fetch the rsma results.
+   */
+  SHashObj *refHash;  // shared by all vgroups
 } SSmaMgmt;
 
 #define SMA_ENV_LOCK(env)  (&(env)->lock)
@@ -73,20 +85,25 @@ struct STSmaStat {
 
 struct SQTaskFile {
   volatile int32_t nRef;
-  int64_t          commitID;
+  int32_t          padding;
+  int64_t          version;
   int64_t          size;
 };
 
 struct SQTaskFReader {
-  SSma      *pSma;
-  SQTaskFile fTask;
-  TdFilePtr  pReadH;
+  SSma     *pSma;
+  int64_t   version;
+  TdFilePtr pReadH;
 };
 struct SQTaskFWriter {
-  SSma      *pSma;
-  SQTaskFile fTask;
-  TdFilePtr  pWriteH;
-  char      *fname;
+  SSma     *pSma;
+  int64_t   version;
+  TdFilePtr pWriteH;
+  char     *fname;
+};
+
+struct SRSmaFS {
+  SArray *aQTaskInf;  // array of SQTaskFile
 };
 
 struct SRSmaStat {
@@ -98,7 +115,7 @@ struct SRSmaStat {
   volatile int32_t nFetchAll;         // active number of fetch all
   int8_t           triggerStat;       // shared by fetch tasks
   int8_t           commitStat;        // 0 not in committing, 1 in committing
-  SArray          *aTaskFile;         // qTaskFiles committed recently(for recovery/snapshot r/w)
+  SRSmaFS          fs;                // for recovery/snapshot r/w
   SHashObj        *infoHash;          // key: suid, value: SRSmaInfo
   tsem_t           notEmpty;          // has items in queue buffer
 };
@@ -118,6 +135,7 @@ struct SSmaStat {
 #define RSMA_TRIGGER_STAT(r) (&(r)->triggerStat)
 #define RSMA_COMMIT_STAT(r)  (&(r)->commitStat)
 #define RSMA_REF_ID(r)       ((r)->refId)
+#define RSMA_FS(r)           (&(r)->fs)
 #define RSMA_FS_LOCK(r)      (&(r)->lock)
 
 struct SRSmaInfoItem {
@@ -132,7 +150,6 @@ struct SRSmaInfoItem {
 struct SRSmaInfo {
   STSchema *pTSchema;
   int64_t   suid;
-  int64_t   refId;     // refId of SRSmaStat
   int64_t   lastRecv;  // ms
   int8_t    assigned;  // 0 idle, 1 assgined for exec
   int8_t    delFlag;
@@ -163,14 +180,6 @@ enum {
   TASK_TRIGGER_STAT_DROPPED = 5,
 };
 
-enum {
-  RSMA_ROLE_CREATE = 0,
-  RSMA_ROLE_DROP = 1,
-  RSMA_ROLE_SUBMIT = 2,
-  RSMA_ROLE_FETCH = 3,
-  RSMA_ROLE_ITERATE = 4,
-};
-
 enum {
   RSMA_RESTORE_REBOOT = 1,
   RSMA_RESTORE_SYNC = 2,
@@ -182,88 +191,34 @@ typedef enum {
   RSMA_EXEC_COMMIT = 3,    // triggered by commit
 } ERsmaExecType;
 
-void  tdDestroySmaEnv(SSmaEnv *pSmaEnv);
-void *tdFreeSmaEnv(SSmaEnv *pSmaEnv);
-
-int32_t tdDropTSma(SSma *pSma, char *pMsg);
-int32_t tdDropTSmaData(SSma *pSma, int64_t indexUid);
-int32_t tdInsertRSmaData(SSma *pSma, char *msg);
-
+// sma
+int32_t tdCheckAndInitSmaEnv(SSma *pSma, int8_t smaType);
+void    tdDestroySmaEnv(SSmaEnv *pSmaEnv);
+void   *tdFreeSmaEnv(SSmaEnv *pSmaEnv);
 int32_t tdRefSmaStat(SSma *pSma, SSmaStat *pStat);
 int32_t tdUnRefSmaStat(SSma *pSma, SSmaStat *pStat);
-int32_t tdRefRSmaInfo(SSma *pSma, SRSmaInfo *pRSmaInfo);
-int32_t tdUnRefRSmaInfo(SSma *pSma, SRSmaInfo *pRSmaInfo);
-
-void   *tdAcquireSmaRef(int32_t rsetId, int64_t refId);
-int32_t tdReleaseSmaRef(int32_t rsetId, int64_t refId);
-
-int32_t tdCheckAndInitSmaEnv(SSma *pSma, int8_t smaType);
-
 int32_t tdLockSma(SSma *pSma);
 int32_t tdUnLockSma(SSma *pSma);
+void   *tdAcquireSmaRef(int32_t rsetId, int64_t refId);
+int32_t tdReleaseSmaRef(int32_t rsetId, int64_t refId);
 
-static FORCE_INLINE int8_t tdSmaStat(STSmaStat *pTStat) {
-  if (pTStat) {
-    return atomic_load_8(&pTStat->state);
-  }
-  return TSDB_SMA_STAT_UNKNOWN;
-}
-
-static FORCE_INLINE bool tdSmaStatIsOK(STSmaStat *pTStat, int8_t *state) {
-  if (!pTStat) {
-    return false;
-  }
-
-  if (state) {
-    *state = atomic_load_8(&pTStat->state);
-    return *state == TSDB_SMA_STAT_OK;
-  }
-  return atomic_load_8(&pTStat->state) == TSDB_SMA_STAT_OK;
-}
-
-static FORCE_INLINE bool tdSmaStatIsExpired(STSmaStat *pTStat) {
-  return pTStat ? (atomic_load_8(&pTStat->state) & TSDB_SMA_STAT_EXPIRED) : true;
-}
-
-static FORCE_INLINE bool tdSmaStatIsDropped(STSmaStat *pTStat) {
-  return pTStat ? (atomic_load_8(&pTStat->state) & TSDB_SMA_STAT_DROPPED) : true;
-}
-
-static FORCE_INLINE void tdSmaStatSetOK(STSmaStat *pTStat) {
-  if (pTStat) {
-    atomic_store_8(&pTStat->state, TSDB_SMA_STAT_OK);
-  }
-}
-
-static FORCE_INLINE void tdSmaStatSetExpired(STSmaStat *pTStat) {
-  if (pTStat) {
-    atomic_or_fetch_8(&pTStat->state, TSDB_SMA_STAT_EXPIRED);
-  }
-}
-
-static FORCE_INLINE void tdSmaStatSetDropped(STSmaStat *pTStat) {
-  if (pTStat) {
-    atomic_or_fetch_8(&pTStat->state, TSDB_SMA_STAT_DROPPED);
-  }
-}
-
-void           tdRSmaQTaskInfoGetFileName(int32_t vid, int64_t version, char *outputName);
-void           tdRSmaQTaskInfoGetFullName(int32_t vid, int64_t version, const char *path, char *outputName);
-int32_t        tdCloneRSmaInfo(SSma *pSma, SRSmaInfo *pInfo);
-void           tdFreeQTaskInfo(qTaskInfo_t *taskHandle, int32_t vgId, int32_t level);
-static int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType);
-void          *tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType);
-void          *tdFreeRSmaInfo(SSma *pSma, SRSmaInfo *pInfo, bool isDeepFree);
-int32_t        tdRSmaPersistExecImpl(SRSmaStat *pRSmaStat, SHashObj *pInfoHash);
-int32_t        tdRSmaProcessExecImpl(SSma *pSma, ERsmaExecType type);
-
-int32_t tdProcessRSmaCreateImpl(SSma *pSma, SRSmaParam *param, int64_t suid, const char *tbName);
-int32_t tdProcessRSmaRestoreImpl(SSma *pSma, int8_t type, int64_t qtaskFileVer);
-int32_t tdRsmaRestore(SSma *pSma, int8_t type, int64_t committedVer);
-
-int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg);
-int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg);
-int32_t tdProcessTSmaGetDaysImpl(SVnodeCfg *pCfg, void *pCont, uint32_t contLen, int32_t *days);
+// rsma
+int32_t tdRefRSmaInfo(SSma *pSma, SRSmaInfo *pRSmaInfo);
+int32_t tdUnRefRSmaInfo(SSma *pSma, SRSmaInfo *pRSmaInfo);
+void   *tdFreeRSmaInfo(SSma *pSma, SRSmaInfo *pInfo, bool isDeepFree);
+int32_t tdRSmaFSOpen(SSma *pSma, int64_t version);
+void    tdRSmaFSClose(SRSmaFS *fs);
+int32_t tdRSmaFSRef(SSma *pSma, SRSmaStat *pStat, int64_t version);
+void    tdRSmaFSUnRef(SSma *pSma, SRSmaStat *pStat, int64_t version);
+int32_t tdRSmaFSUpsertQTaskFile(SRSmaFS *pFS, SQTaskFile *qTaskFile);
+int32_t tdRSmaRestore(SSma *pSma, int8_t type, int64_t committedVer);
+int32_t tdRSmaProcessCreateImpl(SSma *pSma, SRSmaParam *param, int64_t suid, const char *tbName);
+int32_t tdRSmaProcessExecImpl(SSma *pSma, ERsmaExecType type);
+int32_t tdRSmaPersistExecImpl(SRSmaStat *pRSmaStat, SHashObj *pInfoHash);
+int32_t tdRSmaProcessRestoreImpl(SSma *pSma, int8_t type, int64_t qtaskFileVer);
+
+void tdRSmaQTaskInfoGetFileName(int32_t vid, int64_t version, char *outputName);
+void tdRSmaQTaskInfoGetFullName(int32_t vid, int64_t version, const char *path, char *outputName);
 
 // smaFileUtil ================
 

source/dnode/vnode/src/inc/tsdb.h

@@ -42,15 +42,15 @@ typedef struct SMemTable     SMemTable;
 typedef struct STbDataIter   STbDataIter;
 typedef struct SMapData      SMapData;
 typedef struct SBlockIdx     SBlockIdx;
-typedef struct SBlock        SBlock;
-typedef struct SBlockL       SBlockL;
+typedef struct SDataBlk      SDataBlk;
+typedef struct SSstBlk       SSstBlk;
 typedef struct SColData      SColData;
 typedef struct SDiskDataHdr  SDiskDataHdr;
 typedef struct SBlockData    SBlockData;
 typedef struct SDelFile      SDelFile;
 typedef struct SHeadFile     SHeadFile;
 typedef struct SDataFile     SDataFile;
-typedef struct SLastFile     SLastFile;
+typedef struct SSstFile      SSstFile;
 typedef struct SSmaFile      SSmaFile;
 typedef struct SDFileSet     SDFileSet;
 typedef struct SDataFWriter  SDataFWriter;
@@ -64,10 +64,13 @@ typedef struct STsdbReadSnap STsdbReadSnap;
 typedef struct SBlockInfo    SBlockInfo;
 typedef struct SSmaInfo      SSmaInfo;
 typedef struct SBlockCol     SBlockCol;
+typedef struct SVersionRange SVersionRange;
 
-#define TSDB_FILE_DLMT     ((uint32_t)0xF00AFA0F)
-#define TSDB_MAX_SUBBLOCKS 8
-#define TSDB_FHDR_SIZE     512
+#define TSDB_FILE_DLMT         ((uint32_t)0xF00AFA0F)
+#define TSDB_MAX_SUBBLOCKS     8
+#define TSDB_MAX_LAST_FILE     16
+#define TSDB_DEFAULT_LAST_FILE 8
+#define TSDB_FHDR_SIZE         512
 
 #define HAS_NONE  ((int8_t)0x1)
 #define HAS_NULL  ((int8_t)0x2)
@@ -111,15 +114,15 @@ int32_t tTABLEIDCmprFn(const void *p1, const void *p2);
 int32_t tPutBlockCol(uint8_t *p, void *ph);
 int32_t tGetBlockCol(uint8_t *p, void *ph);
 int32_t tBlockColCmprFn(const void *p1, const void *p2);
-// SBlock
-void    tBlockReset(SBlock *pBlock);
-int32_t tPutBlock(uint8_t *p, void *ph);
-int32_t tGetBlock(uint8_t *p, void *ph);
-int32_t tBlockCmprFn(const void *p1, const void *p2);
-bool    tBlockHasSma(SBlock *pBlock);
-// SBlockL
-int32_t tPutBlockL(uint8_t *p, void *ph);
-int32_t tGetBlockL(uint8_t *p, void *ph);
+// SDataBlk
+void    tDataBlkReset(SDataBlk *pBlock);
+int32_t tPutDataBlk(uint8_t *p, void *ph);
+int32_t tGetDataBlk(uint8_t *p, void *ph);
+int32_t tDataBlkCmprFn(const void *p1, const void *p2);
+bool    tDataBlkHasSma(SDataBlk *pDataBlk);
+// SSstBlk
+int32_t tPutSstBlk(uint8_t *p, void *ph);
+int32_t tGetSstBlk(uint8_t *p, void *ph);
 // SBlockIdx
 int32_t tPutBlockIdx(uint8_t *p, void *ph);
 int32_t tGetBlockIdx(uint8_t *p, void *ph);
@@ -170,6 +173,7 @@ int32_t tGetDelData(uint8_t *p, void *ph);
 void    tMapDataReset(SMapData *pMapData);
 void    tMapDataClear(SMapData *pMapData);
 int32_t tMapDataPutItem(SMapData *pMapData, void *pItem, int32_t (*tPutItemFn)(uint8_t *, void *));
+int32_t tMapDataCopy(SMapData *pFrom, SMapData *pTo);
 void    tMapDataGetItemByIdx(SMapData *pMapData, int32_t idx, void *pItem, int32_t (*tGetItemFn)(uint8_t *, void *));
 int32_t tMapDataSearch(SMapData *pMapData, void *pSearchItem, int32_t (*tGetItemFn)(uint8_t *, void *),
                        int32_t (*tItemCmprFn)(const void *, const void *), void *pItem);
@@ -215,7 +219,7 @@ bool    tsdbDelFileIsSame(SDelFile *pDelFile1, SDelFile *pDelFile2);
 int32_t tsdbDFileRollback(STsdb *pTsdb, SDFileSet *pSet, EDataFileT ftype);
 int32_t tPutHeadFile(uint8_t *p, SHeadFile *pHeadFile);
 int32_t tPutDataFile(uint8_t *p, SDataFile *pDataFile);
-int32_t tPutLastFile(uint8_t *p, SLastFile *pLastFile);
+int32_t tPutSstFile(uint8_t *p, SSstFile *pSstFile);
 int32_t tPutSmaFile(uint8_t *p, SSmaFile *pSmaFile);
 int32_t tPutDelFile(uint8_t *p, SDelFile *pDelFile);
 int32_t tGetDelFile(uint8_t *p, SDelFile *pDelFile);
@@ -224,7 +228,7 @@ int32_t tGetDFileSet(uint8_t *p, SDFileSet *pSet);
 
 void tsdbHeadFileName(STsdb *pTsdb, SDiskID did, int32_t fid, SHeadFile *pHeadF, char fname[]);
 void tsdbDataFileName(STsdb *pTsdb, SDiskID did, int32_t fid, SDataFile *pDataF, char fname[]);
-void tsdbLastFileName(STsdb *pTsdb, SDiskID did, int32_t fid, SLastFile *pLastF, char fname[]);
+void tsdbSstFileName(STsdb *pTsdb, SDiskID did, int32_t fid, SSstFile *pSstF, char fname[]);
 void tsdbSmaFileName(STsdb *pTsdb, SDiskID did, int32_t fid, SSmaFile *pSmaF, char fname[]);
 // SDelFile
 void tsdbDelFileName(STsdb *pTsdb, SDelFile *pFile, char fname[]);
@@ -250,7 +254,7 @@ int32_t tsdbDataFWriterClose(SDataFWriter **ppWriter, int8_t sync);
 int32_t tsdbUpdateDFileSetHeader(SDataFWriter *pWriter);
 int32_t tsdbWriteBlockIdx(SDataFWriter *pWriter, SArray *aBlockIdx);
 int32_t tsdbWriteBlock(SDataFWriter *pWriter, SMapData *pMapData, SBlockIdx *pBlockIdx);
-int32_t tsdbWriteBlockL(SDataFWriter *pWriter, SArray *aBlockL);
+int32_t tsdbWriteSstBlk(SDataFWriter *pWriter, SArray *aSstBlk);
 int32_t tsdbWriteBlockData(SDataFWriter *pWriter, SBlockData *pBlockData, SBlockInfo *pBlkInfo, SSmaInfo *pSmaInfo,
                            int8_t cmprAlg, int8_t toLast);
 
@@ -260,10 +264,10 @@ int32_t tsdbDataFReaderOpen(SDataFReader **ppReader, STsdb *pTsdb, SDFileSet *pS
 int32_t tsdbDataFReaderClose(SDataFReader **ppReader);
 int32_t tsdbReadBlockIdx(SDataFReader *pReader, SArray *aBlockIdx);
 int32_t tsdbReadBlock(SDataFReader *pReader, SBlockIdx *pBlockIdx, SMapData *pMapData);
-int32_t tsdbReadBlockL(SDataFReader *pReader, SArray *aBlockL);
-int32_t tsdbReadBlockSma(SDataFReader *pReader, SBlock *pBlock, SArray *aColumnDataAgg);
-int32_t tsdbReadDataBlock(SDataFReader *pReader, SBlock *pBlock, SBlockData *pBlockData);
-int32_t tsdbReadLastBlock(SDataFReader *pReader, SBlockL *pBlockL, SBlockData *pBlockData);
+int32_t tsdbReadSstBlk(SDataFReader *pReader, int32_t iSst, SArray *aSstBlk);
+int32_t tsdbReadBlockSma(SDataFReader *pReader, SDataBlk *pBlock, SArray *aColumnDataAgg);
+int32_t tsdbReadDataBlock(SDataFReader *pReader, SDataBlk *pBlock, SBlockData *pBlockData);
+int32_t tsdbReadSstBlock(SDataFReader *pReader, int32_t iSst, SSstBlk *pSstBlk, SBlockData *pBlockData);
 // SDelFWriter
 int32_t tsdbDelFWriterOpen(SDelFWriter **ppWriter, SDelFile *pFile, STsdb *pTsdb);
 int32_t tsdbDelFWriterClose(SDelFWriter **ppWriter, int8_t sync);
@@ -278,6 +282,8 @@ int32_t tsdbReadDelIdx(SDelFReader *pReader, SArray *aDelIdx);
 // tsdbRead.c ==============================================================================================
 int32_t tsdbTakeReadSnap(STsdb *pTsdb, STsdbReadSnap **ppSnap);
 void    tsdbUntakeReadSnap(STsdb *pTsdb, STsdbReadSnap *pSnap);
+// tsdbMerge.c ==============================================================================================
+int32_t tsdbMerge(STsdb *pTsdb);
 
 #define TSDB_CACHE_NO(c)       ((c).cacheLast == 0)
 #define TSDB_CACHE_LAST_ROW(c) (((c).cacheLast & 1) > 0)
@@ -324,6 +330,11 @@ struct TSDBKEY {
   TSKEY   ts;
 };
 
+struct SVersionRange {
+  uint64_t minVer;
+  uint64_t maxVer;
+};
+
 typedef struct SMemSkipListNode SMemSkipListNode;
 struct SMemSkipListNode {
   int8_t            level;
@@ -416,7 +427,7 @@ struct SSmaInfo {
   int32_t size;
 };
 
-struct SBlock {
+struct SDataBlk {
   TSDBKEY    minKey;
   TSDBKEY    maxKey;
   int64_t    minVer;
@@ -428,7 +439,7 @@ struct SBlock {
   SSmaInfo   smaInfo;
 };
 
-struct SBlockL {
+struct SSstBlk {
   int64_t    suid;
   int64_t    minUid;
   int64_t    maxUid;
@@ -467,12 +478,6 @@ struct SBlockData {
   SArray  *aColData;  // SArray<SColData>
 };
 
-// ================== TSDB global config
-extern bool tsdbForceKeepFile;
-
-#define TSDB_FS_ITER_FORWARD  TSDB_ORDER_ASC
-#define TSDB_FS_ITER_BACKWARD TSDB_ORDER_DESC
-
 struct TABLEID {
   tb_uid_t suid;
   tb_uid_t uid;
@@ -536,7 +541,7 @@ struct SDataFile {
   int64_t size;
 };
 
-struct SLastFile {
+struct SSstFile {
   volatile int32_t nRef;
 
   int64_t commitID;
@@ -556,8 +561,9 @@ struct SDFileSet {
   int32_t    fid;
   SHeadFile *pHeadF;
   SDataFile *pDataF;
-  SLastFile *pLastF;
   SSmaFile  *pSmaF;
+  uint8_t    nSstF;
+  SSstFile  *aSstF[TSDB_MAX_LAST_FILE];
 };
 
 struct SRowIter {
@@ -586,13 +592,13 @@ struct SDataFWriter {
 
   TdFilePtr pHeadFD;
   TdFilePtr pDataFD;
-  TdFilePtr pLastFD;
   TdFilePtr pSmaFD;
+  TdFilePtr pLastFD;
 
   SHeadFile fHead;
   SDataFile fData;
-  SLastFile fLast;
   SSmaFile  fSma;
+  SSstFile  fSst[TSDB_MAX_LAST_FILE];
 
   uint8_t *aBuf[4];
 };
@@ -603,6 +609,36 @@ struct STsdbReadSnap {
   STsdbFS    fs;
 };
 
+struct SDataFReader {
+  STsdb     *pTsdb;
+  SDFileSet *pSet;
+  TdFilePtr  pHeadFD;
+  TdFilePtr  pDataFD;
+  TdFilePtr  pSmaFD;
+  TdFilePtr  aLastFD[TSDB_MAX_LAST_FILE];
+
+  uint8_t *aBuf[3];
+};
+
+typedef struct {
+  int64_t suid;
+  int64_t uid;
+  TSDBROW row;
+} SRowInfo;
+
+typedef struct SMergeTree {
+  int8_t             backward;
+  SRBTree            rbt;
+  SArray            *pIterList;
+  struct SLDataIter *pIter;
+} SMergeTree;
+
+int32_t tMergeTreeOpen(SMergeTree *pMTree, int8_t backward, SDataFReader* pFReader, uint64_t uid, STimeWindow* pTimeWindow, SVersionRange* pVerRange);
+void tMergeTreeAddIter(SMergeTree *pMTree, struct SLDataIter *pIter);
+bool tMergeTreeNext(SMergeTree *pMTree);
+TSDBROW tMergeTreeGetRow(SMergeTree *pMTree);
+void    tMergeTreeClose(SMergeTree *pMTree);
+
 // ========== inline functions ==========
 static FORCE_INLINE int32_t tsdbKeyCmprFn(const void *p1, const void *p2) {
   TSDBKEY *pKey1 = (TSDBKEY *)p1;

source/dnode/vnode/src/inc/vnodeInt.h

@@ -36,6 +36,7 @@
 #include "tlosertree.h"
 #include "tlrucache.h"
 #include "tmsgcb.h"
+#include "trbtree.h"
 #include "tref.h"
 #include "tskiplist.h"
 #include "tstream.h"
@@ -70,8 +71,8 @@ typedef struct SStreamTaskReader  SStreamTaskReader;
 typedef struct SStreamTaskWriter  SStreamTaskWriter;
 typedef struct SStreamStateReader SStreamStateReader;
 typedef struct SStreamStateWriter SStreamStateWriter;
-typedef struct SRsmaSnapReader    SRsmaSnapReader;
-typedef struct SRsmaSnapWriter    SRsmaSnapWriter;
+typedef struct SRSmaSnapReader    SRSmaSnapReader;
+typedef struct SRSmaSnapWriter    SRSmaSnapWriter;
 typedef struct SSnapDataHdr       SSnapDataHdr;
 
 #define VNODE_META_DIR  "meta"
@@ -102,7 +103,7 @@ int             metaCommit(SMeta* pMeta);
 int             metaCreateSTable(SMeta* pMeta, int64_t version, SVCreateStbReq* pReq);
 int             metaAlterSTable(SMeta* pMeta, int64_t version, SVCreateStbReq* pReq);
 int             metaDropSTable(SMeta* pMeta, int64_t verison, SVDropStbReq* pReq, SArray* tbUidList);
-int             metaCreateTable(SMeta* pMeta, int64_t version, SVCreateTbReq* pReq, STableMetaRsp **pMetaRsp);
+int             metaCreateTable(SMeta* pMeta, int64_t version, SVCreateTbReq* pReq, STableMetaRsp** pMetaRsp);
 int             metaDropTable(SMeta* pMeta, int64_t version, SVDropTbReq* pReq, SArray* tbUids);
 int             metaTtlDropTable(SMeta* pMeta, int64_t ttl, SArray* tbUids);
 int             metaAlterTable(SMeta* pMeta, int64_t version, SVAlterTbReq* pReq, STableMetaRsp* pMetaRsp);
@@ -173,7 +174,8 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg);
 // tq-stream
 int32_t tqProcessTaskDeployReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
 int32_t tqProcessTaskDropReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
-int32_t tqProcessStreamTrigger(STQ* pTq, SSubmitReq* data, int64_t ver);
+int32_t tqProcessSubmitReq(STQ* pTq, SSubmitReq* data, int64_t ver);
+int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver);
 int32_t tqProcessTaskRunReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg, bool exec);
 int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg);
@@ -246,14 +248,14 @@ int32_t tqOffsetSnapWrite(STqOffsetWriter* pWriter, uint8_t* pData, uint32_t nDa
 // SStreamTaskReader ======================================
 // SStreamStateWriter =====================================
 // SStreamStateReader =====================================
-// SRsmaSnapReader ========================================
-int32_t rsmaSnapReaderOpen(SSma* pSma, int64_t sver, int64_t ever, SRsmaSnapReader** ppReader);
-int32_t rsmaSnapReaderClose(SRsmaSnapReader** ppReader);
-int32_t rsmaSnapRead(SRsmaSnapReader* pReader, uint8_t** ppData);
-// SRsmaSnapWriter ========================================
-int32_t rsmaSnapWriterOpen(SSma* pSma, int64_t sver, int64_t ever, SRsmaSnapWriter** ppWriter);
-int32_t rsmaSnapWrite(SRsmaSnapWriter* pWriter, uint8_t* pData, uint32_t nData);
-int32_t rsmaSnapWriterClose(SRsmaSnapWriter** ppWriter, int8_t rollback);
+// SRSmaSnapReader ========================================
+int32_t rsmaSnapReaderOpen(SSma* pSma, int64_t sver, int64_t ever, SRSmaSnapReader** ppReader);
+int32_t rsmaSnapReaderClose(SRSmaSnapReader** ppReader);
+int32_t rsmaSnapRead(SRSmaSnapReader* pReader, uint8_t** ppData);
+// SRSmaSnapWriter ========================================
+int32_t rsmaSnapWriterOpen(SSma* pSma, int64_t sver, int64_t ever, SRSmaSnapWriter** ppWriter);
+int32_t rsmaSnapWrite(SRSmaSnapWriter* pWriter, uint8_t* pData, uint32_t nData);
+int32_t rsmaSnapWriterClose(SRSmaSnapWriter** ppWriter, int8_t rollback);
 
 typedef struct {
   int8_t  streamType;  // sma or other

source/dnode/vnode/src/meta/metaQuery.c

@@ -129,10 +129,16 @@ _err:
 
 bool metaIsTableExist(SMeta *pMeta, tb_uid_t uid) {
   // query uid.idx
+  metaRLock(pMeta);
+
   if (tdbTbGet(pMeta->pUidIdx, &uid, sizeof(uid), NULL, NULL) < 0) {
+    metaULock(pMeta);
+
     return false;
   }
 
+  metaULock(pMeta);
+
   return true;
 }
 
@@ -182,9 +188,14 @@ tb_uid_t metaGetTableEntryUidByName(SMeta *pMeta, const char *name) {
 }
 
 int metaGetTableNameByUid(void *meta, uint64_t uid, char *tbName) {
+  int         code = 0;
   SMetaReader mr = {0};
   metaReaderInit(&mr, (SMeta *)meta, 0);
-  metaGetTableEntryByUid(&mr, uid);
+  code = metaGetTableEntryByUid(&mr, uid);
+  if (code < 0) {
+    metaReaderClear(&mr);
+    return -1;
+  }
 
   STR_TO_VARSTR(tbName, mr.me.name);
   metaReaderClear(&mr);
@@ -280,6 +291,38 @@ _query:
       tDecoderClear(&dc);
       goto _exit;
     }
+    { // Traverse to find the previous qualified data
+      TBC      *pCur;
+      tdbTbcOpen(pMeta->pTbDb, &pCur, NULL);
+      STbDbKey key = {.version = sver, .uid = INT64_MAX};
+      int c = 0;
+      tdbTbcMoveTo(pCur, &key, sizeof(key), &c);
+      if(c < 0){
+        tdbTbcMoveToPrev(pCur);
+      }
+
+      void *pKey = NULL;
+      void *pVal = NULL;
+      int   vLen = 0, kLen = 0;
+      while(1){
+        int32_t ret = tdbTbcPrev(pCur, &pKey, &kLen, &pVal, &vLen);
+        if (ret < 0) break;
+
+        STbDbKey *tmp = (STbDbKey*)pKey;
+        if(tmp->uid != uid){
+          continue;
+        }
+        SDecoder   dcNew = {0};
+        SMetaEntry meNew = {0};
+        tDecoderInit(&dcNew, pVal, vLen);
+        metaDecodeEntry(&dcNew, &meNew);
+        pSchema = tCloneSSchemaWrapper(&meNew.stbEntry.schemaRow);
+        tDecoderClear(&dcNew);
+        tdbTbcClose(pCur);
+        goto _exit;
+      }
+      tdbTbcClose(pCur);
+    }
   } else if (me.type == TSDB_CHILD_TABLE) {
     uid = me.ctbEntry.suid;
     tDecoderClear(&dc);

source/dnode/vnode/src/sma/smaCommit.c

@@ -15,13 +15,15 @@
 
 #include "sma.h"
 
+extern SSmaMgmt smaMgmt;
+
 static int32_t tdProcessRSmaSyncPreCommitImpl(SSma *pSma);
 static int32_t tdProcessRSmaSyncCommitImpl(SSma *pSma);
 static int32_t tdProcessRSmaSyncPostCommitImpl(SSma *pSma);
 static int32_t tdProcessRSmaAsyncPreCommitImpl(SSma *pSma);
 static int32_t tdProcessRSmaAsyncCommitImpl(SSma *pSma);
 static int32_t tdProcessRSmaAsyncPostCommitImpl(SSma *pSma);
-static int32_t tdCleanupQTaskInfoFiles(SSma *pSma, SRSmaStat *pRSmaStat);
+static int32_t tdUpdateQTaskInfoFiles(SSma *pSma, SRSmaStat *pRSmaStat);
 
 /**
  * @brief Only applicable to Rollup SMA
@@ -166,114 +168,49 @@ static int32_t tdProcessRSmaSyncCommitImpl(SSma *pSma) {
   return TSDB_CODE_SUCCESS;
 }
 
-static int32_t tdCleanupQTaskInfoFiles(SSma *pSma, SRSmaStat *pRSmaStat) {
-  SVnode       *pVnode = pSma->pVnode;
-  int64_t       committed = pRSmaStat->commitAppliedVer;
-  TdDirPtr      pDir = NULL;
-  TdDirEntryPtr pDirEntry = NULL;
-  char          dir[TSDB_FILENAME_LEN];
-  const char   *pattern = "v[0-9]+qinf\\.v([0-9]+)?$";
-  regex_t       regex;
-  int           code = 0;
-
-  tdGetVndDirName(TD_VID(pVnode), tfsGetPrimaryPath(pVnode->pTfs), VNODE_RSMA_DIR, true, dir);
-
-  // Resource allocation and init
-  if ((code = regcomp(&regex, pattern, REG_EXTENDED)) != 0) {
-    char errbuf[128];
-    regerror(code, &regex, errbuf, sizeof(errbuf));
-    smaWarn("vgId:%d, rsma post commit, regcomp for %s failed since %s", TD_VID(pVnode), dir, errbuf);
-    return TSDB_CODE_FAILED;
-  }
-
-  if ((pDir = taosOpenDir(dir)) == NULL) {
-    regfree(&regex);
-    terrno = TAOS_SYSTEM_ERROR(errno);
-    smaDebug("vgId:%d, rsma post commit, open dir %s failed since %s", TD_VID(pVnode), dir, terrstr());
-    return TSDB_CODE_FAILED;
-  }
-
-  int32_t    dirLen = strlen(dir);
-  char      *dirEnd = POINTER_SHIFT(dir, dirLen);
-  regmatch_t regMatch[2];
-  while ((pDirEntry = taosReadDir(pDir)) != NULL) {
-    char *entryName = taosGetDirEntryName(pDirEntry);
-    if (!entryName) {
-      continue;
-    }
-
-    code = regexec(&regex, entryName, 2, regMatch, 0);
-
-    if (code == 0) {
-      // match
-      int64_t version = -1;
-      sscanf((const char *)POINTER_SHIFT(entryName, regMatch[1].rm_so), "%" PRIi64, &version);
-      if ((version < committed) && (version > -1)) {
-        strncpy(dirEnd, entryName, TSDB_FILENAME_LEN - dirLen);
-        if (taosRemoveFile(dir) != 0) {
-          terrno = TAOS_SYSTEM_ERROR(errno);
-          smaWarn("vgId:%d, committed version:%" PRIi64 ", failed to remove %s since %s", TD_VID(pVnode), committed,
-                  dir, terrstr());
-        } else {
-          smaDebug("vgId:%d, committed version:%" PRIi64 ", success to remove %s", TD_VID(pVnode), committed, dir);
-        }
-      }
-    } else if (code == REG_NOMATCH) {
-      // not match
-      smaTrace("vgId:%d, rsma post commit, not match %s", TD_VID(pVnode), entryName);
-      continue;
-    } else {
-      // has other error
-      char errbuf[128];
-      regerror(code, &regex, errbuf, sizeof(errbuf));
-      smaWarn("vgId:%d, rsma post commit, regexec failed since %s", TD_VID(pVnode), errbuf);
-
-      taosCloseDir(&pDir);
-      regfree(&regex);
-      return TSDB_CODE_FAILED;
-    }
-  }
-
-  taosCloseDir(&pDir);
-  regfree(&regex);
-
-  return TSDB_CODE_SUCCESS;
-}
-
 // SQTaskFile ======================================================
-// int32_t tCmprQTaskFile(void const *lhs, void const *rhs) {
-//   int64_t    *lCommitted = *(int64_t *)lhs;
-//   SQTaskFile *rQTaskF = (SQTaskFile *)rhs;
-
-//   if (lCommitted < rQTaskF->commitID) {
-//     return -1;
-//   } else if (lCommitted > rQTaskF->commitID) {
-//     return 1;
-//   }
 
-//   return 0;
-// }
-
-#if 0
 /**
  * @brief At most time, there is only one qtaskinfo file committed latest in aTaskFile. Sometimes, there would be
  * multiple qtaskinfo files supporting snapshot replication.
  *
  * @param pSma
- * @param pRSmaStat
+ * @param pStat
  * @return int32_t
  */
-static int32_t tdCleanupQTaskInfoFiles(SSma *pSma, SRSmaStat *pRSmaStat) {
-  SVnode *pVnode = pSma->pVnode;
-  int64_t committed = pRSmaStat->commitAppliedVer;
-  SArray *aTaskFile = pRSmaStat->aTaskFile;
+static int32_t tdUpdateQTaskInfoFiles(SSma *pSma, SRSmaStat *pStat) {
+  SVnode  *pVnode = pSma->pVnode;
+  SRSmaFS *pFS = RSMA_FS(pStat);
+  int64_t  committed = pStat->commitAppliedVer;
+  char     qTaskInfoFullName[TSDB_FILENAME_LEN];
+
+  taosWLockLatch(RSMA_FS_LOCK(pStat));
+
+  for (int32_t i = 0; i < taosArrayGetSize(pFS->aQTaskInf);) {
+    SQTaskFile *pTaskF = taosArrayGet(pFS->aQTaskInf, i);
+    if (atomic_sub_fetch_32(&pTaskF->nRef, 1) <= 0) {
+      tdRSmaQTaskInfoGetFullName(TD_VID(pVnode), pTaskF->version, tfsGetPrimaryPath(pVnode->pTfs), qTaskInfoFullName);
+      if (taosRemoveFile(qTaskInfoFullName) < 0) {
+        smaWarn("vgId:%d, cleanup qinf, failed to remove %s since %s", TD_VID(pVnode), qTaskInfoFullName,
+                tstrerror(TAOS_SYSTEM_ERROR(errno)));
+      } else {
+        smaDebug("vgId:%d, cleanup qinf, success to remove %s", TD_VID(pVnode), qTaskInfoFullName);
+      }
+      taosArrayRemove(pFS->aQTaskInf, i);
+      continue;
+    }
+    ++i;
+  }
 
-  void *qTaskFile = taosArraySearch(aTaskFile, committed, tCmprQTaskFile, TD_LE);
-  
+  SQTaskFile qFile = {.nRef = 1, .padding = 0, .version = committed, .size = 0};
+  if (tdRSmaFSUpsertQTaskFile(pFS, &qFile) < 0) {
+    taosWUnLockLatch(RSMA_FS_LOCK(pStat));
+    return TSDB_CODE_FAILED;
+  }
 
+  taosWUnLockLatch(RSMA_FS_LOCK(pStat));
   return TSDB_CODE_SUCCESS;
 }
-#endif
 
 /**
  * @brief post-commit for rollup sma
@@ -290,8 +227,7 @@ static int32_t tdProcessRSmaSyncPostCommitImpl(SSma *pSma) {
 
   SRSmaStat *pRSmaStat = SMA_RSMA_STAT(pSma);
 
-  // cleanup outdated qtaskinfo files
-  tdCleanupQTaskInfoFiles(pSma, pRSmaStat);
+  tdUpdateQTaskInfoFiles(pSma, pRSmaStat);
 
   return TSDB_CODE_SUCCESS;
 }
@@ -427,8 +363,8 @@ static int32_t tdProcessRSmaAsyncPostCommitImpl(SSma *pSma) {
     return TSDB_CODE_SUCCESS;
   }
 
-  SRSmaStat *pRSmaStat = (SRSmaStat *)SMA_ENV_STAT(pEnv);
-  SArray    *rsmaDeleted = NULL;
+  SRSmaStat     *pRSmaStat = (SRSmaStat *)SMA_ENV_STAT(pEnv);
+  SRSmaInfoItem *pItem = NULL;
 
   // step 1: merge qTaskInfo and iQTaskInfo
   // lock
@@ -441,11 +377,7 @@ static int32_t tdProcessRSmaAsyncPostCommitImpl(SSma *pSma) {
     if (RSMA_INFO_IS_DEL(pRSmaInfo)) {
       int32_t refVal = T_REF_VAL_GET(pRSmaInfo);
       if (refVal == 0) {
-        if (!rsmaDeleted) {
-          if ((rsmaDeleted = taosArrayInit(1, sizeof(tb_uid_t)))) {
-            taosArrayPush(rsmaDeleted, pSuid);
-          }
-        }
+        taosHashRemove(RSMA_INFO_HASH(pRSmaStat), pSuid, sizeof(*pSuid));
       } else {
         smaDebug(
             "vgId:%d, rsma async post commit, not free rsma info since ref is %d although already deleted for "
@@ -471,25 +403,10 @@ static int32_t tdProcessRSmaAsyncPostCommitImpl(SSma *pSma) {
 #endif
   }
 
-  for (int32_t i = 0; i < taosArrayGetSize(rsmaDeleted); ++i) {
-    tb_uid_t *pSuid = taosArrayGet(rsmaDeleted, i);
-    void     *pRSmaInfo = taosHashGet(RSMA_INFO_HASH(pRSmaStat), pSuid, sizeof(tb_uid_t));
-    if ((pRSmaInfo = *(SRSmaInfo **)pRSmaInfo)) {
-      tdFreeRSmaInfo(pSma, pRSmaInfo, true);
-      smaDebug(
-          "vgId:%d, rsma async post commit, free rsma info since already deleted and ref is 0 for "
-          "table:%" PRIi64,
-          SMA_VID(pSma), *pSuid);
-    }
-    taosHashRemove(RSMA_INFO_HASH(pRSmaStat), pSuid, sizeof(tb_uid_t));
-  }
-  taosArrayDestroy(rsmaDeleted);
-
   // unlock
   // taosWUnLockLatch(SMA_ENV_LOCK(pEnv));
 
-  // step 2: cleanup outdated qtaskinfo files
-  tdCleanupQTaskInfoFiles(pSma, pRSmaStat);
+  tdUpdateQTaskInfoFiles(pSma, pRSmaStat);
 
   atomic_store_8(RSMA_COMMIT_STAT(pRSmaStat), 0);
 

source/dnode/vnode/src/sma/smaEnv.c

@@ -28,6 +28,8 @@ static int32_t tdInitSmaEnv(SSma *pSma, int8_t smaType, SSmaEnv **ppEnv);
 static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType, const SSma *pSma);
 static int32_t tdRsmaStartExecutor(const SSma *pSma);
 static int32_t tdRsmaStopExecutor(const SSma *pSma);
+static int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType);
+static void   *tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType);
 static void   *tdFreeTSmaStat(STSmaStat *pStat);
 static void    tdDestroyRSmaStat(void *pRSmaStat);
 
@@ -59,12 +61,23 @@ int32_t smaInit() {
       return TSDB_CODE_FAILED;
     }
 
+    int32_t type = (8 == POINTER_BYTES) ? TSDB_DATA_TYPE_UBIGINT : TSDB_DATA_TYPE_UINT;
+    smaMgmt.refHash = taosHashInit(1, taosGetDefaultHashFunction(type), true, HASH_ENTRY_LOCK);
+    if (!smaMgmt.refHash) {
+      taosCloseRef(smaMgmt.rsetId);
+      atomic_store_8(&smaMgmt.inited, 0);
+      smaError("failed to init sma tmr hanle since %s", terrstr());
+      return TSDB_CODE_FAILED;
+    }
+
     // init fetch timer handle
     smaMgmt.tmrHandle = taosTmrInit(10000, 100, 10000, "RSMA");
     if (!smaMgmt.tmrHandle) {
       taosCloseRef(smaMgmt.rsetId);
+      taosHashCleanup(smaMgmt.refHash);
+      smaMgmt.refHash = NULL;
       atomic_store_8(&smaMgmt.inited, 0);
-      smaError("failed to init sma tmr hanle since %s", terrstr());
+      smaError("failed to init sma tmr handle since %s", terrstr());
       return TSDB_CODE_FAILED;
     }
 
@@ -93,6 +106,7 @@ void smaCleanUp() {
 
   if (old == 1) {
     taosCloseRef(smaMgmt.rsetId);
+    taosHashCleanup(smaMgmt.refHash);
     taosTmrCleanUp(smaMgmt.tmrHandle);
     smaInfo("sma mgmt env is cleaned up, rsetId:%d, tmrHandle:%p", smaMgmt.rsetId, smaMgmt.tmrHandle);
     atomic_store_8(&smaMgmt.inited, 0);
@@ -195,6 +209,21 @@ int32_t tdUnRefRSmaInfo(SSma *pSma, SRSmaInfo *pRSmaInfo) {
   return 0;
 }
 
+static void tRSmaInfoHashFreeNode(void *data) {
+  SRSmaInfo     *pRSmaInfo = NULL;
+  SRSmaInfoItem *pItem = NULL;
+
+  if ((pRSmaInfo = *(SRSmaInfo **)data)) {
+    if ((pItem = RSMA_INFO_ITEM((SRSmaInfo *)pRSmaInfo, 0)) && pItem->level) {
+      taosHashRemove(smaMgmt.refHash, &pItem, POINTER_BYTES);
+    }
+    if ((pItem = RSMA_INFO_ITEM((SRSmaInfo *)pRSmaInfo, 1)) && pItem->level) {
+      taosHashRemove(smaMgmt.refHash, &pItem, POINTER_BYTES);
+    }
+    tdFreeRSmaInfo(NULL, pRSmaInfo, true);
+  }
+}
+
 static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType, const SSma *pSma) {
   ASSERT(pSmaStat != NULL);
 
@@ -240,10 +269,16 @@ static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType, const SSma *pS
       if (!RSMA_INFO_HASH(pRSmaStat)) {
         return TSDB_CODE_FAILED;
       }
+      taosHashSetFreeFp(RSMA_INFO_HASH(pRSmaStat), tRSmaInfoHashFreeNode);
 
       if (tdRsmaStartExecutor(pSma) < 0) {
         return TSDB_CODE_FAILED;
       }
+
+      if (!(RSMA_FS(pRSmaStat)->aQTaskInf = taosArrayInit(1, sizeof(SQTaskFile)))) {
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
+        return TSDB_CODE_FAILED;
+      }
     } else if (smaType == TSDB_SMA_TYPE_TIME_RANGE) {
       // TODO
     } else {
@@ -278,14 +313,6 @@ static void tdDestroyRSmaStat(void *pRSmaStat) {
     tsem_destroy(&(pStat->notEmpty));
 
     // step 2: destroy the rsma info and associated fetch tasks
-    if (taosHashGetSize(RSMA_INFO_HASH(pStat)) > 0) {
-      void *infoHash = taosHashIterate(RSMA_INFO_HASH(pStat), NULL);
-      while (infoHash) {
-        SRSmaInfo *pSmaInfo = *(SRSmaInfo **)infoHash;
-        tdFreeRSmaInfo(pSma, pSmaInfo, true);
-        infoHash = taosHashIterate(RSMA_INFO_HASH(pStat), infoHash);
-      }
-    }
     taosHashCleanup(RSMA_INFO_HASH(pStat));
 
     // step 3: wait for all triggered fetch tasks to finish
@@ -307,12 +334,15 @@ static void tdDestroyRSmaStat(void *pRSmaStat) {
     // step 4:
     tdRsmaStopExecutor(pSma);
 
-    // step 5: free pStat
+    // step 5:
+    tdRSmaFSClose(RSMA_FS(pStat));
+
+    // step 6: free pStat
     taosMemoryFreeClear(pStat);
   }
 }
 
-void *tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType) {
+static void *tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType) {
   tdDestroySmaState(pSmaStat, smaType);
   if (smaType == TSDB_SMA_TYPE_TIME_RANGE) {
     taosMemoryFreeClear(pSmaStat);
@@ -329,7 +359,7 @@ void *tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType) {
  * @return int32_t
  */
 
-int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType) {
+static int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType) {
   if (pSmaStat) {
     if (smaType == TSDB_SMA_TYPE_TIME_RANGE) {
       tdDestroyTSmaStat(SMA_STAT_TSMA(pSmaStat));
@@ -337,7 +367,7 @@ int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType) {
       SRSmaStat *pRSmaStat = &pSmaStat->rsmaStat;
       int32_t    vid = SMA_VID(pRSmaStat->pSma);
       int64_t    refId = RSMA_REF_ID(pRSmaStat);
-      if (taosRemoveRef(smaMgmt.rsetId, RSMA_REF_ID(pRSmaStat)) < 0) {
+      if (taosRemoveRef(smaMgmt.rsetId, refId) < 0) {
         smaError("vgId:%d, remove refId:%" PRIi64 " from rsmaRef:%" PRIi32 " failed since %s", vid, refId,
                  smaMgmt.rsetId, terrstr());
       } else {

source/dnode/vnode/src/sma/smaFS.c

+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "sma.h"
+
+// =================================================================================================
+
+static int32_t tdFetchQTaskInfoFiles(SSma *pSma, int64_t version, SArray **output);
+static int32_t tdQTaskInfCmprFn1(const void *p1, const void *p2);
+static int32_t tdQTaskInfCmprFn2(const void *p1, const void *p2);
+/**
+ * @brief Open RSma FS from qTaskInfo files
+ *
+ * @param pSma
+ * @param version
+ * @return int32_t
+ */
+int32_t tdRSmaFSOpen(SSma *pSma, int64_t version) {
+  SVnode    *pVnode = pSma->pVnode;
+  int64_t    commitID = pVnode->state.commitID;
+  SSmaEnv   *pEnv = SMA_RSMA_ENV(pSma);
+  SRSmaStat *pStat = NULL;
+  SArray    *output = NULL;
+  
+  terrno = TSDB_CODE_SUCCESS;
+
+  if (!pEnv) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  if (tdFetchQTaskInfoFiles(pSma, version, &output) < 0) {
+    goto _end;
+  }
+
+  pStat = (SRSmaStat *)SMA_ENV_STAT(pEnv);
+
+  for (int32_t i = 0; i < taosArrayGetSize(output); ++i) {
+    int32_t vid = 0;
+    int64_t version = -1;
+    sscanf((const char *)taosArrayGetP(output, i), "v%dqinfo.v%" PRIi64, &vid, &version);
+    SQTaskFile qTaskFile = {.version = version, .nRef = 1};
+    if ((terrno = tdRSmaFSUpsertQTaskFile(RSMA_FS(pStat), &qTaskFile)) < 0) {
+      goto _end;
+    }
+    smaInfo("vgId:%d, open fs, version:%" PRIi64 ", ref:%" PRIi64, TD_VID(pVnode), qTaskFile.version, qTaskFile.nRef);
+  }
+
+_end:
+  for (int32_t i = 0; i < taosArrayGetSize(output); ++i) {
+    void *ptr = taosArrayGetP(output, i);
+    taosMemoryFreeClear(ptr);
+  }
+  taosArrayDestroy(output);
+
+  if (terrno != TSDB_CODE_SUCCESS) {
+    smaError("vgId:%d, open rsma fs failed since %s", TD_VID(pVnode), terrstr());
+    return TSDB_CODE_FAILED;
+  }
+  return TSDB_CODE_SUCCESS;
+}
+
+void tdRSmaFSClose(SRSmaFS *fs) { taosArrayDestroy(fs->aQTaskInf); }
+
+static int32_t tdQTaskInfCmprFn1(const void *p1, const void *p2) {
+  if (*(int64_t *)p1 < ((SQTaskFile *)p2)->version) {
+    return -1;
+  } else if (*(int64_t *)p1 > ((SQTaskFile *)p2)->version) {
+    return 1;
+  }
+  return 0;
+}
+
+int32_t tdRSmaFSRef(SSma *pSma, SRSmaStat *pStat, int64_t version) {
+  SArray     *aQTaskInf = RSMA_FS(pStat)->aQTaskInf;
+  SQTaskFile *pTaskF = NULL;
+  int32_t     oldVal = 0;
+
+  taosRLockLatch(RSMA_FS_LOCK(pStat));
+  if ((pTaskF = taosArraySearch(aQTaskInf, &version, tdQTaskInfCmprFn1, TD_EQ))) {
+    oldVal = atomic_fetch_add_32(&pTaskF->nRef, 1);
+    ASSERT(oldVal > 0);
+  }
+  taosRUnLockLatch(RSMA_FS_LOCK(pStat));
+  return oldVal;
+}
+
+void tdRSmaFSUnRef(SSma *pSma, SRSmaStat *pStat, int64_t version) {
+  SVnode     *pVnode = pSma->pVnode;
+  SArray     *aQTaskInf = RSMA_FS(pStat)->aQTaskInf;
+  char        qTaskFullName[TSDB_FILENAME_LEN];
+  SQTaskFile *pTaskF = NULL;
+  int32_t     idx = -1;
+
+  taosWLockLatch(RSMA_FS_LOCK(pStat));
+  if ((idx = taosArraySearchIdx(aQTaskInf, &version, tdQTaskInfCmprFn1, TD_EQ)) >= 0) {
+    ASSERT(idx < taosArrayGetSize(aQTaskInf));
+    pTaskF = taosArrayGet(aQTaskInf, idx);
+    if (atomic_sub_fetch_32(&pTaskF->nRef, 1) <= 0) {
+      tdRSmaQTaskInfoGetFullName(TD_VID(pVnode), pTaskF->version, tfsGetPrimaryPath(pVnode->pTfs), qTaskFullName);
+      if (taosRemoveFile(qTaskFullName) < 0) {
+        smaWarn("vgId:%d, failed to remove %s since %s", TD_VID(pVnode), qTaskFullName,
+                tstrerror(TAOS_SYSTEM_ERROR(errno)));
+      } else {
+        smaDebug("vgId:%d, success to remove %s", TD_VID(pVnode), qTaskFullName);
+      }
+      taosArrayRemove(aQTaskInf, idx);
+    }
+  }
+  taosWUnLockLatch(RSMA_FS_LOCK(pStat));
+}
+
+/**
+ * @brief Fetch qtaskfiles LE than version
+ *
+ * @param pSma
+ * @param version
+ * @param output
+ * @return int32_t
+ */
+static int32_t tdFetchQTaskInfoFiles(SSma *pSma, int64_t version, SArray **output) {
+  SVnode       *pVnode = pSma->pVnode;
+  TdDirPtr      pDir = NULL;
+  TdDirEntryPtr pDirEntry = NULL;
+  char          dir[TSDB_FILENAME_LEN];
+  const char   *pattern = "v[0-9]+qinf\\.v([0-9]+)?$";
+  regex_t       regex;
+  int           code = 0;
+
+  tdGetVndDirName(TD_VID(pVnode), tfsGetPrimaryPath(pVnode->pTfs), VNODE_RSMA_DIR, true, dir);
+
+  if (!taosCheckExistFile(dir)) {
+    smaDebug("vgId:%d, fetch qtask files, no need as dir %s not exist", TD_VID(pVnode), dir);
+    return TSDB_CODE_SUCCESS;
+  }
+
+  // Resource allocation and init
+  if ((code = regcomp(&regex, pattern, REG_EXTENDED)) != 0) {
+    terrno = TSDB_CODE_RSMA_REGEX_MATCH;
+    char errbuf[128];
+    regerror(code, &regex, errbuf, sizeof(errbuf));
+    smaWarn("vgId:%d, fetch qtask files, regcomp for %s failed since %s", TD_VID(pVnode), dir, errbuf);
+    return TSDB_CODE_FAILED;
+  }
+
+  if (!(pDir = taosOpenDir(dir))) {
+    regfree(&regex);
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    smaError("vgId:%d, fetch qtask files, open dir %s failed since %s", TD_VID(pVnode), dir, terrstr());
+    return TSDB_CODE_FAILED;
+  }
+
+  int32_t    dirLen = strlen(dir);
+  char      *dirEnd = POINTER_SHIFT(dir, dirLen);
+  regmatch_t regMatch[2];
+  while ((pDirEntry = taosReadDir(pDir))) {
+    char *entryName = taosGetDirEntryName(pDirEntry);
+    if (!entryName) {
+      continue;
+    }
+
+    code = regexec(&regex, entryName, 2, regMatch, 0);
+
+    if (code == 0) {
+      // match
+      smaInfo("vgId:%d, fetch qtask files, max ver:%" PRIi64 ", %s found", TD_VID(pVnode), version, entryName);
+
+      int64_t ver = -1;
+      sscanf((const char *)POINTER_SHIFT(entryName, regMatch[1].rm_so), "%" PRIi64, &ver);
+      if ((ver <= version) && (ver > -1)) {
+        if (!(*output)) {
+          if (!(*output = taosArrayInit(1, POINTER_BYTES))) {
+            terrno = TSDB_CODE_OUT_OF_MEMORY;
+            goto _end;
+          }
+        }
+        char *entryDup = strdup(entryName);
+        if (!entryDup) {
+          terrno = TSDB_CODE_OUT_OF_MEMORY;
+          goto _end;
+        }
+        if (!taosArrayPush(*output, &entryDup)) {
+          terrno = TSDB_CODE_OUT_OF_MEMORY;
+          goto _end;
+        }
+      } else {
+      }
+    } else if (code == REG_NOMATCH) {
+      // not match
+      smaTrace("vgId:%d, fetch qtask files, not match %s", TD_VID(pVnode), entryName);
+      continue;
+    } else {
+      // has other error
+      char errbuf[128];
+      regerror(code, &regex, errbuf, sizeof(errbuf));
+      smaWarn("vgId:%d, fetch qtask files, regexec failed since %s", TD_VID(pVnode), errbuf);
+      terrno = TSDB_CODE_RSMA_REGEX_MATCH;
+      goto _end;
+    }
+  }
+_end:
+  taosCloseDir(&pDir);
+  regfree(&regex);
+  return terrno == 0 ? TSDB_CODE_SUCCESS : TSDB_CODE_FAILED;
+}
+
+static int32_t tdQTaskFileCmprFn2(const void *p1, const void *p2) {
+  if (((SQTaskFile *)p1)->version < ((SQTaskFile *)p2)->version) {
+    return -1;
+  } else if (((SQTaskFile *)p1)->version > ((SQTaskFile *)p2)->version) {
+    return 1;
+  }
+
+  return 0;
+}
+
+int32_t tdRSmaFSUpsertQTaskFile(SRSmaFS *pFS, SQTaskFile *qTaskFile) {
+  int32_t code = 0;
+  int32_t idx = taosArraySearchIdx(pFS->aQTaskInf, qTaskFile, tdQTaskFileCmprFn2, TD_GE);
+
+  if (idx < 0) {
+    idx = taosArrayGetSize(pFS->aQTaskInf);
+  } else {
+    SQTaskFile *pTaskF = (SQTaskFile *)taosArrayGet(pFS->aQTaskInf, idx);
+    int32_t     c = tdQTaskFileCmprFn2(pTaskF, qTaskFile);
+    if (c == 0) {
+      pTaskF->nRef = qTaskFile->nRef;
+      pTaskF->version = qTaskFile->version;
+      pTaskF->size = qTaskFile->size;
+      goto _exit;
+    }
+  }
+
+  if (taosArrayInsert(pFS->aQTaskInf, idx, qTaskFile) == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _exit;
+  }
+
+_exit:
+  return code;
+}
\ No newline at end of file

source/dnode/vnode/src/sma/smaOpen.c

@@ -150,7 +150,7 @@ int32_t smaOpen(SVnode *pVnode) {
     }
 
     // restore the rsma
-    if (tdRsmaRestore(pSma, RSMA_RESTORE_REBOOT, pVnode->state.committed) < 0) {
+    if (tdRSmaRestore(pSma, RSMA_RESTORE_REBOOT, pVnode->state.committed) < 0) {
       goto _err;
     }
   }
@@ -181,8 +181,8 @@ int32_t smaClose(SSma *pSma) {
  * @param committedVer
  * @return int32_t
  */
-int32_t tdRsmaRestore(SSma *pSma, int8_t type, int64_t committedVer) {
+int32_t tdRSmaRestore(SSma *pSma, int8_t type, int64_t committedVer) {
   ASSERT(VND_IS_RSMA(pSma->pVnode));
 
-  return tdProcessRSmaRestoreImpl(pSma, type, committedVer);
+  return tdRSmaProcessRestoreImpl(pSma, type, committedVer);
 }
\ No newline at end of file

source/dnode/vnode/src/sma/smaRollup.c

@@ -20,7 +20,7 @@
 #define RSMA_QTASKEXEC_SMOOTH_SIZE (100)                                                 // cnt
 #define RSMA_SUBMIT_BATCH_SIZE     (1024)                                                // cnt
 #define RSMA_FETCH_DELAY_MAX       (900000)                                              // ms
-#define RSMA_FETCH_ACTIVE_MAX      (1800)                                                // ms
+#define RSMA_FETCH_ACTIVE_MAX      (1000)                                                // ms
 #define RSMA_FETCH_INTERVAL        (5000)                                                // ms
 
 SSmaMgmt smaMgmt = {
@@ -42,10 +42,12 @@ static int32_t    tdExecuteRSmaImpl(SSma *pSma, const void *pMsg, int32_t msgSiz
 static SRSmaInfo *tdAcquireRSmaInfoBySuid(SSma *pSma, int64_t suid);
 static void       tdReleaseRSmaInfo(SSma *pSma, SRSmaInfo *pInfo);
 static void       tdFreeRSmaSubmitItems(SArray *pItems);
-static int32_t    tdRSmaFetchAllResult(SSma *pSma, SRSmaInfo *pInfo, SArray *pSubmitArr);
+static int32_t    tdRSmaFetchAllResult(SSma *pSma, SRSmaInfo *pInfo);
 static int32_t    tdRSmaExecAndSubmitResult(SSma *pSma, qTaskInfo_t taskInfo, SRSmaInfoItem *pItem, STSchema *pTSchema,
                                             int64_t suid);
 static void       tdRSmaFetchTrigger(void *param, void *tmrId);
+static int32_t    tdRSmaInfoClone(SSma *pSma, SRSmaInfo *pInfo);
+static void       tdRSmaQTaskInfoFree(qTaskInfo_t *taskHandle, int32_t vgId, int32_t level);
 static int32_t    tdRSmaQTaskInfoIterInit(SRSmaQTaskInfoIter *pIter, STFile *pTFile);
 static int32_t    tdRSmaQTaskInfoIterNextBlock(SRSmaQTaskInfoIter *pIter, bool *isFinish);
 static int32_t    tdRSmaQTaskInfoRestore(SSma *pSma, int8_t type, SRSmaQTaskInfoIter *pIter);
@@ -96,7 +98,7 @@ static FORCE_INLINE int32_t tdRSmaQTaskInfoContLen(int32_t lenWithHead) {
 
 static FORCE_INLINE void tdRSmaQTaskInfoIterDestroy(SRSmaQTaskInfoIter *pIter) { taosMemoryFreeClear(pIter->pBuf); }
 
-void tdFreeQTaskInfo(qTaskInfo_t *taskHandle, int32_t vgId, int32_t level) {
+static void tdRSmaQTaskInfoFree(qTaskInfo_t *taskHandle, int32_t vgId, int32_t level) {
   // Note: free/kill may in RC
   if (!taskHandle || !(*taskHandle)) return;
   qTaskInfo_t otaskHandle = atomic_load_ptr(taskHandle);
@@ -119,27 +121,27 @@ void tdFreeQTaskInfo(qTaskInfo_t *taskHandle, int32_t vgId, int32_t level) {
  */
 void *tdFreeRSmaInfo(SSma *pSma, SRSmaInfo *pInfo, bool isDeepFree) {
   if (pInfo) {
+    int32_t vid = pSma ? SMA_VID(pSma) : -1;
     for (int32_t i = 0; i < TSDB_RETENTION_L2; ++i) {
       SRSmaInfoItem *pItem = &pInfo->items[i];
 
       if (isDeepFree && pItem->tmrId) {
-        smaDebug("vgId:%d, stop fetch timer %p for table %" PRIi64 " level %d", SMA_VID(pSma), pInfo->suid,
-                 pItem->tmrId, i + 1);
+        smaDebug("vgId:%d, stop fetch timer %p for table %" PRIi64 " level %d", vid, pItem->tmrId, pInfo->suid, i + 1);
         taosTmrStopA(&pItem->tmrId);
       }
 
       if (isDeepFree && pInfo->taskInfo[i]) {
-        tdFreeQTaskInfo(&pInfo->taskInfo[i], SMA_VID(pSma), i + 1);
+        tdRSmaQTaskInfoFree(&pInfo->taskInfo[i], vid, i + 1);
       } else {
-        smaDebug("vgId:%d, table %" PRIi64 " no need to destroy rsma info level %d since empty taskInfo", SMA_VID(pSma),
+        smaDebug("vgId:%d, table %" PRIi64 " no need to destroy rsma info level %d since empty taskInfo", vid,
                  pInfo->suid, i + 1);
       }
 
       if (pInfo->iTaskInfo[i]) {
-        tdFreeQTaskInfo(&pInfo->iTaskInfo[i], SMA_VID(pSma), i + 1);
+        tdRSmaQTaskInfoFree(&pInfo->iTaskInfo[i], vid, i + 1);
       } else {
-        smaDebug("vgId:%d, table %" PRIi64 " no need to destroy rsma info level %d since empty iTaskInfo",
-                 SMA_VID(pSma), pInfo->suid, i + 1);
+        smaDebug("vgId:%d, table %" PRIi64 " no need to destroy rsma info level %d since empty iTaskInfo", vid,
+                 pInfo->suid, i + 1);
       }
     }
     if (isDeepFree) {
@@ -313,10 +315,17 @@ static int32_t tdSetRSmaInfoItemParams(SSma *pSma, SRSmaParam *param, SRSmaStat
     }
 
     pItem->level = idx == 0 ? TSDB_RETENTION_L1 : TSDB_RETENTION_L2;
+    ASSERT(pItem->level > 0);
+
+    SRSmaRef rsmaRef = {.refId = pStat->refId, .suid = pRSmaInfo->suid};
+    taosHashPut(smaMgmt.refHash, &pItem, POINTER_BYTES, &rsmaRef, sizeof(rsmaRef));
+
     taosTmrReset(tdRSmaFetchTrigger, pItem->maxDelay, pItem, smaMgmt.tmrHandle, &pItem->tmrId);
-    smaInfo("vgId:%d, table:%" PRIi64 " level:%" PRIi8 " maxdelay:%" PRIi64 " watermark:%" PRIi64
+
+    smaInfo("vgId:%d, item:%p table:%" PRIi64 " level:%" PRIi8 " maxdelay:%" PRIi64 " watermark:%" PRIi64
             ", finally maxdelay:%" PRIi32,
-            TD_VID(pVnode), pRSmaInfo->suid, idx + 1, param->maxdelay[idx], param->watermark[idx], pItem->maxDelay);
+            TD_VID(pVnode), pItem, pRSmaInfo->suid, idx + 1, param->maxdelay[idx], param->watermark[idx],
+            pItem->maxDelay);
   }
   return TSDB_CODE_SUCCESS;
 }
@@ -330,7 +339,7 @@ static int32_t tdSetRSmaInfoItemParams(SSma *pSma, SRSmaParam *param, SRSmaStat
  * @param tbName
  * @return int32_t
  */
-int32_t tdProcessRSmaCreateImpl(SSma *pSma, SRSmaParam *param, int64_t suid, const char *tbName) {
+int32_t tdRSmaProcessCreateImpl(SSma *pSma, SRSmaParam *param, int64_t suid, const char *tbName) {
   if ((param->qmsgLen[0] == 0) && (param->qmsgLen[1] == 0)) {
     smaDebug("vgId:%d, no qmsg1/qmsg2 for rollup table %s %" PRIi64, SMA_VID(pSma), tbName, suid);
     return TSDB_CODE_SUCCESS;
@@ -383,7 +392,6 @@ int32_t tdProcessRSmaCreateImpl(SSma *pSma, SRSmaParam *param, int64_t suid, con
     goto _err;
   }
   pRSmaInfo->suid = suid;
-  pRSmaInfo->refId = RSMA_REF_ID(pStat);
   T_REF_INIT_VAL(pRSmaInfo, 1);
 
   if (tdSetRSmaInfoItemParams(pSma, param, pStat, pRSmaInfo, 0) < 0) {
@@ -427,7 +435,7 @@ int32_t tdProcessRSmaCreate(SSma *pSma, SVCreateStbReq *pReq) {
     return TSDB_CODE_SUCCESS;
   }
 
-  return tdProcessRSmaCreateImpl(pSma, &pReq->rsmaParam, pReq->suid, pReq->name);
+  return tdRSmaProcessCreateImpl(pSma, &pReq->rsmaParam, pReq->suid, pReq->name);
 }
 
 /**
@@ -654,15 +662,15 @@ static int32_t tdRSmaExecAndSubmitResult(SSma *pSma, qTaskInfo_t taskInfo, SRSma
 
     if (taosArrayGetSize(pResList) == 0) {
       if (terrno == 0) {
-        // smaDebug("vgId:%d, no rsma %" PRIi8 " data fetched yet", SMA_VID(pSma), pItem->level);
+        // smaDebug("vgId:%d, no rsma level %" PRIi8 " data fetched yet", SMA_VID(pSma), pItem->level);
       } else {
-        smaDebug("vgId:%d, no rsma %" PRIi8 " data fetched since %s", SMA_VID(pSma), pItem->level, terrstr());
+        smaDebug("vgId:%d, no rsma level %" PRIi8 " data fetched since %s", SMA_VID(pSma), pItem->level, terrstr());
         goto _err;
       }
 
       break;
     } else {
-      smaDebug("vgId:%d, rsma %" PRIi8 " data fetched", SMA_VID(pSma), pItem->level);
+      smaDebug("vgId:%d, rsma level %" PRIi8 " data fetched", SMA_VID(pSma), pItem->level);
     }
 #if 0
     char flag[10] = {0};
@@ -676,21 +684,22 @@ static int32_t tdRSmaExecAndSubmitResult(SSma *pSma, qTaskInfo_t taskInfo, SRSma
 
       // TODO: the schema update should be handled later(TD-17965)
       if (buildSubmitReqFromDataBlock(&pReq, output, pTSchema, SMA_VID(pSma), suid) < 0) {
-        smaError("vgId:%d, build submit req for rsma stable %" PRIi64 " level %" PRIi8 " failed since %s",
-                 SMA_VID(pSma), suid, pItem->level, terrstr());
+        smaError("vgId:%d, build submit req for rsma table %" PRIi64 " level %" PRIi8 " failed since %s", SMA_VID(pSma),
+                 suid, pItem->level, terrstr());
         goto _err;
       }
 
       if (pReq && tdProcessSubmitReq(sinkTsdb, output->info.version, pReq) < 0) {
         taosMemoryFreeClear(pReq);
-        smaError("vgId:%d, process submit req for rsma stable %" PRIi64 " level %" PRIi8 " failed since %s",
+        smaError("vgId:%d, process submit req for rsma table %" PRIi64 " level %" PRIi8 " failed since %s",
                  SMA_VID(pSma), suid, pItem->level, terrstr());
         goto _err;
       }
-      taosMemoryFreeClear(pReq);
 
-      smaDebug("vgId:%d, process submit req for rsma table %" PRIi64 " level %" PRIi8 " version:%" PRIi64,
-               SMA_VID(pSma), suid, pItem->level, output->info.version);
+      smaDebug("vgId:%d, process submit req for rsma table %" PRIi64 " level %" PRIi8 " ver %" PRIi64 " len %" PRIu32,
+               SMA_VID(pSma), suid, pItem->level, output->info.version, htonl(pReq->header.contLen));
+
+      taosMemoryFreeClear(pReq);
     }
   }
 
@@ -817,6 +826,95 @@ static int32_t tdExecuteRSmaImpl(SSma *pSma, const void *pMsg, int32_t msgSize,
   return TSDB_CODE_SUCCESS;
 }
 
+static int32_t tdCloneQTaskInfo(SSma *pSma, qTaskInfo_t dstTaskInfo, qTaskInfo_t srcTaskInfo, SRSmaParam *param,
+                                tb_uid_t suid, int8_t idx) {
+  SVnode *pVnode = pSma->pVnode;
+  char   *pOutput = NULL;
+  int32_t len = 0;
+
+  if ((terrno = qSerializeTaskStatus(srcTaskInfo, &pOutput, &len)) < 0) {
+    smaError("vgId:%d, rsma clone, table %" PRIi64 " serialize qTaskInfo failed since %s", TD_VID(pVnode), suid,
+             terrstr());
+    goto _err;
+  }
+
+  SReadHandle handle = {
+      .meta = pVnode->pMeta,
+      .vnode = pVnode,
+      .initTqReader = 1,
+  };
+  ASSERT(!dstTaskInfo);
+  dstTaskInfo = qCreateStreamExecTaskInfo(param->qmsg[idx], &handle);
+  if (!dstTaskInfo) {
+    terrno = TSDB_CODE_RSMA_QTASKINFO_CREATE;
+    goto _err;
+  }
+
+  if (qDeserializeTaskStatus(dstTaskInfo, pOutput, len) < 0) {
+    smaError("vgId:%d, rsma clone, restore rsma task for table:%" PRIi64 " failed since %s", TD_VID(pVnode), suid,
+             terrstr());
+    goto _err;
+  }
+
+  smaDebug("vgId:%d, rsma clone, restore rsma task for table:%" PRIi64 " succeed", TD_VID(pVnode), suid);
+
+  taosMemoryFreeClear(pOutput);
+  return TSDB_CODE_SUCCESS;
+_err:
+  taosMemoryFreeClear(pOutput);
+  tdRSmaQTaskInfoFree(dstTaskInfo, TD_VID(pVnode), idx + 1);
+  smaError("vgId:%d, rsma clone, restore rsma task for table:%" PRIi64 " failed since %s", TD_VID(pVnode), suid,
+           terrstr());
+  return TSDB_CODE_FAILED;
+}
+
+/**
+ * @brief Clone qTaskInfo of SRSmaInfo
+ *
+ * @param pSma
+ * @param pInfo
+ * @return int32_t
+ */
+static int32_t tdRSmaInfoClone(SSma *pSma, SRSmaInfo *pInfo) {
+  SRSmaParam *param = NULL;
+  if (!pInfo) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  SMetaReader mr = {0};
+  metaReaderInit(&mr, SMA_META(pSma), 0);
+  smaDebug("vgId:%d, rsma clone qTaskInfo for suid:%" PRIi64, SMA_VID(pSma), pInfo->suid);
+  if (metaGetTableEntryByUid(&mr, pInfo->suid) < 0) {
+    smaError("vgId:%d, rsma clone, failed to get table meta for %" PRIi64 " since %s", SMA_VID(pSma), pInfo->suid,
+             terrstr());
+    goto _err;
+  }
+  ASSERT(mr.me.type == TSDB_SUPER_TABLE);
+  ASSERT(mr.me.uid == pInfo->suid);
+  if (TABLE_IS_ROLLUP(mr.me.flags)) {
+    param = &mr.me.stbEntry.rsmaParam;
+    for (int32_t i = 0; i < TSDB_RETENTION_L2; ++i) {
+      if (!pInfo->iTaskInfo[i]) {
+        continue;
+      }
+      if (tdCloneQTaskInfo(pSma, pInfo->taskInfo[i], pInfo->iTaskInfo[i], param, pInfo->suid, i) < 0) {
+        goto _err;
+      }
+    }
+    smaDebug("vgId:%d, rsma clone env success for %" PRIi64, SMA_VID(pSma), pInfo->suid);
+  } else {
+    terrno = TSDB_CODE_RSMA_INVALID_SCHEMA;
+    goto _err;
+  }
+
+  metaReaderClear(&mr);
+  return TSDB_CODE_SUCCESS;
+_err:
+  metaReaderClear(&mr);
+  smaError("vgId:%d, rsma clone env failed for %" PRIi64 " since %s", SMA_VID(pSma), pInfo->suid, terrstr());
+  return TSDB_CODE_FAILED;
+}
+
 /**
  * @brief During async commit, the SRSmaInfo object would be COW from iRSmaInfoHash and write lock should be applied.
  *
@@ -848,7 +946,7 @@ static SRSmaInfo *tdAcquireRSmaInfoBySuid(SSma *pSma, int64_t suid) {
       return NULL;
     }
     if (!pRSmaInfo->taskInfo[0]) {
-      if (tdCloneRSmaInfo(pSma, pRSmaInfo) < 0) {
+      if (tdRSmaInfoClone(pSma, pRSmaInfo) < 0) {
         // taosRUnLockLatch(SMA_ENV_LOCK(pEnv));
         return NULL;
       }
@@ -1006,7 +1104,7 @@ static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma, int64_t *nTables) {
                  " qmsgLen:%" PRIi32,
                  TD_VID(pVnode), suid, i, param->maxdelay[i], param->watermark[i], param->qmsgLen[i]);
       }
-      if (tdProcessRSmaCreateImpl(pSma, &mr.me.stbEntry.rsmaParam, suid, mr.me.name) < 0) {
+      if (tdRSmaProcessCreateImpl(pSma, &mr.me.stbEntry.rsmaParam, suid, mr.me.name) < 0) {
         smaError("vgId:%d, rsma restore env failed for %" PRIi64 " since %s", TD_VID(pVnode), suid, terrstr());
         goto _err;
       }
@@ -1118,7 +1216,7 @@ static int32_t tdRSmaRestoreTSDataReload(SSma *pSma) {
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t tdProcessRSmaRestoreImpl(SSma *pSma, int8_t type, int64_t qtaskFileVer) {
+int32_t tdRSmaProcessRestoreImpl(SSma *pSma, int8_t type, int64_t qtaskFileVer) {
   // step 1: iterate all stables to restore the rsma env
   int64_t nTables = 0;
   if (tdRSmaRestoreQTaskInfoInit(pSma, &nTables) < 0) {
@@ -1139,6 +1237,12 @@ int32_t tdProcessRSmaRestoreImpl(SSma *pSma, int8_t type, int64_t qtaskFileVer)
   if (tdRSmaRestoreTSDataReload(pSma) < 0) {
     goto _err;
   }
+
+  // step 4: open SRSmaFS for qTaskFiles
+  if (tdRSmaFSOpen(pSma, qtaskFileVer) < 0) {
+    goto _err;
+  }
+
   smaInfo("vgId:%d, restore rsma task %" PRIi8 " from qtaskf %" PRIi64 " succeed", SMA_VID(pSma), type, qtaskFileVer);
   return TSDB_CODE_SUCCESS;
 _err:
@@ -1468,26 +1572,46 @@ _err:
  * @param tmrId
  */
 static void tdRSmaFetchTrigger(void *param, void *tmrId) {
-  SRSmaInfoItem *pItem = param;
+  SRSmaRef      *pRSmaRef = NULL;
   SSma          *pSma = NULL;
-  SRSmaInfo     *pRSmaInfo = tdGetRSmaInfoByItem(pItem);
+  SRSmaStat     *pStat = NULL;
+  SRSmaInfo     *pRSmaInfo = NULL;
+  SRSmaInfoItem *pItem = NULL;
 
-  if (RSMA_INFO_IS_DEL(pRSmaInfo)) {
-    smaDebug("rsma fetch task not start since rsma info already deleted, rsetId:%" PRIi64 " refId:%d)", smaMgmt.rsetId,
-             pRSmaInfo->refId);
+  if (!(pRSmaRef = taosHashGet(smaMgmt.refHash, &param, POINTER_BYTES))) {
+    smaDebug("rsma fetch task not start since rsma info item:%p not exist in refHash:%p, rsetId:%" PRIi64, param,
+             *(int64_t *)&param, smaMgmt.refHash, smaMgmt.rsetId);
     return;
   }
 
-  SRSmaStat *pStat = (SRSmaStat *)tdAcquireSmaRef(smaMgmt.rsetId, pRSmaInfo->refId);
-
-  if (!pStat) {
+  if (!(pStat = (SRSmaStat *)tdAcquireSmaRef(smaMgmt.rsetId, pRSmaRef->refId))) {
     smaDebug("rsma fetch task not start since rsma stat already destroyed, rsetId:%" PRIi64 " refId:%d)",
-             smaMgmt.rsetId, pRSmaInfo->refId);
+             smaMgmt.rsetId, pRSmaRef->refId); // pRSmaRef freed in taosHashRemove
+    taosHashRemove(smaMgmt.refHash, &param, POINTER_BYTES); 
     return;
   }
 
   pSma = pStat->pSma;
 
+  if (!(pRSmaInfo = tdAcquireRSmaInfoBySuid(pSma, pRSmaRef->suid))) {
+    smaDebug("rsma fetch task not start since rsma info not exist, rsetId:%" PRIi64 " refId:%d)", smaMgmt.rsetId,
+             pRSmaRef->refId);  // pRSmaRef freed in taosHashRemove
+    tdReleaseSmaRef(smaMgmt.rsetId, pRSmaRef->refId);
+    taosHashRemove(smaMgmt.refHash, &param, POINTER_BYTES);
+    return;
+  }
+
+  if (RSMA_INFO_IS_DEL(pRSmaInfo)) {
+    smaDebug("rsma fetch task not start since rsma info already deleted, rsetId:%" PRIi64 " refId:%d)", smaMgmt.rsetId,
+             pRSmaRef->refId);  // pRSmaRef freed in taosHashRemove
+    tdReleaseRSmaInfo(pSma, pRSmaInfo);
+    tdReleaseSmaRef(smaMgmt.rsetId, pRSmaRef->refId);
+    taosHashRemove(smaMgmt.refHash, &param, POINTER_BYTES);
+    return;
+  }
+
+  pItem = *(SRSmaInfoItem **)&param;
+
   // if rsma trigger stat in paused, cancelled or finished, not start fetch task
   int8_t rsmaTriggerStat = atomic_load_8(RSMA_TRIGGER_STAT(pStat));
   switch (rsmaTriggerStat) {
@@ -1495,11 +1619,12 @@ static void tdRSmaFetchTrigger(void *param, void *tmrId) {
     case TASK_TRIGGER_STAT_CANCELLED: {
       smaDebug("vgId:%d, rsma fetch task not start for level %" PRIi8 " since stat is %" PRIi8
                ", rsetId rsetId:%" PRIi64 " refId:%d",
-               SMA_VID(pSma), pItem->level, rsmaTriggerStat, smaMgmt.rsetId, pRSmaInfo->refId);
+               SMA_VID(pSma), pItem->level, rsmaTriggerStat, smaMgmt.rsetId, pRSmaRef->refId);
       if (rsmaTriggerStat == TASK_TRIGGER_STAT_PAUSED) {
         taosTmrReset(tdRSmaFetchTrigger, RSMA_FETCH_INTERVAL, pItem, smaMgmt.tmrHandle, &pItem->tmrId);
       }
-      tdReleaseSmaRef(smaMgmt.rsetId, pRSmaInfo->refId);
+      tdReleaseRSmaInfo(pSma, pRSmaInfo);
+      tdReleaseSmaRef(smaMgmt.rsetId, pRSmaRef->refId);
       return;
     }
     default:
@@ -1511,7 +1636,7 @@ static void tdRSmaFetchTrigger(void *param, void *tmrId) {
   switch (fetchTriggerStat) {
     case TASK_TRIGGER_STAT_ACTIVE: {
       smaDebug("vgId:%d, rsma fetch task planned for level:%" PRIi8 " suid:%" PRIi64 " since stat is active",
-               SMA_VID(pSma), pItem->level, pRSmaInfo->suid);
+              SMA_VID(pSma), pItem->level, pRSmaInfo->suid);
       // async process
       pItem->fetchLevel = pItem->level;
 #if 0
@@ -1544,11 +1669,11 @@ static void tdRSmaFetchTrigger(void *param, void *tmrId) {
 
 _end:
   taosTmrReset(tdRSmaFetchTrigger, pItem->maxDelay, pItem, smaMgmt.tmrHandle, &pItem->tmrId);
-  tdReleaseSmaRef(smaMgmt.rsetId, pRSmaInfo->refId);
+  tdReleaseRSmaInfo(pSma, pRSmaInfo);
+  tdReleaseSmaRef(smaMgmt.rsetId, pRSmaRef->refId);
 }
 
 static void tdFreeRSmaSubmitItems(SArray *pItems) {
-  ASSERT(taosArrayGetSize(pItems) > 0);
   for (int32_t i = 0; i < taosArrayGetSize(pItems); ++i) {
     taosFreeQitem(*(void **)taosArrayGet(pItems, i));
   }
@@ -1560,10 +1685,9 @@ static void tdFreeRSmaSubmitItems(SArray *pItems) {
  *
  * @param pSma
  * @param pInfo
- * @param pSubmitArr
  * @return int32_t
  */
-static int32_t tdRSmaFetchAllResult(SSma *pSma, SRSmaInfo *pInfo, SArray *pSubmitArr) {
+static int32_t tdRSmaFetchAllResult(SSma *pSma, SRSmaInfo *pInfo) {
   SSDataBlock dataBlock = {.info.type = STREAM_GET_ALL};
   for (int8_t i = 1; i <= TSDB_RETENTION_L2; ++i) {
     SRSmaInfoItem *pItem = RSMA_INFO_ITEM(pInfo, i - 1);
@@ -1609,10 +1733,8 @@ static int32_t tdRSmaFetchAllResult(SSma *pSma, SRSmaInfo *pInfo, SArray *pSubmi
   }
 
 _end:
-  tdReleaseRSmaInfo(pSma, pInfo);
   return TSDB_CODE_SUCCESS;
 _err:
-  tdReleaseRSmaInfo(pSma, pInfo);
   return TSDB_CODE_FAILED;
 }
 
@@ -1709,7 +1831,7 @@ int32_t tdRSmaProcessExecImpl(SSma *pSma, ERsmaExecType type) {
               if (oldStat == 0 ||
                   ((oldStat == 2) && atomic_load_8(RSMA_TRIGGER_STAT(pRSmaStat)) < TASK_TRIGGER_STAT_PAUSED)) {
                 atomic_fetch_add_32(&pRSmaStat->nFetchAll, 1);
-                tdRSmaFetchAllResult(pSma, pInfo, pSubmitArr);
+                tdRSmaFetchAllResult(pSma, pInfo);
                 if (0 == atomic_sub_fetch_32(&pRSmaStat->nFetchAll, 1)) {
                   atomic_store_8(RSMA_COMMIT_STAT(pRSmaStat), 0);
                 }

source/dnode/vnode/src/sma/smaSnapshot.c

@@ -15,11 +15,13 @@
 
 #include "sma.h"
 
-static int32_t rsmaSnapReadQTaskInfo(SRsmaSnapReader* pReader, uint8_t** ppData);
-static int32_t rsmaSnapWriteQTaskInfo(SRsmaSnapWriter* pWriter, uint8_t* pData, uint32_t nData);
+static int32_t rsmaSnapReadQTaskInfo(SRSmaSnapReader* pReader, uint8_t** ppData);
+static int32_t rsmaSnapWriteQTaskInfo(SRSmaSnapWriter* pWriter, uint8_t* pData, uint32_t nData);
+static int32_t rsmaQTaskInfSnapReaderOpen(SRSmaSnapReader* pReader, int64_t version);
+static int32_t rsmaQTaskInfSnapReaderClose(SQTaskFReader** ppReader);
 
-// SRsmaSnapReader ========================================
-struct SRsmaSnapReader {
+// SRSmaSnapReader ========================================
+struct SRSmaSnapReader {
   SSma*   pSma;
   int64_t sver;
   int64_t ever;
@@ -33,13 +35,13 @@ struct SRsmaSnapReader {
   SQTaskFReader* pQTaskFReader;
 };
 
-int32_t rsmaSnapReaderOpen(SSma* pSma, int64_t sver, int64_t ever, SRsmaSnapReader** ppReader) {
+int32_t rsmaSnapReaderOpen(SSma* pSma, int64_t sver, int64_t ever, SRSmaSnapReader** ppReader) {
   int32_t          code = 0;
   SVnode*          pVnode = pSma->pVnode;
-  SRsmaSnapReader* pReader = NULL;
+  SRSmaSnapReader* pReader = NULL;
 
   // alloc
-  pReader = (SRsmaSnapReader*)taosMemoryCalloc(1, sizeof(*pReader));
+  pReader = (SRSmaSnapReader*)taosMemoryCalloc(1, sizeof(*pReader));
   if (pReader == NULL) {
     code = TSDB_CODE_OUT_OF_MEMORY;
     goto _err;
@@ -48,7 +50,7 @@ int32_t rsmaSnapReaderOpen(SSma* pSma, int64_t sver, int64_t ever, SRsmaSnapRead
   pReader->sver = sver;
   pReader->ever = ever;
 
-  // rsma1/rsma2
+  // open rsma1/rsma2
   for (int32_t i = 0; i < TSDB_RETENTION_L2; ++i) {
     if (pSma->pRSmaTsdb[i]) {
       code = tsdbSnapReaderOpen(pSma->pRSmaTsdb[i], sver, ever, i == 0 ? SNAP_DATA_RSMA1 : SNAP_DATA_RSMA2,
@@ -59,51 +61,112 @@ int32_t rsmaSnapReaderOpen(SSma* pSma, int64_t sver, int64_t ever, SRsmaSnapRead
     }
   }
 
-  // qtaskinfo
-  // 1. add ref to qtaskinfo.v${ever} if exists and then start to replicate
+  // open qtaskinfo
+  if ((code = rsmaQTaskInfSnapReaderOpen(pReader, ever)) < 0) {
+    goto _err;
+  }
+
+  *ppReader = pReader;
+
+  return TSDB_CODE_SUCCESS;
+_err:
+  smaError("vgId:%d, vnode snapshot rsma reader open failed since %s", TD_VID(pVnode), tstrerror(code));
+  return TSDB_CODE_FAILED;
+}
+
+static int32_t rsmaQTaskInfSnapReaderOpen(SRSmaSnapReader* pReader, int64_t version) {
+  int32_t    code = 0;
+  SSma*      pSma = pReader->pSma;
+  SVnode*    pVnode = pSma->pVnode;
+  SSmaEnv*   pEnv = NULL;
+  SRSmaStat* pStat = NULL;
+
+  if (!(pEnv = SMA_RSMA_ENV(pSma))) {
+    smaInfo("vgId:%d, vnode snapshot rsma reader for qtaskinfo version %" PRIi64 " not need as env is NULL",
+            TD_VID(pVnode), version);
+    return TSDB_CODE_SUCCESS;
+  }
+
+  pStat = (SRSmaStat*)SMA_ENV_STAT(pEnv);
+
+  int32_t ref = tdRSmaFSRef(pReader->pSma, pStat, version);
+  if (ref < 1) {
+    smaInfo("vgId:%d, vnode snapshot rsma reader for qtaskinfo version %" PRIi64 " not need as ref is %d",
+            TD_VID(pVnode), version, ref);
+    return TSDB_CODE_SUCCESS;
+  }
+
   char qTaskInfoFullName[TSDB_FILENAME_LEN];
-  tdRSmaQTaskInfoGetFullName(TD_VID(pVnode), ever, tfsGetPrimaryPath(pVnode->pTfs), qTaskInfoFullName);
+  tdRSmaQTaskInfoGetFullName(TD_VID(pVnode), version, tfsGetPrimaryPath(pVnode->pTfs), qTaskInfoFullName);
 
   if (!taosCheckExistFile(qTaskInfoFullName)) {
-    smaInfo("vgId:%d, vnode snapshot rsma reader for qtaskinfo not need as %s not exists", TD_VID(pVnode),
-            qTaskInfoFullName);
-  } else {
-    pReader->pQTaskFReader = taosMemoryCalloc(1, sizeof(SQTaskFReader));
-    if (!pReader->pQTaskFReader) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _err;
-    }
+    tdRSmaFSUnRef(pSma, pStat, version);
+    smaInfo("vgId:%d, vnode snapshot rsma reader for qtaskinfo version %" PRIi64 " not need as %s not exists",
+            TD_VID(pVnode), qTaskInfoFullName);
+    return TSDB_CODE_SUCCESS;
+  }
 
-    TdFilePtr qTaskF = taosOpenFile(qTaskInfoFullName, TD_FILE_READ);
-    if (!qTaskF) {
-      code = TAOS_SYSTEM_ERROR(errno);
-      goto _err;
-    }
-    pReader->pQTaskFReader->pReadH = qTaskF;
-#if 0
-    SQTaskFile* pQTaskF = &pReader->pQTaskFReader->fTask;
-    pQTaskF->nRef = 1;
-#endif
+  pReader->pQTaskFReader = taosMemoryCalloc(1, sizeof(SQTaskFReader));
+  if (!pReader->pQTaskFReader) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _end;
   }
 
-  *ppReader = pReader;
-  smaInfo("vgId:%d, vnode snapshot rsma reader opened %s succeed", TD_VID(pVnode), qTaskInfoFullName);
+  TdFilePtr fp = taosOpenFile(qTaskInfoFullName, TD_FILE_READ);
+  if (!fp) {
+    code = TAOS_SYSTEM_ERROR(errno);
+    taosMemoryFreeClear(pReader->pQTaskFReader);
+    goto _end;
+  }
+
+  pReader->pQTaskFReader->pReadH = fp;
+  pReader->pQTaskFReader->pSma = pSma;
+  pReader->pQTaskFReader->version = pReader->ever;
+
+_end:
+  if (code < 0) {
+    tdRSmaFSUnRef(pSma, pStat, version);
+    smaError("vgId:%d, vnode snapshot rsma reader open %s succeed", TD_VID(pVnode), qTaskInfoFullName);
+    return TSDB_CODE_FAILED;
+  }
+
+  smaInfo("vgId:%d, vnode snapshot rsma reader open %s succeed", TD_VID(pVnode), qTaskInfoFullName);
+  return TSDB_CODE_SUCCESS;
+}
+
+static int32_t rsmaQTaskInfSnapReaderClose(SQTaskFReader** ppReader) {
+  if (!(*ppReader)) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  SSma*      pSma = (*ppReader)->pSma;
+  SRSmaStat* pStat = SMA_RSMA_STAT(pSma);
+  int64_t    version = (*ppReader)->version;
+
+  taosCloseFile(&(*ppReader)->pReadH);
+  tdRSmaFSUnRef(pSma, pStat, version);
+  taosMemoryFreeClear(*ppReader);
+  smaInfo("vgId:%d, vnode snapshot rsma reader closed for qTaskInfo version %" PRIi64, SMA_VID(pSma), version);
+
   return TSDB_CODE_SUCCESS;
-_err:
-  smaError("vgId:%d, vnode snapshot rsma reader opened failed since %s", TD_VID(pVnode), tstrerror(code));
-  return TSDB_CODE_FAILED;
 }
 
-static int32_t rsmaSnapReadQTaskInfo(SRsmaSnapReader* pReader, uint8_t** ppBuf) {
+static int32_t rsmaSnapReadQTaskInfo(SRSmaSnapReader* pReader, uint8_t** ppBuf) {
   int32_t        code = 0;
   SSma*          pSma = pReader->pSma;
   int64_t        n = 0;
   uint8_t*       pBuf = NULL;
   SQTaskFReader* qReader = pReader->pQTaskFReader;
 
+  if (!qReader) {
+    *ppBuf = NULL;
+    smaInfo("vgId:%d, vnode snapshot rsma reader qtaskinfo, qTaskReader is NULL", SMA_VID(pSma));
+    return 0;
+  }
+
   if (!qReader->pReadH) {
     *ppBuf = NULL;
-    smaInfo("vgId:%d, vnode snapshot rsma reader qtaskinfo, readh is empty", SMA_VID(pSma));
+    smaInfo("vgId:%d, vnode snapshot rsma reader qtaskinfo, readh is NULL", SMA_VID(pSma));
     return 0;
   }
 
@@ -153,7 +216,7 @@ _err:
   return code;
 }
 
-int32_t rsmaSnapRead(SRsmaSnapReader* pReader, uint8_t** ppData) {
+int32_t rsmaSnapRead(SRSmaSnapReader* pReader, uint8_t** ppData) {
   int32_t code = 0;
 
   *ppData = NULL;
@@ -205,9 +268,9 @@ _err:
   return code;
 }
 
-int32_t rsmaSnapReaderClose(SRsmaSnapReader** ppReader) {
+int32_t rsmaSnapReaderClose(SRSmaSnapReader** ppReader) {
   int32_t          code = 0;
-  SRsmaSnapReader* pReader = *ppReader;
+  SRSmaSnapReader* pReader = *ppReader;
 
   for (int32_t i = 0; i < TSDB_RETENTION_L2; ++i) {
     if (pReader->pDataReader[i]) {
@@ -215,11 +278,7 @@ int32_t rsmaSnapReaderClose(SRsmaSnapReader** ppReader) {
     }
   }
 
-  if (pReader->pQTaskFReader) {
-    taosCloseFile(&pReader->pQTaskFReader->pReadH);
-    taosMemoryFreeClear(pReader->pQTaskFReader);
-    smaInfo("vgId:%d, vnode snapshot rsma reader closed for qTaskInfo", SMA_VID(pReader->pSma));
-  }
+  rsmaQTaskInfSnapReaderClose(&pReader->pQTaskFReader);
 
   smaInfo("vgId:%d, vnode snapshot rsma reader closed", SMA_VID(pReader->pSma));
 
@@ -227,8 +286,8 @@ int32_t rsmaSnapReaderClose(SRsmaSnapReader** ppReader) {
   return code;
 }
 
-// SRsmaSnapWriter ========================================
-struct SRsmaSnapWriter {
+// SRSmaSnapWriter ========================================
+struct SRSmaSnapWriter {
   SSma*   pSma;
   int64_t sver;
   int64_t ever;
@@ -244,13 +303,13 @@ struct SRsmaSnapWriter {
   SQTaskFWriter* pQTaskFWriter;
 };
 
-int32_t rsmaSnapWriterOpen(SSma* pSma, int64_t sver, int64_t ever, SRsmaSnapWriter** ppWriter) {
+int32_t rsmaSnapWriterOpen(SSma* pSma, int64_t sver, int64_t ever, SRSmaSnapWriter** ppWriter) {
   int32_t          code = 0;
-  SRsmaSnapWriter* pWriter = NULL;
+  SRSmaSnapWriter* pWriter = NULL;
   SVnode*          pVnode = pSma->pVnode;
 
   // alloc
-  pWriter = (SRsmaSnapWriter*)taosMemoryCalloc(1, sizeof(*pWriter));
+  pWriter = (SRSmaSnapWriter*)taosMemoryCalloc(1, sizeof(*pWriter));
   if (pWriter == NULL) {
     code = TSDB_CODE_OUT_OF_MEMORY;
     goto _err;
@@ -301,9 +360,9 @@ _err:
   return code;
 }
 
-int32_t rsmaSnapWriterClose(SRsmaSnapWriter** ppWriter, int8_t rollback) {
+int32_t rsmaSnapWriterClose(SRSmaSnapWriter** ppWriter, int8_t rollback) {
   int32_t          code = 0;
-  SRsmaSnapWriter* pWriter = *ppWriter;
+  SRSmaSnapWriter* pWriter = *ppWriter;
   SVnode*          pVnode = pWriter->pSma->pVnode;
 
   if (rollback) {
@@ -332,7 +391,7 @@ int32_t rsmaSnapWriterClose(SRsmaSnapWriter** ppWriter, int8_t rollback) {
               pWriter->pQTaskFWriter->fname, qTaskInfoFullName);
 
       // rsma restore
-      if ((code = tdRsmaRestore(pWriter->pSma, RSMA_RESTORE_SYNC, pWriter->ever)) < 0) {
+      if ((code = tdRSmaRestore(pWriter->pSma, RSMA_RESTORE_SYNC, pWriter->ever)) < 0) {
         goto _err;
       }
       smaInfo("vgId:%d, vnode snapshot rsma writer restore from %s succeed", SMA_VID(pWriter->pSma), qTaskInfoFullName);
@@ -349,7 +408,7 @@ _err:
   return code;
 }
 
-int32_t rsmaSnapWrite(SRsmaSnapWriter* pWriter, uint8_t* pData, uint32_t nData) {
+int32_t rsmaSnapWrite(SRSmaSnapWriter* pWriter, uint8_t* pData, uint32_t nData) {
   int32_t       code = 0;
   SSnapDataHdr* pHdr = (SSnapDataHdr*)pData;
 
@@ -377,7 +436,7 @@ _err:
   return code;
 }
 
-static int32_t rsmaSnapWriteQTaskInfo(SRsmaSnapWriter* pWriter, uint8_t* pData, uint32_t nData) {
+static int32_t rsmaSnapWriteQTaskInfo(SRSmaSnapWriter* pWriter, uint8_t* pData, uint32_t nData) {
   int32_t        code = 0;
   SQTaskFWriter* qWriter = pWriter->pQTaskFWriter;
 

source/dnode/vnode/src/sma/smaTimeRange.c

@@ -20,6 +20,10 @@
 #define SMA_STORAGE_MINUTES_DAY  1440
 #define SMA_STORAGE_SPLIT_FACTOR 14400  // least records in tsma file
 
+static int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg);
+static int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg);
+static int32_t tdProcessTSmaGetDaysImpl(SVnodeCfg *pCfg, void *pCont, uint32_t contLen, int32_t *days);
+
 // TODO: Who is responsible for resource allocate and release?
 int32_t tdProcessTSmaInsert(SSma *pSma, int64_t indexUid, const char *msg) {
   int32_t code = TSDB_CODE_SUCCESS;
@@ -59,7 +63,7 @@ int32_t smaGetTSmaDays(SVnodeCfg *pCfg, void *pCont, uint32_t contLen, int32_t *
  * @param days unit is minute
  * @return int32_t
  */
-int32_t tdProcessTSmaGetDaysImpl(SVnodeCfg *pCfg, void *pCont, uint32_t contLen, int32_t *days) {
+static int32_t tdProcessTSmaGetDaysImpl(SVnodeCfg *pCfg, void *pCont, uint32_t contLen, int32_t *days) {
   SDecoder coder = {0};
   tDecoderInit(&coder, pCont, contLen);
 
@@ -106,7 +110,7 @@ _err:
  * @param pMsg
  * @return int32_t
  */
-int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg) {
+static int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg) {
   SSmaCfg *pCfg = (SSmaCfg *)pMsg;
 
   if (TD_VID(pSma->pVnode) == pCfg->dstVgId) {
@@ -145,7 +149,7 @@ int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg) {
  * @param msg
  * @return int32_t
  */
-int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg) {
+static int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg) {
   const SArray *pDataBlocks = (const SArray *)msg;
   // TODO: destroy SSDataBlocks(msg)
   if (!pDataBlocks) {

source/dnode/vnode/src/sma/smaUtil.c

@@ -305,93 +305,4 @@ int32_t tdReleaseSmaRef(int32_t rsetId, int64_t refId) {
   smaDebug("rsma release ref for rsetId:%" PRIi64 " refId:%d success", rsetId, refId);
 
   return TSDB_CODE_SUCCESS;
-}
-
-static int32_t tdCloneQTaskInfo(SSma *pSma, qTaskInfo_t dstTaskInfo, qTaskInfo_t srcTaskInfo, SRSmaParam *param,
-                                tb_uid_t suid, int8_t idx) {
-  SVnode *pVnode = pSma->pVnode;
-  char   *pOutput = NULL;
-  int32_t len = 0;
-
-  if ((terrno = qSerializeTaskStatus(srcTaskInfo, &pOutput, &len)) < 0) {
-    smaError("vgId:%d, rsma clone, table %" PRIi64 " serialize qTaskInfo failed since %s", TD_VID(pVnode), suid,
-             terrstr());
-    goto _err;
-  }
-
-  SReadHandle handle = {
-      .meta = pVnode->pMeta,
-      .vnode = pVnode,
-      .initTqReader = 1,
-  };
-  ASSERT(!dstTaskInfo);
-  dstTaskInfo = qCreateStreamExecTaskInfo(param->qmsg[idx], &handle);
-  if (!dstTaskInfo) {
-    terrno = TSDB_CODE_RSMA_QTASKINFO_CREATE;
-    goto _err;
-  }
-
-  if (qDeserializeTaskStatus(dstTaskInfo, pOutput, len) < 0) {
-    smaError("vgId:%d, rsma clone, restore rsma task for table:%" PRIi64 " failed since %s", TD_VID(pVnode), suid,
-             terrstr());
-    goto _err;
-  }
-
-  smaDebug("vgId:%d, rsma clone, restore rsma task for table:%" PRIi64 " succeed", TD_VID(pVnode), suid);
-
-  taosMemoryFreeClear(pOutput);
-  return TSDB_CODE_SUCCESS;
-_err:
-  taosMemoryFreeClear(pOutput);
-  tdFreeQTaskInfo(dstTaskInfo, TD_VID(pVnode), idx + 1);
-  smaError("vgId:%d, rsma clone, restore rsma task for table:%" PRIi64 " failed since %s", TD_VID(pVnode), suid,
-           terrstr());
-  return TSDB_CODE_FAILED;
-}
-
-/**
- * @brief Clone qTaskInfo of SRSmaInfo
- *
- * @param pSma
- * @param pInfo
- * @return int32_t
- */
-int32_t tdCloneRSmaInfo(SSma *pSma, SRSmaInfo *pInfo) {
-  SRSmaParam *param = NULL;
-  if (!pInfo) {
-    return TSDB_CODE_SUCCESS;
-  }
-
-  SMetaReader mr = {0};
-  metaReaderInit(&mr, SMA_META(pSma), 0);
-  smaDebug("vgId:%d, rsma clone qTaskInfo for suid:%" PRIi64, SMA_VID(pSma), pInfo->suid);
-  if (metaGetTableEntryByUid(&mr, pInfo->suid) < 0) {
-    smaError("vgId:%d, rsma clone, failed to get table meta for %" PRIi64 " since %s", SMA_VID(pSma), pInfo->suid,
-             terrstr());
-    goto _err;
-  }
-  ASSERT(mr.me.type == TSDB_SUPER_TABLE);
-  ASSERT(mr.me.uid == pInfo->suid);
-  if (TABLE_IS_ROLLUP(mr.me.flags)) {
-    param = &mr.me.stbEntry.rsmaParam;
-    for (int32_t i = 0; i < TSDB_RETENTION_L2; ++i) {
-      if (!pInfo->iTaskInfo[i]) {
-        continue;
-      }
-      if (tdCloneQTaskInfo(pSma, pInfo->taskInfo[i], pInfo->iTaskInfo[i], param, pInfo->suid, i) < 0) {
-        goto _err;
-      }
-    }
-    smaDebug("vgId:%d, rsma clone env success for %" PRIi64, SMA_VID(pSma), pInfo->suid);
-  } else {
-    terrno = TSDB_CODE_RSMA_INVALID_SCHEMA;
-    goto _err;
-  }
-
-  metaReaderClear(&mr);
-  return TSDB_CODE_SUCCESS;
-_err:
-  metaReaderClear(&mr);
-  smaError("vgId:%d, rsma clone env failed for %" PRIi64 " since %s", SMA_VID(pSma), pInfo->suid, terrstr());
-  return TSDB_CODE_FAILED;
 }
\ No newline at end of file

source/dnode/vnode/src/tq/tq.c

@@ -816,7 +816,86 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, int64_t version, char* msg, int32_t msg
   return streamMetaAddSerializedTask(pTq->pStreamMeta, version, msg, msgLen);
 }
 
-int32_t tqProcessStreamTrigger(STQ* pTq, SSubmitReq* pReq, int64_t ver) {
+int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver) {
+  bool        failed = false;
+  SDecoder*   pCoder = &(SDecoder){0};
+  SDeleteRes* pRes = &(SDeleteRes){0};
+
+  pRes->uidList = taosArrayInit(0, sizeof(tb_uid_t));
+  if (pRes->uidList == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    failed = true;
+  }
+
+  tDecoderInit(pCoder, pReq, len);
+  tDecodeDeleteRes(pCoder, pRes);
+  tDecoderClear(pCoder);
+
+  int32_t sz = taosArrayGetSize(pRes->uidList);
+  if (sz == 0) {
+    taosArrayDestroy(pRes->uidList);
+    return 0;
+  }
+  SSDataBlock* pDelBlock = createSpecialDataBlock(STREAM_DELETE_DATA);
+  blockDataEnsureCapacity(pDelBlock, sz);
+  pDelBlock->info.rows = sz;
+  pDelBlock->info.version = ver;
+
+  for (int32_t i = 0; i < sz; i++) {
+    // start key column
+    SColumnInfoData* pStartCol = taosArrayGet(pDelBlock->pDataBlock, START_TS_COLUMN_INDEX);
+    colDataAppend(pStartCol, i, (const char*)&pRes->skey, false);  // end key column
+    SColumnInfoData* pEndCol = taosArrayGet(pDelBlock->pDataBlock, END_TS_COLUMN_INDEX);
+    colDataAppend(pEndCol, i, (const char*)&pRes->ekey, false);
+    // uid column
+    SColumnInfoData* pUidCol = taosArrayGet(pDelBlock->pDataBlock, UID_COLUMN_INDEX);
+    int64_t*         pUid = taosArrayGet(pRes->uidList, i);
+    colDataAppend(pUidCol, i, (const char*)pUid, false);
+
+    colDataAppendNULL(taosArrayGet(pDelBlock->pDataBlock, GROUPID_COLUMN_INDEX), i);
+    colDataAppendNULL(taosArrayGet(pDelBlock->pDataBlock, CALCULATE_START_TS_COLUMN_INDEX), i);
+    colDataAppendNULL(taosArrayGet(pDelBlock->pDataBlock, CALCULATE_END_TS_COLUMN_INDEX), i);
+  }
+
+  taosArrayDestroy(pRes->uidList);
+
+  void* pIter = NULL;
+  while (1) {
+    pIter = taosHashIterate(pTq->pStreamMeta->pTasks, pIter);
+    if (pIter == NULL) break;
+    SStreamTask* pTask = *(SStreamTask**)pIter;
+    if (pTask->taskLevel != TASK_LEVEL__SOURCE) continue;
+
+    qDebug("delete req enqueue stream task: %d, ver: %" PRId64, pTask->taskId, ver);
+
+    SStreamDataBlock* pStreamBlock = taosAllocateQitem(sizeof(SStreamDataBlock), DEF_QITEM);
+    pStreamBlock->type = STREAM_INPUT__DATA_BLOCK;
+    pStreamBlock->blocks = taosArrayInit(0, sizeof(SSDataBlock));
+    SSDataBlock block = {0};
+    assignOneDataBlock(&block, pDelBlock);
+    block.info.type = STREAM_DELETE_DATA;
+    taosArrayPush(pStreamBlock->blocks, &block);
+
+    if (!failed) {
+      if (streamTaskInput(pTask, (SStreamQueueItem*)pStreamBlock) < 0) {
+        qError("stream task input del failed, task id %d", pTask->taskId);
+        continue;
+      }
+
+      if (streamSchedExec(pTask) < 0) {
+        qError("stream task launch failed, task id %d", pTask->taskId);
+        continue;
+      }
+    } else {
+      streamTaskInputFail(pTask);
+    }
+  }
+  blockDataDestroy(pDelBlock);
+
+  return 0;
+}
+
+int32_t tqProcessSubmitReq(STQ* pTq, SSubmitReq* pReq, int64_t ver) {
   void*              pIter = NULL;
   bool               failed = false;
   SStreamDataSubmit* pSubmit = NULL;

source/dnode/vnode/src/tq/tqPush.c

@@ -213,20 +213,25 @@ int32_t tqPushMsgNew(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_
 #endif
 
 int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver) {
-  if (vnodeIsRoleLeader(pTq->pVnode) && msgType == TDMT_VND_SUBMIT) {
-    if (taosHashGetSize(pTq->pStreamMeta->pTasks) == 0) return 0;
-
-    void* data = taosMemoryMalloc(msgLen);
-    if (data == NULL) {
-      terrno = TSDB_CODE_OUT_OF_MEMORY;
-      tqError("failed to copy data for stream since out of memory");
-      return -1;
+  if (vnodeIsRoleLeader(pTq->pVnode)) {
+    if (msgType == TDMT_VND_SUBMIT) {
+      if (taosHashGetSize(pTq->pStreamMeta->pTasks) == 0) return 0;
+
+      void* data = taosMemoryMalloc(msgLen);
+      if (data == NULL) {
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
+        tqError("failed to copy data for stream since out of memory");
+        return -1;
+      }
+      memcpy(data, msg, msgLen);
+      SSubmitReq* pReq = (SSubmitReq*)data;
+      pReq->version = ver;
+
+      tqProcessSubmitReq(pTq, data, ver);
+    }
+    if (msgType == TDMT_VND_DELETE) {
+      tqProcessDelReq(pTq, POINTER_SHIFT(msg, sizeof(SMsgHead)), msgLen - sizeof(SMsgHead), ver);
     }
-    memcpy(data, msg, msgLen);
-    SSubmitReq* pReq = (SSubmitReq*)data;
-    pReq->version = ver;
-
-    tqProcessStreamTrigger(pTq, data, ver);
   }
 
   return 0;

source/dnode/vnode/src/tsdb/tsdbCache.c

@@ -418,31 +418,16 @@ typedef enum {
 } SFSLASTNEXTROWSTATES;
 
 typedef struct {
-  SFSLASTNEXTROWSTATES state;         // [input]
-  STsdb               *pTsdb;         // [input]
-  SBlockIdx           *pBlockIdxExp;  // [input]
-  STSchema            *pTSchema;      // [input]
-  tb_uid_t             suid;
+  SFSLASTNEXTROWSTATES state;  // [input]
+  STsdb               *pTsdb;  // [input]
   tb_uid_t             uid;
   int32_t              nFileSet;
   int32_t              iFileSet;
   SArray              *aDFileSet;
   SDataFReader        *pDataFReader;
-  SArray              *aBlockL;
-  SBlockL             *pBlockL;
-  SBlockData          *pBlockDataL;
-  SBlockData           blockDataL;
-  int32_t              nRow;
-  int32_t              iRow;
   TSDBROW              row;
-  /*
-  SArray    *aBlockIdx;
-  SBlockIdx *pBlockIdx;
-  SMapData   blockMap;
-  int32_t    nBlock;
-  int32_t    iBlock;
-  SBlock     block;
-  */
+
+  SMergeTree mergeTree;
 } SFSLastNextRowIter;
 
 static int32_t getNextRowFromFSLast(void *iter, TSDBROW **ppRow) {
@@ -451,22 +436,16 @@ static int32_t getNextRowFromFSLast(void *iter, TSDBROW **ppRow) {
 
   switch (state->state) {
     case SFSLASTNEXTROW_FS:
-      // state->aDFileSet = state->pTsdb->pFS->cState->aDFileSet;
       state->nFileSet = taosArrayGetSize(state->aDFileSet);
       state->iFileSet = state->nFileSet;
 
-      state->pBlockDataL = NULL;
-
     case SFSLASTNEXTROW_FILESET: {
       SDFileSet *pFileSet = NULL;
     _next_fileset:
       if (--state->iFileSet >= 0) {
         pFileSet = (SDFileSet *)taosArrayGet(state->aDFileSet, state->iFileSet);
       } else {
-        if (state->pBlockDataL) {
-          tBlockDataDestroy(state->pBlockDataL, 1);
-          state->pBlockDataL = NULL;
-        }
+        // tMergeTreeClose(&state->mergeTree);
 
         *ppRow = NULL;
         return code;
@@ -475,68 +454,24 @@ static int32_t getNextRowFromFSLast(void *iter, TSDBROW **ppRow) {
       code = tsdbDataFReaderOpen(&state->pDataFReader, state->pTsdb, pFileSet);
       if (code) goto _err;
 
-      if (!state->aBlockL) {
-        state->aBlockL = taosArrayInit(0, sizeof(SBlockL));
-      } else {
-        taosArrayClear(state->aBlockL);
-      }
-
-      code = tsdbReadBlockL(state->pDataFReader, state->aBlockL);
-      if (code) goto _err;
-
-      // SBlockL *pBlockL = (SBlockL *)taosArrayGet(state->aBlockL, state->iBlockL);
-
-      state->pBlockL = taosArraySearch(state->aBlockL, state->pBlockIdxExp, tCmprBlockL, TD_EQ);
-      if (!state->pBlockL) {
+      tMergeTreeOpen(&state->mergeTree, 1, state->pDataFReader, state->uid,
+                     &(STimeWindow){.skey = TSKEY_MIN, .ekey = TSKEY_MAX},
+                     &(SVersionRange){.minVer = 0, .maxVer = UINT64_MAX});
+      bool hasVal = tMergeTreeNext(&state->mergeTree);
+      if (!hasVal) {
+        state->state = SFSLASTNEXTROW_FILESET;
+        // tMergeTreeClose(&state->mergeTree);
         goto _next_fileset;
       }
-
-      int64_t suid = state->pBlockL->suid;
-      int64_t uid = state->pBlockL->maxUid;
-
-      if (!state->pBlockDataL) {
-        state->pBlockDataL = &state->blockDataL;
-
-        tBlockDataCreate(state->pBlockDataL);
-      }
-      code = tBlockDataInit(state->pBlockDataL, suid, suid ? 0 : uid, state->pTSchema);
-      if (code) goto _err;
-    }
-    case SFSLASTNEXTROW_BLOCKDATA:
-      code = tsdbReadLastBlock(state->pDataFReader, state->pBlockL, state->pBlockDataL);
-      if (code) goto _err;
-
-      state->nRow = state->blockDataL.nRow;
-      state->iRow = state->nRow - 1;
-
-      if (!state->pBlockDataL->uid) {
-        while (state->pBlockIdxExp->uid != state->pBlockDataL->aUid[state->iRow]) {
-          --state->iRow;
-        }
-      }
-
       state->state = SFSLASTNEXTROW_BLOCKROW;
+    }
     case SFSLASTNEXTROW_BLOCKROW:
-      if (state->pBlockDataL->uid) {
-        if (state->iRow >= 0) {
-          state->row = tsdbRowFromBlockData(state->pBlockDataL, state->iRow);
-          *ppRow = &state->row;
-
-          if (--state->iRow < 0) {
-            state->state = SFSLASTNEXTROW_FILESET;
-          }
-        }
-      } else {
-        if (state->iRow >= 0 && state->pBlockIdxExp->uid == state->pBlockDataL->aUid[state->iRow]) {
-          state->row = tsdbRowFromBlockData(state->pBlockDataL, state->iRow);
-          *ppRow = &state->row;
-
-          if (--state->iRow < 0 || state->pBlockIdxExp->uid != state->pBlockDataL->aUid[state->iRow]) {
-            state->state = SFSLASTNEXTROW_FILESET;
-          }
-        }
+      state->row = tMergeTreeGetRow(&state->mergeTree);
+      *ppRow = &state->row;
+      bool hasVal = tMergeTreeNext(&state->mergeTree);
+      if (!hasVal) {
+        state->state = SFSLASTNEXTROW_FILESET;
       }
-
       return code;
     default:
       ASSERT(0);
@@ -548,15 +483,6 @@ _err:
     tsdbDataFReaderClose(&state->pDataFReader);
     state->pDataFReader = NULL;
   }
-  if (state->aBlockL) {
-    taosArrayDestroy(state->aBlockL);
-    state->aBlockL = NULL;
-  }
-  if (state->pBlockDataL) {
-    tBlockDataDestroy(state->pBlockDataL, 1);
-    state->pBlockDataL = NULL;
-  }
-
   *ppRow = NULL;
 
   return code;
@@ -574,14 +500,6 @@ int32_t clearNextRowFromFSLast(void *iter) {
     tsdbDataFReaderClose(&state->pDataFReader);
     state->pDataFReader = NULL;
   }
-  if (state->aBlockL) {
-    taosArrayDestroy(state->aBlockL);
-    state->aBlockL = NULL;
-  }
-  if (state->pBlockDataL) {
-    tBlockDataDestroy(state->pBlockDataL, 1);
-    state->pBlockDataL = NULL;
-  }
 
   return code;
 }
@@ -609,7 +527,7 @@ typedef struct SFSNextRowIter {
   SMapData         blockMap;
   int32_t          nBlock;
   int32_t          iBlock;
-  SBlock           block;
+  SDataBlk         block;
   SBlockData       blockData;
   SBlockData      *pBlockData;
   int32_t          nRow;
@@ -684,13 +602,13 @@ static int32_t getNextRowFromFS(void *iter, TSDBROW **ppRow) {
     }
     case SFSNEXTROW_BLOCKDATA:
       if (state->iBlock >= 0) {
-        SBlock block = {0};
+        SDataBlk block = {0};
 
-        tBlockReset(&block);
+        tDataBlkReset(&block);
         // tBlockDataReset(&state->blockData);
         tBlockDataReset(state->pBlockData);
 
-        tMapDataGetItemByIdx(&state->blockMap, state->iBlock, &block, tGetBlock);
+        tMapDataGetItemByIdx(&state->blockMap, state->iBlock, &block, tGetDataBlk);
         /* code = tsdbReadBlockData(state->pDataFReader, &state->blockIdx, &block, &state->blockData, NULL, NULL); */
         tBlockDataReset(state->pBlockData);
         code = tBlockDataInit(state->pBlockData, state->suid, state->uid, state->pTSchema);
@@ -972,9 +890,6 @@ static int32_t nextRowIterOpen(CacheNextRowIter *pIter, tb_uid_t uid, STsdb *pTs
   pIter->fsLastState.state = (SFSLASTNEXTROWSTATES)SFSNEXTROW_FS;
   pIter->fsLastState.pTsdb = pTsdb;
   pIter->fsLastState.aDFileSet = pIter->pReadSnap->fs.aDFileSet;
-  pIter->fsLastState.pBlockIdxExp = &pIter->idx;
-  pIter->fsLastState.pTSchema = pTSchema;
-  pIter->fsLastState.suid = suid;
   pIter->fsLastState.uid = uid;
 
   pIter->fsState.state = SFSNEXTROW_FS;
@@ -1372,25 +1287,33 @@ int32_t tsdbCacheGetLastH(SLRUCache *pCache, tb_uid_t uid, STsdb *pTsdb, LRUHand
   // getTableCacheKeyS(uid, "l", key, &keyLen);
   getTableCacheKey(uid, 1, key, &keyLen);
   LRUHandle *h = taosLRUCacheLookup(pCache, key, keyLen);
-  if (h) {
-  } else {
-    SArray *pLastArray = NULL;
-    code = mergeLast(uid, pTsdb, &pLastArray);
-    // if table's empty or error, return code of -1
-    // if (code < 0 || pRow == NULL) {
-    if (code < 0 || pLastArray == NULL) {
-      *handle = NULL;
-      return 0;
-    }
-
-    _taos_lru_deleter_t deleter = deleteTableCacheLast;
-    LRUStatus           status =
-        taosLRUCacheInsert(pCache, key, keyLen, pLastArray, pLastArray->capacity, deleter, NULL, TAOS_LRU_PRIORITY_LOW);
-    if (status != TAOS_LRU_STATUS_OK) {
-      code = -1;
-    }
+  if (!h) {
+    taosThreadMutexLock(&pTsdb->lruMutex);
 
     h = taosLRUCacheLookup(pCache, key, keyLen);
+    if (!h) {
+      SArray *pLastArray = NULL;
+      code = mergeLast(uid, pTsdb, &pLastArray);
+      // if table's empty or error, return code of -1
+      // if (code < 0 || pRow == NULL) {
+      if (code < 0 || pLastArray == NULL) {
+        *handle = NULL;
+        return 0;
+      }
+
+      _taos_lru_deleter_t deleter = deleteTableCacheLast;
+      LRUStatus status = taosLRUCacheInsert(pCache, key, keyLen, pLastArray, pLastArray->capacity, deleter, NULL,
+                                            TAOS_LRU_PRIORITY_LOW);
+      if (status != TAOS_LRU_STATUS_OK) {
+        code = -1;
+      }
+
+      taosThreadMutexUnlock(&pTsdb->lruMutex);
+
+      h = taosLRUCacheLookup(pCache, key, keyLen);
+    } else {
+      taosThreadMutexUnlock(&pTsdb->lruMutex);
+    }
   }
 
   *handle = h;
@@ -1411,3 +1334,5 @@ void tsdbCacheSetCapacity(SVnode *pVnode, size_t capacity) {
 }
 
 size_t tsdbCacheGetCapacity(SVnode *pVnode) { return taosLRUCacheGetCapacity(pVnode->pTsdb->lruCache); }
+
+size_t tsdbCacheGetUsage(SVnode *pVnode) { return taosLRUCacheGetUsage(pVnode->pTsdb->lruCache); }

source/dnode/vnode/src/tsdb/tsdbCompact.c

+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "tsdb.h"
+
+typedef struct {
+  STsdb  *pTsdb;
+  STsdbFS fs;
+} STsdbCompactor;
+
+int32_t tsdbCompact(STsdb *pTsdb) {
+  int32_t code = 0;
+  // TODO
+  return code;
+}

source/dnode/vnode/src/tsdb/tsdbCompress.c

+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "tsdb.h"
+
+// Integer =====================================================
+typedef struct {
+  int8_t   rawCopy;
+  int64_t  prevVal;
+  int32_t  nVal;
+  int32_t  nBuf;
+  uint8_t *pBuf;
+} SIntCompressor;
+
+#define I64_SAFE_ADD(a, b) (((a) >= 0 && (b) <= INT64_MAX - (b)) || ((a) < 0 && (b) >= INT64_MIN - (a)))
+#define SIMPLE8B_MAX       ((uint64_t)1152921504606846974LL)
+
+static int32_t tsdbCmprI64(SIntCompressor *pCompressor, int64_t val) {
+  int32_t code = 0;
+
+  // raw copy
+  if (pCompressor->rawCopy) {
+    memcpy(pCompressor->pBuf + pCompressor->nBuf, &val, sizeof(val));
+    pCompressor->nBuf += sizeof(val);
+    pCompressor->nVal++;
+    goto _exit;
+  }
+
+  if (!I64_SAFE_ADD(val, pCompressor->prevVal)) {
+    pCompressor->rawCopy = 1;
+    // TODO: decompress and copy
+    pCompressor->nVal++;
+    goto _exit;
+  }
+
+  int64_t diff = val - pCompressor->prevVal;
+  uint8_t zigzag = ZIGZAGE(int64_t, diff);
+
+  if (zigzag >= SIMPLE8B_MAX) {
+    pCompressor->rawCopy = 1;
+    // TODO: decompress and copy
+    pCompressor->nVal++;
+    goto _exit;
+  }
+
+_exit:
+  return code;
+}
+
+// Timestamp =====================================================
+
+// Float =====================================================
\ No newline at end of file

source/dnode/vnode/src/tsdb/tsdbDiskData.c

+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "tsdb.h"
+
+typedef struct SDiskColBuilder SDiskColBuilder;
+struct SDiskColBuilder {
+  uint8_t  flags;
+  uint8_t *pBitMap;
+  int32_t *aOffset;
+  int32_t  nData;
+  uint8_t *pData;
+};
+
+int32_t tDiskColAddVal(SDiskColBuilder *pBuilder, SColVal *pColVal) {
+  int32_t code = 0;
+  // TODO
+  return code;
+}
+
+// ================================================================
+typedef struct SDiskDataBuilder SDiskDataBuilder;
+struct SDiskDataBuilder {
+  SDiskDataHdr hdr;
+  SArray      *aBlockCol;  // SArray<SBlockCol>
+};
+
+int32_t tDiskDataBuilderCreate(SDiskDataBuilder **ppBuilder) {
+  int32_t code = 0;
+  // TODO
+  return code;
+}
+
+void tDiskDataBuilderDestroy(SDiskDataBuilder *pBuilder) {
+  // TODO
+}
+
+void tDiskDataBuilderInit(SDiskDataBuilder *pBuilder, int64_t suid, int64_t uid, STSchema *pTSchema, int8_t cmprAlg) {
+  pBuilder->hdr = (SDiskDataHdr){.delimiter = TSDB_FILE_DLMT,  //
+                                 .fmtVer = 0,
+                                 .suid = suid,
+                                 .uid = uid,
+                                 .cmprAlg = cmprAlg};
+}
+
+void tDiskDataBuilderReset(SDiskDataBuilder *pBuilder) {
+  // TODO
+}
+
+int32_t tDiskDataBuilderAddRow(SDiskDataBuilder *pBuilder, TSDBROW *pRow, STSchema *pTSchema, int64_t uid) {
+  int32_t code = 0;
+
+  // uid (todo)
+
+  // version (todo)
+
+  // TSKEY (todo)
+
+  SRowIter iter = {0};
+  tRowIterInit(&iter, pRow, pTSchema);
+
+  for (int32_t iDiskCol = 0; iDiskCol < 0; iDiskCol++) {
+  }
+
+  return code;
+}
+
+int32_t tDiskDataBuilderGet(SDiskDataBuilder *pBuilder, uint8_t **ppData) {
+  int32_t code = 0;
+  // TODO
+  return code;
+}
\ No newline at end of file

source/dnode/vnode/src/tsdb/tsdbFS.c

@@ -110,7 +110,7 @@ _err:
 //       taosRemoveFile(fname);
 //     }
 
-//     // last
+//     // sst
 //     if (isSameDisk && pFrom->pLastF->commitID == pTo->pLastF->commitID) {
 //       if (pFrom->pLastF->size > pTo->pLastF->size) {
 //         code = tsdbDFileRollback(pFS->pTsdb, pTo, TSDB_LAST_FILE);
@@ -140,7 +140,7 @@ _err:
 //     tsdbDataFileName(pFS->pTsdb, pFrom->diskId, pFrom->fid, pFrom->pDataF, fname);
 //     taosRemoveFile(fname);
 
-//     // last
+//     // sst
 //     tsdbLastFileName(pFS->pTsdb, pFrom->diskId, pFrom->fid, pFrom->pLastF, fname);
 //     taosRemoveFile(fname);
 
@@ -254,8 +254,10 @@ void tsdbFSDestroy(STsdbFS *pFS) {
     SDFileSet *pSet = (SDFileSet *)taosArrayGet(pFS->aDFileSet, iSet);
     taosMemoryFree(pSet->pHeadF);
     taosMemoryFree(pSet->pDataF);
-    taosMemoryFree(pSet->pLastF);
     taosMemoryFree(pSet->pSmaF);
+    for (int32_t iSst = 0; iSst < pSet->nSstF; iSst++) {
+      taosMemoryFree(pSet->aSstF[iSst]);
+    }
   }
 
   taosArrayDestroy(pFS->aDFileSet);
@@ -309,17 +311,6 @@ static int32_t tsdbScanAndTryFixFS(STsdb *pTsdb) {
       if (code) goto _err;
     }
 
-    // last ===========
-    tsdbLastFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pLastF, fname);
-    if (taosStatFile(fname, &size, NULL)) {
-      code = TAOS_SYSTEM_ERROR(errno);
-      goto _err;
-    }
-    if (size != pSet->pLastF->size) {
-      code = TSDB_CODE_FILE_CORRUPTED;
-      goto _err;
-    }
-
     // sma =============
     tsdbSmaFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pSmaF, fname);
     if (taosStatFile(fname, &size, NULL)) {
@@ -333,6 +324,19 @@ static int32_t tsdbScanAndTryFixFS(STsdb *pTsdb) {
       code = tsdbDFileRollback(pTsdb, pSet, TSDB_SMA_FILE);
       if (code) goto _err;
     }
+
+    // sst ===========
+    for (int32_t iSst = 0; iSst < pSet->nSstF; iSst++) {
+      tsdbSstFileName(pTsdb, pSet->diskId, pSet->fid, pSet->aSstF[iSst], fname);
+      if (taosStatFile(fname, &size, NULL)) {
+        code = TAOS_SYSTEM_ERROR(errno);
+        goto _err;
+      }
+      if (size != pSet->aSstF[iSst]->size) {
+        code = TSDB_CODE_FILE_CORRUPTED;
+        goto _err;
+      }
+    }
   }
 
   {
@@ -382,41 +386,15 @@ static int32_t tsdbRecoverFS(STsdb *pTsdb, uint8_t *pData, int64_t nData) {
   taosArrayClear(pTsdb->fs.aDFileSet);
   n += tGetU32v(pData + n, &nSet);
   for (uint32_t iSet = 0; iSet < nSet; iSet++) {
-    SDFileSet fSet;
-
-    // head
-    fSet.pHeadF = (SHeadFile *)taosMemoryCalloc(1, sizeof(SHeadFile));
-    if (fSet.pHeadF == NULL) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _err;
-    }
-    fSet.pHeadF->nRef = 1;
-
-    // data
-    fSet.pDataF = (SDataFile *)taosMemoryCalloc(1, sizeof(SDataFile));
-    if (fSet.pDataF == NULL) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _err;
-    }
-    fSet.pDataF->nRef = 1;
+    SDFileSet fSet = {0};
 
-    // last
-    fSet.pLastF = (SLastFile *)taosMemoryCalloc(1, sizeof(SLastFile));
-    if (fSet.pLastF == NULL) {
+    int32_t nt = tGetDFileSet(pData + n, &fSet);
+    if (nt < 0) {
       code = TSDB_CODE_OUT_OF_MEMORY;
       goto _err;
     }
-    fSet.pLastF->nRef = 1;
 
-    // sma
-    fSet.pSmaF = (SSmaFile *)taosMemoryCalloc(1, sizeof(SSmaFile));
-    if (fSet.pSmaF == NULL) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _err;
-    }
-    fSet.pSmaF->nRef = 1;
-
-    n += tGetDFileSet(pData + n, &fSet);
+    n += nt;
 
     if (taosArrayPush(pTsdb->fs.aDFileSet, &fSet) == NULL) {
       code = TSDB_CODE_OUT_OF_MEMORY;
@@ -532,13 +510,15 @@ int32_t tsdbFSClose(STsdb *pTsdb) {
     ASSERT(pSet->pDataF->nRef == 1);
     taosMemoryFree(pSet->pDataF);
 
-    // last
-    ASSERT(pSet->pLastF->nRef == 1);
-    taosMemoryFree(pSet->pLastF);
-
     // sma
     ASSERT(pSet->pSmaF->nRef == 1);
     taosMemoryFree(pSet->pSmaF);
+
+    // sst
+    for (int32_t iSst = 0; iSst < pSet->nSstF; iSst++) {
+      ASSERT(pSet->aSstF[iSst]->nRef == 1);
+      taosMemoryFree(pSet->aSstF[iSst]);
+    }
   }
 
   taosArrayDestroy(pTsdb->fs.aDFileSet);
@@ -586,15 +566,7 @@ int32_t tsdbFSCopy(STsdb *pTsdb, STsdbFS *pFS) {
     }
     *fSet.pDataF = *pSet->pDataF;
 
-    // data
-    fSet.pLastF = (SLastFile *)taosMemoryMalloc(sizeof(SLastFile));
-    if (fSet.pLastF == NULL) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _exit;
-    }
-    *fSet.pLastF = *pSet->pLastF;
-
-    // last
+    // sma
     fSet.pSmaF = (SSmaFile *)taosMemoryMalloc(sizeof(SSmaFile));
     if (fSet.pSmaF == NULL) {
       code = TSDB_CODE_OUT_OF_MEMORY;
@@ -602,6 +574,16 @@ int32_t tsdbFSCopy(STsdb *pTsdb, STsdbFS *pFS) {
     }
     *fSet.pSmaF = *pSet->pSmaF;
 
+    // sst
+    for (fSet.nSstF = 0; fSet.nSstF < pSet->nSstF; fSet.nSstF++) {
+      fSet.aSstF[fSet.nSstF] = (SSstFile *)taosMemoryMalloc(sizeof(SSstFile));
+      if (fSet.aSstF[fSet.nSstF] == NULL) {
+        code = TSDB_CODE_OUT_OF_MEMORY;
+        goto _exit;
+      }
+      *fSet.aSstF[fSet.nSstF] = *pSet->aSstF[fSet.nSstF];
+    }
+
     if (taosArrayPush(pFS->aDFileSet, &fSet) == NULL) {
       code = TSDB_CODE_OUT_OF_MEMORY;
       goto _exit;
@@ -651,14 +633,38 @@ int32_t tsdbFSUpsertFSet(STsdbFS *pFS, SDFileSet *pSet) {
     if (c == 0) {
       *pDFileSet->pHeadF = *pSet->pHeadF;
       *pDFileSet->pDataF = *pSet->pDataF;
-      *pDFileSet->pLastF = *pSet->pLastF;
       *pDFileSet->pSmaF = *pSet->pSmaF;
+      // sst
+      if (pSet->nSstF > pDFileSet->nSstF) {
+        ASSERT(pSet->nSstF == pDFileSet->nSstF + 1);
+
+        pDFileSet->aSstF[pDFileSet->nSstF] = (SSstFile *)taosMemoryMalloc(sizeof(SSstFile));
+        if (pDFileSet->aSstF[pDFileSet->nSstF] == NULL) {
+          code = TSDB_CODE_OUT_OF_MEMORY;
+          goto _exit;
+        }
+        *pDFileSet->aSstF[pDFileSet->nSstF] = *pSet->aSstF[pSet->nSstF - 1];
+        pDFileSet->nSstF++;
+      } else if (pSet->nSstF < pDFileSet->nSstF) {
+        ASSERT(pSet->nSstF == 1);
+        for (int32_t iSst = 1; iSst < pDFileSet->nSstF; iSst++) {
+          taosMemoryFree(pDFileSet->aSstF[iSst]);
+        }
+
+        *pDFileSet->aSstF[0] = *pSet->aSstF[0];
+        pDFileSet->nSstF = 1;
+      } else {
+        for (int32_t iSst = 0; iSst < pSet->nSstF; iSst++) {
+          *pDFileSet->aSstF[iSst] = *pSet->aSstF[iSst];
+        }
+      }
 
       goto _exit;
     }
   }
 
-  SDFileSet fSet = {.diskId = pSet->diskId, .fid = pSet->fid};
+  ASSERT(pSet->nSstF == 1);
+  SDFileSet fSet = {.diskId = pSet->diskId, .fid = pSet->fid, .nSstF = 1};
 
   // head
   fSet.pHeadF = (SHeadFile *)taosMemoryMalloc(sizeof(SHeadFile));
@@ -676,21 +682,21 @@ int32_t tsdbFSUpsertFSet(STsdbFS *pFS, SDFileSet *pSet) {
   }
   *fSet.pDataF = *pSet->pDataF;
 
-  // data
-  fSet.pLastF = (SLastFile *)taosMemoryMalloc(sizeof(SLastFile));
-  if (fSet.pLastF == NULL) {
+  // sma
+  fSet.pSmaF = (SSmaFile *)taosMemoryMalloc(sizeof(SSmaFile));
+  if (fSet.pSmaF == NULL) {
     code = TSDB_CODE_OUT_OF_MEMORY;
     goto _exit;
   }
-  *fSet.pLastF = *pSet->pLastF;
+  *fSet.pSmaF = *pSet->pSmaF;
 
-  // last
-  fSet.pSmaF = (SSmaFile *)taosMemoryMalloc(sizeof(SSmaFile));
-  if (fSet.pSmaF == NULL) {
+  // sst
+  fSet.aSstF[0] = (SSstFile *)taosMemoryMalloc(sizeof(SSstFile));
+  if (fSet.aSstF[0] == NULL) {
     code = TSDB_CODE_OUT_OF_MEMORY;
     goto _exit;
   }
-  *fSet.pSmaF = *pSet->pSmaF;
+  *fSet.aSstF[0] = *pSet->aSstF[0];
 
   if (taosArrayInsert(pFS->aDFileSet, idx, &fSet) == NULL) {
     code = TSDB_CODE_OUT_OF_MEMORY;
@@ -836,27 +842,6 @@ int32_t tsdbFSCommit2(STsdb *pTsdb, STsdbFS *pFSNew) {
       pSetOld->pDataF->size = pSetNew->pDataF->size;
     }
 
-    // last
-    fSet.pLastF = pSetOld->pLastF;
-    if ((!sameDisk) || (pSetOld->pLastF->commitID != pSetNew->pLastF->commitID)) {
-      pSetOld->pLastF = (SLastFile *)taosMemoryMalloc(sizeof(SLastFile));
-      if (pSetOld->pLastF == NULL) {
-        code = TSDB_CODE_OUT_OF_MEMORY;
-        goto _err;
-      }
-      *pSetOld->pLastF = *pSetNew->pLastF;
-      pSetOld->pLastF->nRef = 1;
-
-      nRef = atomic_sub_fetch_32(&fSet.pLastF->nRef, 1);
-      if (nRef == 0) {
-        tsdbLastFileName(pTsdb, pSetOld->diskId, pSetOld->fid, fSet.pLastF, fname);
-        taosRemoveFile(fname);
-        taosMemoryFree(fSet.pLastF);
-      }
-    } else {
-      ASSERT(pSetOld->pLastF->size == pSetNew->pLastF->size);
-    }
-
     // sma
     fSet.pSmaF = pSetOld->pSmaF;
     if ((!sameDisk) || (pSetOld->pSmaF->commitID != pSetNew->pSmaF->commitID)) {
@@ -879,6 +864,84 @@ int32_t tsdbFSCommit2(STsdb *pTsdb, STsdbFS *pFSNew) {
       pSetOld->pSmaF->size = pSetNew->pSmaF->size;
     }
 
+    // sst
+    if (sameDisk) {
+      if (pSetNew->nSstF > pSetOld->nSstF) {
+        ASSERT(pSetNew->nSstF = pSetOld->nSstF + 1);
+        pSetOld->aSstF[pSetOld->nSstF] = (SSstFile *)taosMemoryMalloc(sizeof(SSstFile));
+        if (pSetOld->aSstF[pSetOld->nSstF] == NULL) {
+          code = TSDB_CODE_OUT_OF_MEMORY;
+          goto _err;
+        }
+        *pSetOld->aSstF[pSetOld->nSstF] = *pSetNew->aSstF[pSetOld->nSstF];
+        pSetOld->aSstF[pSetOld->nSstF]->nRef = 1;
+        pSetOld->nSstF++;
+      } else if (pSetNew->nSstF < pSetOld->nSstF) {
+        ASSERT(pSetNew->nSstF == 1);
+        for (int32_t iSst = 0; iSst < pSetOld->nSstF; iSst++) {
+          SSstFile *pSstFile = pSetOld->aSstF[iSst];
+          nRef = atomic_sub_fetch_32(&pSstFile->nRef, 1);
+          if (nRef == 0) {
+            tsdbSstFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSstFile, fname);
+            taosRemoveFile(fname);
+            taosMemoryFree(pSstFile);
+          }
+          pSetOld->aSstF[iSst] = NULL;
+        }
+
+        pSetOld->nSstF = 1;
+        pSetOld->aSstF[0] = (SSstFile *)taosMemoryMalloc(sizeof(SSstFile));
+        if (pSetOld->aSstF[0] == NULL) {
+          code = TSDB_CODE_OUT_OF_MEMORY;
+          goto _err;
+        }
+        *pSetOld->aSstF[0] = *pSetNew->aSstF[0];
+        pSetOld->aSstF[0]->nRef = 1;
+      } else {
+        for (int32_t iSst = 0; iSst < pSetOld->nSstF; iSst++) {
+          if (pSetOld->aSstF[iSst]->commitID != pSetNew->aSstF[iSst]->commitID) {
+            SSstFile *pSstFile = pSetOld->aSstF[iSst];
+            nRef = atomic_sub_fetch_32(&pSstFile->nRef, 1);
+            if (nRef == 0) {
+              tsdbSstFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSstFile, fname);
+              taosRemoveFile(fname);
+              taosMemoryFree(pSstFile);
+            }
+
+            pSetOld->aSstF[iSst] = (SSstFile *)taosMemoryMalloc(sizeof(SSstFile));
+            if (pSetOld->aSstF[iSst] == NULL) {
+              code = TSDB_CODE_OUT_OF_MEMORY;
+              goto _err;
+            }
+            *pSetOld->aSstF[iSst] = *pSetNew->aSstF[iSst];
+            pSetOld->aSstF[iSst]->nRef = 1;
+          } else {
+            ASSERT(pSetOld->aSstF[iSst]->size == pSetOld->aSstF[iSst]->size);
+            ASSERT(pSetOld->aSstF[iSst]->offset == pSetOld->aSstF[iSst]->offset);
+          }
+        }
+      }
+    } else {
+      ASSERT(pSetOld->nSstF == pSetNew->nSstF);
+      for (int32_t iSst = 0; iSst < pSetOld->nSstF; iSst++) {
+        SSstFile *pSstFile = pSetOld->aSstF[iSst];
+        nRef = atomic_sub_fetch_32(&pSstFile->nRef, 1);
+        if (nRef == 0) {
+          tsdbSstFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSstFile, fname);
+          taosRemoveFile(fname);
+          taosMemoryFree(pSstFile);
+        }
+
+        pSetOld->aSstF[iSst] = (SSstFile *)taosMemoryMalloc(sizeof(SSstFile));
+        if (pSetOld->aSstF[iSst] == NULL) {
+          code = TSDB_CODE_OUT_OF_MEMORY;
+          goto _err;
+        }
+        *pSetOld->aSstF[iSst] = *pSetNew->aSstF[iSst];
+        pSetOld->aSstF[iSst]->nRef = 1;
+      }
+    }
+
     if (!sameDisk) {
       pSetOld->diskId = pSetNew->diskId;
     }
@@ -902,13 +965,6 @@ int32_t tsdbFSCommit2(STsdb *pTsdb, STsdbFS *pFSNew) {
       taosMemoryFree(pSetOld->pDataF);
     }
 
-    nRef = atomic_sub_fetch_32(&pSetOld->pLastF->nRef, 1);
-    if (nRef == 0) {
-      tsdbLastFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSetOld->pLastF, fname);
-      taosRemoveFile(fname);
-      taosMemoryFree(pSetOld->pLastF);
-    }
-
     nRef = atomic_sub_fetch_32(&pSetOld->pSmaF->nRef, 1);
     if (nRef == 0) {
       tsdbSmaFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSetOld->pSmaF, fname);
@@ -916,12 +972,20 @@ int32_t tsdbFSCommit2(STsdb *pTsdb, STsdbFS *pFSNew) {
       taosMemoryFree(pSetOld->pSmaF);
     }
 
+    for (int8_t iSst = 0; iSst < pSetOld->nSstF; iSst++) {
+      nRef = atomic_sub_fetch_32(&pSetOld->aSstF[iSst]->nRef, 1);
+      if (nRef == 0) {
+        tsdbSstFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSetOld->aSstF[iSst], fname);
+        taosRemoveFile(fname);
+        taosMemoryFree(pSetOld->aSstF[iSst]);
+      }
+    }
+
     taosArrayRemove(pTsdb->fs.aDFileSet, iOld);
     continue;
 
   _add_new:
-    fSet.diskId = pSetNew->diskId;
-    fSet.fid = pSetNew->fid;
+    fSet = (SDFileSet){.diskId = pSetNew->diskId, .fid = pSetNew->fid, .nSstF = 1};
 
     // head
     fSet.pHeadF = (SHeadFile *)taosMemoryMalloc(sizeof(SHeadFile));
@@ -941,15 +1005,6 @@ int32_t tsdbFSCommit2(STsdb *pTsdb, STsdbFS *pFSNew) {
     *fSet.pDataF = *pSetNew->pDataF;
     fSet.pDataF->nRef = 1;
 
-    // last
-    fSet.pLastF = (SLastFile *)taosMemoryMalloc(sizeof(SLastFile));
-    if (fSet.pLastF == NULL) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _err;
-    }
-    *fSet.pLastF = *pSetNew->pLastF;
-    fSet.pLastF->nRef = 1;
-
     // sma
     fSet.pSmaF = (SSmaFile *)taosMemoryMalloc(sizeof(SSmaFile));
     if (fSet.pSmaF == NULL) {
@@ -959,6 +1014,16 @@ int32_t tsdbFSCommit2(STsdb *pTsdb, STsdbFS *pFSNew) {
     *fSet.pSmaF = *pSetNew->pSmaF;
     fSet.pSmaF->nRef = 1;
 
+    // sst
+    ASSERT(pSetNew->nSstF == 1);
+    fSet.aSstF[0] = (SSstFile *)taosMemoryMalloc(sizeof(SSstFile));
+    if (fSet.aSstF[0] == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _err;
+    }
+    *fSet.aSstF[0] = *pSetNew->aSstF[0];
+    fSet.aSstF[0]->nRef = 1;
+
     if (taosArrayInsert(pTsdb->fs.aDFileSet, iOld, &fSet) == NULL) {
       code = TSDB_CODE_OUT_OF_MEMORY;
       goto _err;
@@ -1002,12 +1067,14 @@ int32_t tsdbFSRef(STsdb *pTsdb, STsdbFS *pFS) {
     nRef = atomic_fetch_add_32(&pSet->pDataF->nRef, 1);
     ASSERT(nRef > 0);
 
-    nRef = atomic_fetch_add_32(&pSet->pLastF->nRef, 1);
-    ASSERT(nRef > 0);
-
     nRef = atomic_fetch_add_32(&pSet->pSmaF->nRef, 1);
     ASSERT(nRef > 0);
 
+    for (int32_t iSst = 0; iSst < pSet->nSstF; iSst++) {
+      nRef = atomic_fetch_add_32(&pSet->aSstF[iSst]->nRef, 1);
+      ASSERT(nRef > 0);
+    }
+
     if (taosArrayPush(pFS->aDFileSet, &fSet) == NULL) {
       code = TSDB_CODE_OUT_OF_MEMORY;
       goto _exit;
@@ -1053,15 +1120,6 @@ void tsdbFSUnref(STsdb *pTsdb, STsdbFS *pFS) {
       taosMemoryFree(pSet->pDataF);
     }
 
-    // last
-    nRef = atomic_sub_fetch_32(&pSet->pLastF->nRef, 1);
-    ASSERT(nRef >= 0);
-    if (nRef == 0) {
-      tsdbLastFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pLastF, fname);
-      taosRemoveFile(fname);
-      taosMemoryFree(pSet->pLastF);
-    }
-
     // sma
     nRef = atomic_sub_fetch_32(&pSet->pSmaF->nRef, 1);
     ASSERT(nRef >= 0);
@@ -1070,6 +1128,18 @@ void tsdbFSUnref(STsdb *pTsdb, STsdbFS *pFS) {
       taosRemoveFile(fname);
       taosMemoryFree(pSet->pSmaF);
     }
+
+    // sst
+    for (int32_t iSst = 0; iSst < pSet->nSstF; iSst++) {
+      nRef = atomic_sub_fetch_32(&pSet->aSstF[iSst]->nRef, 1);
+      ASSERT(nRef >= 0);
+      if (nRef == 0) {
+        tsdbSstFileName(pTsdb, pSet->diskId, pSet->fid, pSet->aSstF[iSst], fname);
+        taosRemoveFile(fname);
+        taosMemoryFree(pSet->aSstF[iSst]);
+        /* code */
+      }
+    }
   }
 
   taosArrayDestroy(pFS->aDFileSet);

source/dnode/vnode/src/tsdb/tsdbFile.c

@@ -53,22 +53,22 @@ static int32_t tGetDataFile(uint8_t *p, SDataFile *pDataFile) {
   return n;
 }
 
-int32_t tPutLastFile(uint8_t *p, SLastFile *pLastFile) {
+int32_t tPutSstFile(uint8_t *p, SSstFile *pSstFile) {
   int32_t n = 0;
 
-  n += tPutI64v(p ? p + n : p, pLastFile->commitID);
-  n += tPutI64v(p ? p + n : p, pLastFile->size);
-  n += tPutI64v(p ? p + n : p, pLastFile->offset);
+  n += tPutI64v(p ? p + n : p, pSstFile->commitID);
+  n += tPutI64v(p ? p + n : p, pSstFile->size);
+  n += tPutI64v(p ? p + n : p, pSstFile->offset);
 
   return n;
 }
 
-static int32_t tGetLastFile(uint8_t *p, SLastFile *pLastFile) {
+static int32_t tGetSstFile(uint8_t *p, SSstFile *pSstFile) {
   int32_t n = 0;
 
-  n += tGetI64v(p + n, &pLastFile->commitID);
-  n += tGetI64v(p + n, &pLastFile->size);
-  n += tGetI64v(p + n, &pLastFile->offset);
+  n += tGetI64v(p + n, &pSstFile->commitID);
+  n += tGetI64v(p + n, &pSstFile->size);
+  n += tGetI64v(p + n, &pSstFile->offset);
 
   return n;
 }
@@ -102,9 +102,9 @@ void tsdbDataFileName(STsdb *pTsdb, SDiskID did, int32_t fid, SDataFile *pDataF,
            TD_DIRSEP, pTsdb->path, TD_DIRSEP, TD_VID(pTsdb->pVnode), fid, pDataF->commitID, ".data");
 }
 
-void tsdbLastFileName(STsdb *pTsdb, SDiskID did, int32_t fid, SLastFile *pLastF, char fname[]) {
+void tsdbSstFileName(STsdb *pTsdb, SDiskID did, int32_t fid, SSstFile *pSstF, char fname[]) {
   snprintf(fname, TSDB_FILENAME_LEN - 1, "%s%s%s%sv%df%dver%" PRId64 "%s", tfsGetDiskPath(pTsdb->pVnode->pTfs, did),
-           TD_DIRSEP, pTsdb->path, TD_DIRSEP, TD_VID(pTsdb->pVnode), fid, pLastF->commitID, ".last");
+           TD_DIRSEP, pTsdb->path, TD_DIRSEP, TD_VID(pTsdb->pVnode), fid, pSstF->commitID, ".sst");
 }
 
 void tsdbSmaFileName(STsdb *pTsdb, SDiskID did, int32_t fid, SSmaFile *pSmaF, char fname[]) {
@@ -194,9 +194,11 @@ int32_t tPutDFileSet(uint8_t *p, SDFileSet *pSet) {
   n += tPutDataFile(p ? p + n : p, pSet->pDataF);
   n += tPutSmaFile(p ? p + n : p, pSet->pSmaF);
 
-  // last
-  n += tPutU8(p ? p + n : p, 1);  // for future compatibility
-  n += tPutLastFile(p ? p + n : p, pSet->pLastF);
+  // sst
+  n += tPutU8(p ? p + n : p, pSet->nSstF);
+  for (int32_t iSst = 0; iSst < pSet->nSstF; iSst++) {
+    n += tPutSstFile(p ? p + n : p, pSet->aSstF[iSst]);
+  }
 
   return n;
 }
@@ -208,15 +210,40 @@ int32_t tGetDFileSet(uint8_t *p, SDFileSet *pSet) {
   n += tGetI32v(p + n, &pSet->diskId.id);
   n += tGetI32v(p + n, &pSet->fid);
 
-  // data
+  // head
+  pSet->pHeadF = (SHeadFile *)taosMemoryCalloc(1, sizeof(SHeadFile));
+  if (pSet->pHeadF == NULL) {
+    return -1;
+  }
+  pSet->pHeadF->nRef = 1;
   n += tGetHeadFile(p + n, pSet->pHeadF);
+
+  // data
+  pSet->pDataF = (SDataFile *)taosMemoryCalloc(1, sizeof(SDataFile));
+  if (pSet->pDataF == NULL) {
+    return -1;
+  }
+  pSet->pDataF->nRef = 1;
   n += tGetDataFile(p + n, pSet->pDataF);
+
+  // sma
+  pSet->pSmaF = (SSmaFile *)taosMemoryCalloc(1, sizeof(SSmaFile));
+  if (pSet->pSmaF == NULL) {
+    return -1;
+  }
+  pSet->pSmaF->nRef = 1;
   n += tGetSmaFile(p + n, pSet->pSmaF);
 
-  // last
-  uint8_t nLast;
-  n += tGetU8(p + n, &nLast);
-  n += tGetLastFile(p + n, pSet->pLastF);
+  // sst
+  n += tGetU8(p + n, &pSet->nSstF);
+  for (int32_t iSst = 0; iSst < pSet->nSstF; iSst++) {
+    pSet->aSstF[iSst] = (SSstFile *)taosMemoryCalloc(1, sizeof(SSstFile));
+    if (pSet->aSstF[iSst] == NULL) {
+      return -1;
+    }
+    pSet->aSstF[iSst]->nRef = 1;
+    n += tGetSstFile(p + n, pSet->aSstF[iSst]);
+  }
 
   return n;
 }

source/dnode/vnode/src/tsdb/tsdbMergeTree.c

+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "tsdb.h"
+
+// SLDataIter =================================================
+typedef struct SLDataIter {
+  SRBTreeNode   node;
+  SSstBlk      *pSstBlk;
+  SDataFReader *pReader;
+  int32_t       iSst;
+  int8_t        backward;
+  SArray       *aSstBlk;
+  int32_t       iSstBlk;
+  SBlockData    bData[2];
+  int32_t       loadIndex;
+  int32_t       iRow;
+  SRowInfo      rInfo;
+  uint64_t      uid;
+  STimeWindow   timeWindow;
+  SVersionRange verRange;
+} SLDataIter;
+
+static SBlockData* getCurrentBlock(SLDataIter* pIter) {
+  return &pIter->bData[pIter->loadIndex];
+}
+
+static SBlockData* getNextBlock(SLDataIter* pIter) {
+  pIter->loadIndex ^= 1;
+  return getCurrentBlock(pIter);
+}
+
+int32_t tLDataIterOpen(struct SLDataIter **pIter, SDataFReader *pReader, int32_t iSst, int8_t backward, uint64_t uid,
+                       STimeWindow *pTimeWindow, SVersionRange *pRange) {
+  int32_t code = 0;
+  *pIter = taosMemoryCalloc(1, sizeof(SLDataIter));
+  if (*pIter == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _exit;
+  }
+
+  (*pIter)->uid = uid;
+  (*pIter)->timeWindow = *pTimeWindow;
+  (*pIter)->verRange = *pRange;
+  (*pIter)->pReader = pReader;
+  (*pIter)->iSst = iSst;
+  (*pIter)->backward = backward;
+  (*pIter)->aSstBlk = taosArrayInit(0, sizeof(SSstBlk));
+  if ((*pIter)->aSstBlk == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _exit;
+  }
+
+  code = tBlockDataCreate(&(*pIter)->bData[0]);
+  if (code) {
+    goto _exit;
+  }
+
+  code = tBlockDataCreate(&(*pIter)->bData[1]);
+  if (code) {
+    goto _exit;
+  }
+
+  code = tsdbReadSstBlk(pReader, iSst, (*pIter)->aSstBlk);
+  if (code) {
+    goto _exit;
+  }
+
+  size_t size = taosArrayGetSize((*pIter)->aSstBlk);
+
+  // find the start block
+  int32_t index = -1;
+  if (!backward) {  // asc
+    for (int32_t i = 0; i < size; ++i) {
+      SSstBlk *p = taosArrayGet((*pIter)->aSstBlk, i);
+      if (p->minUid <= uid && p->maxUid >= uid) {
+        index = i;
+        break;
+      }
+    }
+  } else {  // desc
+    for (int32_t i = size - 1; i >= 0; --i) {
+      SSstBlk *p = taosArrayGet((*pIter)->aSstBlk, i);
+      if (p->minUid <= uid && p->maxUid >= uid) {
+        index = i;
+        break;
+      }
+    }
+  }
+
+  (*pIter)->iSstBlk = index;
+  if (index != -1) {
+    (*pIter)->pSstBlk = taosArrayGet((*pIter)->aSstBlk, (*pIter)->iSstBlk);
+  }
+
+_exit:
+  return code;
+}
+
+void tLDataIterClose(SLDataIter *pIter) {
+  tBlockDataDestroy(&pIter->bData[0], 1);
+  tBlockDataDestroy(&pIter->bData[1], 1);
+  taosArrayDestroy(pIter->aSstBlk);
+  taosMemoryFree(pIter);
+}
+
+extern int32_t tsdbReadSstBlockEx(SDataFReader *pReader, int32_t iSst, SSstBlk *pSstBlk, SBlockData *pBlockData);
+
+void tLDataIterNextBlock(SLDataIter *pIter) {
+  int32_t step = pIter->backward ? -1 : 1;
+  pIter->iSstBlk += step;
+
+  int32_t index = -1;
+  size_t  size = taosArrayGetSize(pIter->aSstBlk);
+  for (int32_t i = pIter->iSstBlk; i < size && i >= 0; i += step) {
+    SSstBlk *p = taosArrayGet(pIter->aSstBlk, i);
+    if ((!pIter->backward) && p->minUid > pIter->uid) {
+      break;
+    }
+
+    if (pIter->backward && p->maxUid < pIter->uid) {
+      break;
+    }
+
+    if (p->minUid <= pIter->uid && p->maxUid >= pIter->uid) {
+      index = i;
+      break;
+    }
+  }
+
+  if (index == -1) {
+    pIter->pSstBlk = NULL;
+  } else {
+    pIter->pSstBlk = (SSstBlk *)taosArrayGet(pIter->aSstBlk, pIter->iSstBlk);
+  }
+}
+
+static void findNextValidRow(SLDataIter *pIter) {
+  int32_t step = pIter->backward ? -1 : 1;
+
+  bool    hasVal = false;
+  int32_t i = pIter->iRow;
+  SBlockData* pBlockData = getCurrentBlock(pIter);
+
+  for (; i < pBlockData->nRow && i >= 0; i += step) {
+    if (pBlockData->aUid != NULL) {
+      if (!pIter->backward) {
+        if (pBlockData->aUid[i] < pIter->uid) {
+          continue;
+        } else if (pBlockData->aUid[i] > pIter->uid) {
+          break;
+        }
+      } else {
+        if (pBlockData->aUid[i] > pIter->uid) {
+          continue;
+        } else if (pBlockData->aUid[i] < pIter->uid) {
+          break;
+        }
+      }
+    }
+
+    int64_t ts = pBlockData->aTSKEY[i];
+    if (!pIter->backward) {               // asc
+      if (ts > pIter->timeWindow.ekey) {  // no more data
+        break;
+      } else if (ts < pIter->timeWindow.skey) {
+        continue;
+      }
+    } else {
+      if (ts < pIter->timeWindow.skey) {
+        break;
+      } else if (ts > pIter->timeWindow.ekey) {
+        continue;
+      }
+    }
+
+    int64_t ver = pBlockData->aVersion[i];
+    if (ver < pIter->verRange.minVer) {
+      continue;
+    }
+
+    // todo opt handle desc case
+    if (ver > pIter->verRange.maxVer) {
+      continue;
+    }
+
+    //  todo handle delete soon
+#if 0
+    TSDBKEY k = {.ts = ts, .version = ver};
+        if (hasBeenDropped(pBlockScanInfo->delSkyline, &pBlockScanInfo->lastBlockDelIndex, &k, pLastBlockReader->order)) {
+          continue;
+        }
+#endif
+
+    hasVal = true;
+    break;
+  }
+
+  pIter->iRow = (hasVal) ? i : -1;
+}
+
+bool tLDataIterNextRow(SLDataIter *pIter) {
+  int32_t code = 0;
+  int32_t step = pIter->backward ? -1 : 1;
+
+  // no qualified last file block in current file, no need to fetch row
+  if (pIter->pSstBlk == NULL) {
+    return false;
+  }
+
+  int32_t iBlockL = pIter->iSstBlk;
+  SBlockData* pBlockData = getCurrentBlock(pIter);
+
+  if (pBlockData->nRow == 0 && pIter->pSstBlk != NULL) {  // current block not loaded yet
+    pBlockData = getNextBlock(pIter);
+    code = tsdbReadSstBlockEx(pIter->pReader, pIter->iSst, pIter->pSstBlk, pBlockData);
+    if (code != TSDB_CODE_SUCCESS) {
+      goto _exit;
+    }
+
+    pIter->iRow = (pIter->backward) ? pBlockData->nRow : -1;
+  }
+
+  pIter->iRow += step;
+
+  while (1) {
+    findNextValidRow(pIter);
+
+    if (pIter->iRow >= pBlockData->nRow || pIter->iRow < 0) {
+      tLDataIterNextBlock(pIter);
+      if (pIter->pSstBlk == NULL) {  // no more data
+        goto _exit;
+      }
+    } else {
+      break;
+    }
+
+    if (iBlockL != pIter->iSstBlk) {
+      pBlockData = getNextBlock(pIter);
+      code = tsdbReadSstBlockEx(pIter->pReader, pIter->iSst, pIter->pSstBlk, pBlockData);
+      if (code) {
+        goto _exit;
+      }
+      pIter->iRow = pIter->backward ? (pBlockData->nRow - 1) : 0;
+    }
+  }
+
+  pIter->rInfo.suid = pBlockData->suid;
+  pIter->rInfo.uid = pBlockData->uid;
+  pIter->rInfo.row = tsdbRowFromBlockData(pBlockData, pIter->iRow);
+
+_exit:
+  if (code != TSDB_CODE_SUCCESS) {
+    terrno = code;
+  }
+
+  return (code == TSDB_CODE_SUCCESS) && (pIter->pSstBlk != NULL);
+}
+
+SRowInfo *tLDataIterGet(SLDataIter *pIter) { return &pIter->rInfo; }
+
+// SMergeTree =================================================
+static FORCE_INLINE int32_t tLDataIterCmprFn(const void *p1, const void *p2) {
+  SLDataIter *pIter1 = (SLDataIter *)(((uint8_t *)p1) - sizeof(SRBTreeNode));
+  SLDataIter *pIter2 = (SLDataIter *)(((uint8_t *)p2) - sizeof(SRBTreeNode));
+
+  TSDBKEY key1 = TSDBROW_KEY(&pIter1->rInfo.row);
+  TSDBKEY key2 = TSDBROW_KEY(&pIter2->rInfo.row);
+
+  if (key1.ts < key2.ts) {
+    return -1;
+  } else if (key1.ts > key2.ts) {
+    return 1;
+  } else {
+    if (key1.version < key2.version) {
+      return -1;
+    } else if (key1.version > key2.version) {
+      return 1;
+    } else {
+      return 0;
+    }
+  }
+}
+
+int32_t tMergeTreeOpen(SMergeTree *pMTree, int8_t backward, SDataFReader *pFReader, uint64_t uid,
+                       STimeWindow *pTimeWindow, SVersionRange *pVerRange) {
+  pMTree->backward = backward;
+  pMTree->pIter = NULL;
+  pMTree->pIterList = taosArrayInit(4, POINTER_BYTES);
+  if (pMTree->pIterList == NULL) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  tRBTreeCreate(&pMTree->rbt, tLDataIterCmprFn);
+  int32_t code = TSDB_CODE_OUT_OF_MEMORY;
+
+  struct SLDataIter *pIterList[TSDB_DEFAULT_LAST_FILE] = {0};
+  for (int32_t i = 0; i < pFReader->pSet->nSstF; ++i) {  // open all last file
+    code = tLDataIterOpen(&pIterList[i], pFReader, i, pMTree->backward, uid, pTimeWindow, pVerRange);
+    if (code != TSDB_CODE_SUCCESS) {
+      goto _end;
+    }
+
+    bool hasVal = tLDataIterNextRow(pIterList[i]);
+    if (hasVal) {
+      taosArrayPush(pMTree->pIterList, &pIterList[i]);
+      tMergeTreeAddIter(pMTree, pIterList[i]);
+    } else {
+      tLDataIterClose(pIterList[i]);
+    }
+  }
+
+  return code;
+
+_end:
+  tMergeTreeClose(pMTree);
+  return code;
+}
+
+void tMergeTreeAddIter(SMergeTree *pMTree, SLDataIter *pIter) { tRBTreePut(&pMTree->rbt, (SRBTreeNode *)pIter); }
+
+bool tMergeTreeNext(SMergeTree *pMTree) {
+  int32_t code = TSDB_CODE_SUCCESS;
+  if (pMTree->pIter) {
+    SLDataIter *pIter = pMTree->pIter;
+
+    bool hasVal = tLDataIterNextRow(pIter);
+    if (!hasVal) {
+      pMTree->pIter = NULL;
+    }
+
+    // compare with min in RB Tree
+    pIter = (SLDataIter *)tRBTreeMin(&pMTree->rbt);
+    if (pMTree->pIter && pIter) {
+      int32_t c = pMTree->rbt.cmprFn(RBTREE_NODE_PAYLOAD(&pMTree->pIter->node), RBTREE_NODE_PAYLOAD(&pIter->node));
+      if (c > 0) {
+        tRBTreePut(&pMTree->rbt, (SRBTreeNode *)pMTree->pIter);
+        pMTree->pIter = NULL;
+      } else {
+        ASSERT(c);
+      }
+    }
+  }
+
+  if (pMTree->pIter == NULL) {
+    pMTree->pIter = (SLDataIter *)tRBTreeMin(&pMTree->rbt);
+    if (pMTree->pIter) {
+      tRBTreeDrop(&pMTree->rbt, (SRBTreeNode *)pMTree->pIter);
+    }
+  }
+
+  return pMTree->pIter != NULL;
+}
+
+TSDBROW tMergeTreeGetRow(SMergeTree *pMTree) { return pMTree->pIter->rInfo.row; }
+
+void tMergeTreeClose(SMergeTree *pMTree) {
+  size_t size = taosArrayGetSize(pMTree->pIterList);
+  for (int32_t i = 0; i < size; ++i) {
+    SLDataIter *pIter = taosArrayGetP(pMTree->pIterList, i);
+    tLDataIterClose(pIter);
+  }
+
+  pMTree->pIterList = taosArrayDestroy(pMTree->pIterList);
+  pMTree->pIter = NULL;
+}

source/dnode/vnode/src/tsdb/tsdbRetention.c

@@ -60,7 +60,7 @@ int32_t tsdbDoRetention(STsdb *pTsdb, int64_t now) {
     if (expLevel < 0) {
       taosMemoryFree(pSet->pHeadF);
       taosMemoryFree(pSet->pDataF);
-      taosMemoryFree(pSet->pLastF);
+      taosMemoryFree(pSet->aSstF[0]);
       taosMemoryFree(pSet->pSmaF);
       taosArrayRemove(fs.aDFileSet, iSet);
       iSet--;

source/dnode/vnode/src/vnd/vnodeCfg.c

@@ -14,6 +14,7 @@
  */
 
 #include "vnd.h"
+#include "tutil.h"
 
 const SVnodeCfg vnodeCfgDefault = {.vgId = -1,
                                    .dbname = "",
@@ -106,9 +107,12 @@ int vnodeEncodeConfig(const void *pObj, SJson *pJson) {
   if (tjsonAddIntegerToObject(pJson, "wal.retentionSize", pCfg->walCfg.retentionSize) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "wal.segSize", pCfg->walCfg.segSize) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "wal.level", pCfg->walCfg.level) < 0) return -1;
+  if (tjsonAddIntegerToObject(pJson, "sstTrigger", pCfg->sstTrigger) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "hashBegin", pCfg->hashBegin) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "hashEnd", pCfg->hashEnd) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "hashMethod", pCfg->hashMethod) < 0) return -1;
+  if (tjsonAddIntegerToObject(pJson, "hashPrefix", pCfg->hashPrefix) < 0) return -1;
+  if (tjsonAddIntegerToObject(pJson, "hashSuffix", pCfg->hashSuffix) < 0) return -1;
 
   if (tjsonAddIntegerToObject(pJson, "syncCfg.replicaNum", pCfg->syncCfg.replicaNum) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "syncCfg.myIndex", pCfg->syncCfg.myIndex) < 0) return -1;
@@ -205,12 +209,18 @@ int vnodeDecodeConfig(const SJson *pJson, void *pObj) {
   if (code < 0) return -1;
   tjsonGetNumberValue(pJson, "wal.level", pCfg->walCfg.level, code);
   if (code < 0) return -1;
+  tjsonGetNumberValue(pJson, "sstTrigger", pCfg->sstTrigger, code);
+  if (code < 0) return -1;
   tjsonGetNumberValue(pJson, "hashBegin", pCfg->hashBegin, code);
   if (code < 0) return -1;
   tjsonGetNumberValue(pJson, "hashEnd", pCfg->hashEnd, code);
   if (code < 0) return -1;
   tjsonGetNumberValue(pJson, "hashMethod", pCfg->hashMethod, code);
   if (code < 0) return -1;
+  tjsonGetNumberValue(pJson, "hashPrefix", pCfg->hashPrefix, code);
+  if (code < 0) return -1;
+  tjsonGetNumberValue(pJson, "hashSuffix", pCfg->hashSuffix, code);
+  if (code < 0) return -1;
 
   tjsonGetNumberValue(pJson, "syncCfg.replicaNum", pCfg->syncCfg.replicaNum, code);
   if (code < 0) return -1;
@@ -247,7 +257,8 @@ int vnodeValidateTableHash(SVnode *pVnode, char *tableFName) {
 
   switch (pVnode->config.hashMethod) {
     default:
-      hashValue = MurmurHash3_32(tableFName, strlen(tableFName));
+      hashValue = taosGetTbHashVal(tableFName, strlen(tableFName), pVnode->config.hashMethod, pVnode->config.hashPrefix,
+                                   pVnode->config.hashSuffix);
       break;
   }
 

source/dnode/vnode/src/vnd/vnodeQuery.c

@@ -368,6 +368,7 @@ _exit:
 int32_t vnodeGetLoad(SVnode *pVnode, SVnodeLoad *pLoad) {
   pLoad->vgId = TD_VID(pVnode);
   pLoad->syncState = syncGetMyRole(pVnode->sync);
+  pLoad->cacheUsage = tsdbCacheGetUsage(pVnode);
   pLoad->numOfTables = metaGetTbNum(pVnode->pMeta);
   pLoad->numOfTimeSeries = metaGetTimeSeriesNum(pVnode->pMeta);
   pLoad->totalStorage = (int64_t)3 * 1073741824;
@@ -424,8 +425,8 @@ int32_t vnodeGetCtbIdList(SVnode *pVnode, int64_t suid, SArray *list) {
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t vnodeGetStbIdList(SVnode* pVnode, int64_t suid, SArray* list) {
-  SMStbCursor* pCur = metaOpenStbCursor(pVnode->pMeta, suid);
+int32_t vnodeGetStbIdList(SVnode *pVnode, int64_t suid, SArray *list) {
+  SMStbCursor *pCur = metaOpenStbCursor(pVnode->pMeta, suid);
   if (!pCur) {
     return TSDB_CODE_FAILED;
   }
@@ -467,9 +468,13 @@ static int32_t vnodeGetStbColumnNum(SVnode *pVnode, tb_uid_t suid, int *num) {
   STSchema *pTSchema = metaGetTbTSchema(pVnode->pMeta, suid, -1);
   // metaGetTbTSchemaEx(pVnode->pMeta, suid, suid, -1, &pTSchema);
 
-  *num = pTSchema->numOfCols;
+  if (pTSchema) {
+    *num = pTSchema->numOfCols;
 
-  taosMemoryFree(pTSchema);
+    taosMemoryFree(pTSchema);
+  } else {
+    *num = 2;
+  }
 
   return TSDB_CODE_SUCCESS;
 }

source/dnode/vnode/src/vnd/vnodeSnapshot.c

@@ -39,7 +39,7 @@ struct SVSnapReader {
   SStreamStateReader *pStreamStateReader;
   // rsma
   int8_t           rsmaDone;
-  SRsmaSnapReader *pRsmaReader;
+  SRSmaSnapReader *pRsmaReader;
 };
 
 int32_t vnodeSnapReaderOpen(SVnode *pVnode, int64_t sver, int64_t ever, SVSnapReader **ppReader) {
@@ -241,7 +241,7 @@ struct SVSnapWriter {
   SStreamTaskWriter  *pStreamTaskWriter;
   SStreamStateWriter *pStreamStateWriter;
   // rsma
-  SRsmaSnapWriter *pRsmaSnapWriter;
+  SRSmaSnapWriter *pRsmaSnapWriter;
 };
 
 int32_t vnodeSnapWriterOpen(SVnode *pVnode, int64_t sver, int64_t ever, SVSnapWriter **ppWriter) {

source/dnode/vnode/src/vnd/vnodeSvr.c

@@ -371,7 +371,7 @@ void vnodeUpdateMetaRsp(SVnode *pVnode, STableMetaRsp *pMetaRsp) {
   if (NULL == pMetaRsp) {
     return;
   }
-  
+
   strcpy(pMetaRsp->dbFName, pVnode->config.dbname);
   pMetaRsp->dbId = pVnode->config.dbId;
   pMetaRsp->vgId = TD_VID(pVnode);
@@ -527,7 +527,7 @@ static int32_t vnodeProcessCreateTbReq(SVnode *pVnode, int64_t version, void *pR
       cRsp.code = TSDB_CODE_SUCCESS;
       tdFetchTbUidList(pVnode->pSma, &pStore, pCreateReq->ctb.suid, pCreateReq->uid);
       taosArrayPush(tbUids, &pCreateReq->uid);
-      vnodeUpdateMetaRsp(pVnode, cRsp.pMeta); 
+      vnodeUpdateMetaRsp(pVnode, cRsp.pMeta);
     }
 
     taosArrayPush(rsp.pArray, &cRsp);
@@ -1107,6 +1107,7 @@ static int32_t vnodeProcessDeleteReq(SVnode *pVnode, int64_t version, void *pReq
 
   tDecoderInit(pCoder, pReq, len);
   tDecodeDeleteRes(pCoder, pRes);
+  ASSERT(taosArrayGetSize(pRes->uidList) == 0 || (pRes->skey != 0 && pRes->ekey != 0));
 
   for (int32_t iUid = 0; iUid < taosArrayGetSize(pRes->uidList); iUid++) {
     code = tsdbDeleteTableData(pVnode->pTsdb, version, pRes->suid, *(uint64_t *)taosArrayGet(pRes->uidList, iUid),

source/libs/executor/inc/executil.h

@@ -87,9 +87,7 @@ struct SqlFunctionCtx;
 
 size_t getResultRowSize(struct SqlFunctionCtx* pCtx, int32_t numOfOutput);
 void   initResultRowInfo(SResultRowInfo* pResultRowInfo);
-
-void initResultRow(SResultRow* pResultRow);
-void closeResultRow(SResultRow* pResultRow);
+void   closeResultRow(SResultRow* pResultRow);
 
 struct SResultRowEntryInfo* getResultEntryInfo(const SResultRow* pRow, int32_t index, const int32_t* offset);
 

source/libs/executor/src/dataDeleter.c

@@ -168,7 +168,9 @@ static void getDataLength(SDataSinkHandle* pHandle, int64_t* pLen, bool* pQueryE
   taosReadQitem(pDeleter->pDataBlocks, (void**)&pBuf);
   memcpy(&pDeleter->nextOutput, pBuf, sizeof(SDataDeleterBuf));
   taosFreeQitem(pBuf);
-  *pLen = ((SDataCacheEntry*)(pDeleter->nextOutput.pData))->dataLen;
+
+  SDataCacheEntry* pEntry = (SDataCacheEntry*)pDeleter->nextOutput.pData;
+  *pLen = pEntry->dataLen;
   *pQueryEnd = pDeleter->queryEnd;
   qDebug("got data len %" PRId64 ", row num %d in sink", *pLen, ((SDataCacheEntry*)(pDeleter->nextOutput.pData))->numOfRows);
 }

source/libs/executor/src/executil.c

@@ -1302,7 +1302,6 @@ int32_t initQueryTableDataCond(SQueryTableDataCond* pCond, const STableScanPhysi
   pCond->type = TIMEWINDOW_RANGE_CONTAINED;
   pCond->startVersion = -1;
   pCond->endVersion = -1;
-  pCond->schemaVersion = -1;
   //  pCond->type = pTableScanNode->scanFlag;
 
   int32_t j = 0;

source/libs/executor/src/executorimpl.c

@@ -3731,7 +3731,6 @@ static int32_t initTableblockDistQueryCond(uint64_t uid, SQueryTableDataCond* pC
   pCond->type = TIMEWINDOW_RANGE_CONTAINED;
   pCond->startVersion = -1;
   pCond->endVersion = -1;
-  pCond->schemaVersion = -1;
 
   return TSDB_CODE_SUCCESS;
 }

source/libs/parser/inc/parAst.h

@@ -60,7 +60,9 @@ typedef enum EDatabaseOptionType {
   DB_OPTION_WAL_RETENTION_SIZE,
   DB_OPTION_WAL_ROLL_PERIOD,
   DB_OPTION_WAL_SEGMENT_SIZE,
-  DB_OPTION_SST_TRIGGER
+  DB_OPTION_SST_TRIGGER,
+  DB_OPTION_TABLE_PREFIX,
+  DB_OPTION_TABLE_SUFFIX
 } EDatabaseOptionType;
 
 typedef enum ETableOptionType {