Merge pull request #11115 from taosdata/feature/meta

Feature/meta

source/dnode/vnode/CMakeLists.txt

@@ -46,14 +46,11 @@ target_link_libraries(
     PUBLIC stream
     PUBLIC qworker
     PUBLIC sync
+    # TODO: get rid of BDB
+    PUBLIC bdb
+    PUBLIC tdb
 )
 
-if(${META_DB_IMPL} STREQUAL "BDB")
-    target_link_libraries(vnode PUBLIC bdb)
-elseif(${META_DB_IMPL} STREQUAL "TDB")
-    target_link_libraries(vnode PUBLIC tdb)
-endif()
-
 if(${BUILD_TEST})
     add_subdirectory(test)
 endif(${BUILD_TEST})

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

@@ -15,7 +15,7 @@
 
 #include "metaDef.h"
 
-#include "tdb.h"
+#include "tdbInt.h"
 
 struct SMetaDB {
   TENV *pEnv;
@@ -25,121 +25,627 @@ struct SMetaDB {
   TDB  *pStbIdx;
   TDB  *pNtbIdx;
   TDB  *pCtbIdx;
-  // tag index hash table
-  // suid+colid --> TDB *
-  struct {
-  } tagIdxHt;
 };
 
-#define A(op, flag)                 \
+typedef struct __attribute__((__packed__)) {
-  do {                              \
+  tb_uid_t uid;
-    if ((ret = op) != 0) goto flag; \
+  int32_t  sver;
-  } while (0)
+} SSchemaDbKey;
+
+typedef struct {
+  char    *name;
+  tb_uid_t uid;
+} SNameIdxKey;
+
+typedef struct {
+  tb_uid_t suid;
+  tb_uid_t uid;
+} SCtbIdxKey;
+
+static int   metaEncodeTbInfo(void **buf, STbCfg *pTbCfg);
+static void *metaDecodeTbInfo(void *buf, STbCfg *pTbCfg);
+static int   metaEncodeSchema(void **buf, SSchemaWrapper *pSW);
+static void *metaDecodeSchema(void *buf, SSchemaWrapper *pSW);
+
+static inline int metaUidCmpr(const void *arg1, int len1, const void *arg2, int len2) {
+  tb_uid_t uid1, uid2;
+
+  ASSERT(len1 == sizeof(tb_uid_t));
+  ASSERT(len2 == sizeof(tb_uid_t));
+
+  uid1 = ((tb_uid_t *)arg1)[0];
+  uid2 = ((tb_uid_t *)arg2)[0];
+
+  if (uid1 < uid2) {
+    return -1;
+  }
+  if (uid1 == uid2) {
+    return 0;
+  } else {
+    return 1;
+  }
+}
+
+static inline int metaSchemaKeyCmpr(const void *arg1, int len1, const void *arg2, int len2) {
+  int           c;
+  SSchemaDbKey *pKey1 = (SSchemaDbKey *)arg1;
+  SSchemaDbKey *pKey2 = (SSchemaDbKey *)arg2;
+
+  c = metaUidCmpr(arg1, sizeof(tb_uid_t), arg2, sizeof(tb_uid_t));
+  if (c) return c;
+
+  if (pKey1->sver > pKey2->sver) {
+    return 1;
+  } else if (pKey1->sver == pKey2->sver) {
+    return 0;
+  } else {
+    return -1;
+  }
+}
+
+static inline int metaNameIdxCmpr(const void *arg1, int len1, const void *arg2, int len2) {
+  return strcmp((char *)arg1, (char *)arg2);
+}
+
+static inline int metaCtbIdxCmpr(const void *arg1, int len1, const void *arg2, int len2) {
+  int         c;
+  SCtbIdxKey *pKey1 = (SCtbIdxKey *)arg1;
+  SCtbIdxKey *pKey2 = (SCtbIdxKey *)arg2;
+
+  c = metaUidCmpr(arg1, sizeof(tb_uid_t), arg2, sizeof(tb_uid_t));
+  if (c) return c;
+
+  return metaUidCmpr(&pKey1->uid, sizeof(tb_uid_t), &pKey2->uid, sizeof(tb_uid_t));
+}
 
 int metaOpenDB(SMeta *pMeta) {
-  SMetaDB *pDb;
+  SMetaDB *pMetaDb;
-  TENV *   pEnv;
-  TDB *    pTbDB;
-  TDB *    pSchemaDB;
-  TDB *    pNameIdx;
-  TDB *    pStbIdx;
-  TDB *    pNtbIdx;
-  TDB *    pCtbIdx;
   int      ret;
 
-  pDb = (SMetaDB *)taosMemoryCalloc(1, sizeof(*pDb));
+  // allocate DB handle
-  if (pDb == NULL) {
+  pMetaDb = taosMemoryCalloc(1, sizeof(*pMetaDb));
+  if (pMetaDb == NULL) {
+    // TODO
+    ASSERT(0);
     return -1;
   }
 
-  // Create and open the ENV
+  // open the ENV
-  A((tdbEnvCreate(&pEnv)), _err);
+  ret = tdbEnvOpen(pMeta->path, 4096, 256, &(pMetaDb->pEnv));
-#if 0
+  if (ret < 0) {
-  // Set options of the environment
+    // TODO
-  A(tdbEnvSetPageSize(pEnv, 8192), _err);
+    ASSERT(0);
-  A(tdbEnvSetCacheSize(pEnv, 16 * 1024 * 1024), _err);
-#endif
-  A((tdbEnvOpen(&pEnv)), _err);
-
-  // Create and open each DB
-  A(tdbCreate(&pTbDB), _err);
-  A(tdbOpen(&pTbDB, "table.db", NULL, pEnv), _err);
-
-  A(tdbCreate(&pSchemaDB), _err);
-  A(tdbOpen(&pSchemaDB, "schema.db", NULL, pEnv), _err);
-
-  A(tdbCreate(&pNameIdx), _err);
-  A(tdbOpen(&pNameIdx, "name.db", NULL, pEnv), _err);
-  // tdbAssociate();
-
-  pDb->pEnv = pEnv;
-  pDb->pTbDB = pTbDB;
-  pDb->pSchemaDB = pSchemaDB;
-  pMeta->pDB = pDb;
-  return 0;
-
-_err:
     return -1;
   }
 
+  // open table DB
+  ret = tdbDbOpen("table.db", sizeof(tb_uid_t), TDB_VARIANT_LEN, metaUidCmpr, pMetaDb->pEnv, &(pMetaDb->pTbDB));
+  if (ret < 0) {
+    // TODO
+    ASSERT(0);
+    return -1;
+  }
+
+  // open schema DB
+  ret = tdbDbOpen("schema.db", sizeof(SSchemaDbKey), TDB_VARIANT_LEN, metaSchemaKeyCmpr, pMetaDb->pEnv,
+                  &(pMetaDb->pSchemaDB));
+  if (ret < 0) {
+    // TODO
+    ASSERT(0);
+    return -1;
+  }
+
+  ret = tdbDbOpen("name.idx", TDB_VARIANT_LEN, 0, metaNameIdxCmpr, pMetaDb->pEnv, &(pMetaDb->pNameIdx));
+  if (ret < 0) {
+    // TODO
+    ASSERT(0);
+    return -1;
+  }
+
+  ret = tdbDbOpen("stb.idx", sizeof(tb_uid_t), 0, metaUidCmpr, pMetaDb->pEnv, &(pMetaDb->pStbIdx));
+  if (ret < 0) {
+    // TODO
+    ASSERT(0);
+    return -1;
+  }
+
+  ret = tdbDbOpen("ntb.idx", sizeof(tb_uid_t), 0, metaUidCmpr, pMetaDb->pEnv, &(pMetaDb->pNtbIdx));
+  if (ret < 0) {
+    // TODO
+    ASSERT(0);
+    return -1;
+  }
+
+  ret = tdbDbOpen("ctb.idx", sizeof(SCtbIdxKey), 0, metaCtbIdxCmpr, pMetaDb->pEnv, &(pMetaDb->pCtbIdx));
+  if (ret < 0) {
+    // TODO
+    ASSERT(0);
+    return -1;
+  }
+
+  pMeta->pDB = pMetaDb;
+  return 0;
+}
+
 void metaCloseDB(SMeta *pMeta) {
-  // TODO
+  if (pMeta->pDB) {
+    tdbDbClose(pMeta->pDB->pCtbIdx);
+    tdbDbClose(pMeta->pDB->pNtbIdx);
+    tdbDbClose(pMeta->pDB->pStbIdx);
+    tdbDbClose(pMeta->pDB->pNameIdx);
+    tdbDbClose(pMeta->pDB->pSchemaDB);
+    tdbDbClose(pMeta->pDB->pTbDB);
+    taosMemoryFree(pMeta->pDB);
+  }
 }
 
 int metaSaveTableToDB(SMeta *pMeta, STbCfg *pTbCfg) {
-  // TODO
+  tb_uid_t       uid;
+  SMetaDB       *pMetaDb;
+  void          *pKey;
+  void          *pVal;
+  int            kLen;
+  int            vLen;
+  int            ret;
+  char           buf[512];
+  void          *pBuf;
+  SCtbIdxKey     ctbIdxKey;
+  SSchemaDbKey   schemaDbKey;
+  SSchemaWrapper schemaWrapper;
+
+  pMetaDb = pMeta->pDB;
+
+  // TODO: make this operation pre-process
+  if (pTbCfg->type == META_SUPER_TABLE) {
+    uid = pTbCfg->stbCfg.suid;
+  } else {
+    uid = metaGenerateUid(pMeta);
+  }
+
+  // save to table.db
+  pKey = &uid;
+  kLen = sizeof(uid);
+  pVal = pBuf = buf;
+  metaEncodeTbInfo(&pBuf, pTbCfg);
+  vLen = POINTER_DISTANCE(pBuf, buf);
+  ret = tdbDbInsert(pMetaDb->pTbDB, pKey, kLen, pVal, vLen);
+  if (ret < 0) {
+    return -1;
+  }
+
+  // save to schema.db for META_SUPER_TABLE and META_NORMAL_TABLE
+  if (pTbCfg->type != META_CHILD_TABLE) {
+    schemaDbKey.uid = uid;
+    schemaDbKey.sver = 0;  // TODO
+    pKey = &schemaDbKey;
+    kLen = sizeof(schemaDbKey);
+
+    if (pTbCfg->type == META_SUPER_TABLE) {
+      schemaWrapper.nCols = pTbCfg->stbCfg.nCols;
+      schemaWrapper.pSchema = pTbCfg->stbCfg.pSchema;
+    } else {
+      schemaWrapper.nCols = pTbCfg->ntbCfg.nCols;
+      schemaWrapper.pSchema = pTbCfg->ntbCfg.pSchema;
+    }
+    pVal = pBuf = buf;
+    metaEncodeSchema(&pBuf, &schemaWrapper);
+    vLen = POINTER_DISTANCE(pBuf, buf);
+    ret = tdbDbInsert(pMetaDb->pSchemaDB, pKey, kLen, pVal, vLen);
+    if (ret < 0) {
+      return -1;
+    }
+  }
+
+  // update name.idx
+  int nameLen = strlen(pTbCfg->name);
+  memcpy(buf, pTbCfg->name, nameLen + 1);
+  ((tb_uid_t *)(buf + nameLen + 1))[0] = uid;
+  pKey = buf;
+  kLen = nameLen + 1 + sizeof(uid);
+  pVal = NULL;
+  vLen = 0;
+  ret = tdbDbInsert(pMetaDb->pNameIdx, pKey, kLen, pVal, vLen);
+  if (ret < 0) {
+    return -1;
+  }
+
+  // update other index
+  if (pTbCfg->type == META_SUPER_TABLE) {
+    pKey = &uid;
+    kLen = sizeof(uid);
+    pVal = NULL;
+    vLen = 0;
+    ret = tdbDbInsert(pMetaDb->pStbIdx, pKey, kLen, pVal, vLen);
+    if (ret < 0) {
+      return -1;
+    }
+  } else if (pTbCfg->type == META_CHILD_TABLE) {
+    ctbIdxKey.suid = pTbCfg->ctbCfg.suid;
+    ctbIdxKey.uid = uid;
+    pKey = &ctbIdxKey;
+    kLen = sizeof(ctbIdxKey);
+    pVal = NULL;
+    vLen = 0;
+    ret = tdbDbInsert(pMetaDb->pCtbIdx, pKey, kLen, pVal, vLen);
+    if (ret < 0) {
+      return -1;
+    }
+  } else if (pTbCfg->type == META_NORMAL_TABLE) {
+    pKey = &uid;
+    kLen = sizeof(uid);
+    pVal = NULL;
+    vLen = 0;
+    ret = tdbDbInsert(pMetaDb->pNtbIdx, pKey, kLen, pVal, vLen);
+    if (ret < 0) {
+      return -1;
+    }
+  }
+
   return 0;
 }
 
 int metaRemoveTableFromDb(SMeta *pMeta, tb_uid_t uid) {
   // TODO
+  ASSERT(0);
   return 0;
 }
 
 STbCfg *metaGetTbInfoByUid(SMeta *pMeta, tb_uid_t uid) {
-  // TODO
+  int      ret;
+  SMetaDB *pMetaDb = pMeta->pDB;
+  void    *pKey;
+  void    *pVal;
+  int      kLen;
+  int      vLen;
+  STbCfg  *pTbCfg;
+
+  // Fetch
+  pKey = &uid;
+  kLen = sizeof(uid);
+  pVal = NULL;
+  ret = tdbDbGet(pMetaDb->pTbDB, pKey, kLen, &pVal, &vLen);
+  if (ret < 0) {
     return NULL;
   }
 
+  // Decode
+  pTbCfg = taosMemoryMalloc(sizeof(*pTbCfg));
+  metaDecodeTbInfo(pVal, pTbCfg);
+
+  TDB_FREE(pVal);
+
+  return pTbCfg;
+}
+
 STbCfg *metaGetTbInfoByName(SMeta *pMeta, char *tbname, tb_uid_t *uid) {
-  // TODO
+  void *pKey;
+  void *pVal;
+  void *ppKey;
+  int   pkLen;
+  int   kLen;
+  int   vLen;
+  int   ret;
+
+  pKey = tbname;
+  kLen = strlen(tbname) + 1;
+  pVal = NULL;
+  ppKey = NULL;
+  ret = tdbDbPGet(pMeta->pDB->pNameIdx, pKey, kLen, &ppKey, &pkLen, &pVal, &vLen);
+  if (ret < 0) {
     return NULL;
   }
 
+  ASSERT(pkLen == kLen + sizeof(uid));
+
+  *uid = *(tb_uid_t *)POINTER_SHIFT(ppKey, kLen);
+  TDB_FREE(ppKey);
+  TDB_FREE(pVal);
+
+  return metaGetTbInfoByUid(pMeta, *uid);
+}
+
 SSchemaWrapper *metaGetTableSchema(SMeta *pMeta, tb_uid_t uid, int32_t sver, bool isinline) {
-  // TODO
+  void           *pKey;
+  void           *pVal;
+  int             kLen;
+  int             vLen;
+  int             ret;
+  SSchemaDbKey    schemaDbKey;
+  SSchemaWrapper *pSchemaWrapper;
+  void           *pBuf;
+
+  // fetch
+  schemaDbKey.uid = uid;
+  schemaDbKey.sver = sver;
+  pKey = &schemaDbKey;
+  kLen = sizeof(schemaDbKey);
+  pVal = NULL;
+  ret = tdbDbGet(pMeta->pDB->pSchemaDB, pKey, kLen, &pVal, &vLen);
+  if (ret < 0) {
     return NULL;
   }
 
+  // decode
+  pBuf = pVal;
+  pSchemaWrapper = taosMemoryMalloc(sizeof(*pSchemaWrapper));
+  metaDecodeSchema(pBuf, pSchemaWrapper);
+
+  TDB_FREE(pVal);
+
+  return pSchemaWrapper;
+}
+
 STSchema *metaGetTbTSchema(SMeta *pMeta, tb_uid_t uid, int32_t sver) {
-  // TODO
+  tb_uid_t        quid;
+  SSchemaWrapper *pSW;
+  STSchemaBuilder sb;
+  SSchema        *pSchema;
+  STSchema       *pTSchema;
+  STbCfg         *pTbCfg;
+
+  pTbCfg = metaGetTbInfoByUid(pMeta, uid);
+  if (pTbCfg->type == META_CHILD_TABLE) {
+    quid = pTbCfg->ctbCfg.suid;
+  } else {
+    quid = uid;
+  }
+
+  pSW = metaGetTableSchema(pMeta, quid, sver, true);
+  if (pSW == NULL) {
     return NULL;
   }
 
+  tdInitTSchemaBuilder(&sb, 0);
+  for (int i = 0; i < pSW->nCols; i++) {
+    pSchema = pSW->pSchema + i;
+    tdAddColToSchema(&sb, pSchema->type, pSchema->colId, pSchema->bytes);
+  }
+  pTSchema = tdGetSchemaFromBuilder(&sb);
+  tdDestroyTSchemaBuilder(&sb);
+
+  return pTSchema;
+}
+
+struct SMTbCursor {
+  TDBC *pDbc;
+};
+
 SMTbCursor *metaOpenTbCursor(SMeta *pMeta) {
-  // TODO
+  SMTbCursor *pTbCur = NULL;
+  SMetaDB    *pDB = pMeta->pDB;
+
+  pTbCur = (SMTbCursor *)taosMemoryCalloc(1, sizeof(*pTbCur));
+  if (pTbCur == NULL) {
     return NULL;
   }
 
+  tdbDbcOpen(pDB->pTbDB, &pTbCur->pDbc);
+
+  return pTbCur;
+}
+
 void metaCloseTbCursor(SMTbCursor *pTbCur) {
-  // TODO
+  if (pTbCur) {
+    if (pTbCur->pDbc) {
+      tdbDbcClose(pTbCur->pDbc);
+    }
+    taosMemoryFree(pTbCur);
+  }
 }
 
 char *metaTbCursorNext(SMTbCursor *pTbCur) {
-  // TODO
+  void  *pKey = NULL;
+  void  *pVal = NULL;
+  int    kLen;
+  int    vLen;
+  int    ret;
+  void  *pBuf;
+  STbCfg tbCfg;
+
+  for (;;) {
+    ret = tdbDbNext(pTbCur->pDbc, &pKey, &kLen, &pVal, &vLen);
+    if (ret < 0) break;
+    pBuf = pVal;
+    metaDecodeTbInfo(pBuf, &tbCfg);
+    if (tbCfg.type == META_SUPER_TABLE) {
+      taosMemoryFree(tbCfg.name);
+      taosMemoryFree(tbCfg.stbCfg.pTagSchema);
+      continue;
+      ;
+    } else if (tbCfg.type == META_CHILD_TABLE) {
+      kvRowFree(tbCfg.ctbCfg.pTag);
+    }
+
+    return tbCfg.name;
+  }
+
   return NULL;
 }
 
+struct SMCtbCursor {
+  TDBC    *pCur;
+  tb_uid_t suid;
+  void    *pKey;
+  void    *pVal;
+  int      kLen;
+  int      vLen;
+};
+
 SMCtbCursor *metaOpenCtbCursor(SMeta *pMeta, tb_uid_t uid) {
-  // TODO
+  SMCtbCursor *pCtbCur = NULL;
+  SMetaDB     *pDB = pMeta->pDB;
+  int          ret;
+
+  pCtbCur = (SMCtbCursor *)taosMemoryCalloc(1, sizeof(*pCtbCur));
+  if (pCtbCur == NULL) {
     return NULL;
   }
 
+  pCtbCur->suid = uid;
+  ret = tdbDbcOpen(pDB->pCtbIdx, &pCtbCur->pCur);
+  if (ret < 0) {
+    taosMemoryFree(pCtbCur);
+    return NULL;
+  }
+
+  // TODO: move the cursor to the suid there
+
+  return pCtbCur;
+}
+
 void metaCloseCtbCurosr(SMCtbCursor *pCtbCur) {
-  // TODO
+  if (pCtbCur) {
+    if (pCtbCur->pCur) {
+      tdbDbcClose(pCtbCur->pCur);
+
+      TDB_FREE(pCtbCur->pKey);
+      TDB_FREE(pCtbCur->pVal);
+    }
+
+    taosMemoryFree(pCtbCur);
+  }
 }
 
 tb_uid_t metaCtbCursorNext(SMCtbCursor *pCtbCur) {
-  // TODO
+  int         ret;
+  SCtbIdxKey *pCtbIdxKey;
+
+  ret = tdbDbNext(pCtbCur->pCur, &pCtbCur->pKey, &pCtbCur->kLen, &pCtbCur->pVal, &pCtbCur->vLen);
+  if (ret < 0) {
     return 0;
   }
+
+  pCtbIdxKey = pCtbCur->pVal;
+
+  return pCtbIdxKey->uid;
+}
+
+int metaGetTbNum(SMeta *pMeta) {
+  // TODO
+  // ASSERT(0);
+  return 0;
+}
+
+STSmaWrapper *metaGetSmaInfoByTable(SMeta *pMeta, tb_uid_t uid) {
+  // TODO
+  ASSERT(0);
+  return NULL;
+}
+
+int metaRemoveSmaFromDb(SMeta *pMeta, int64_t indexUid) {
+  // TODO
+  ASSERT(0);
+  return 0;
+}
+
+int metaSaveSmaToDB(SMeta *pMeta, STSma *pSmaCfg) {
+  // TODO
+  ASSERT(0);
+  return 0;
+}
+
+STSma *metaGetSmaInfoByIndex(SMeta *pMeta, int64_t indexUid) {
+  // TODO
+  ASSERT(0);
+  return NULL;
+}
+
+const char *metaSmaCursorNext(SMSmaCursor *pCur) {
+  // TODO
+  ASSERT(0);
+  return NULL;
+}
+
+void metaCloseSmaCurosr(SMSmaCursor *pCur) {
+  // TODO
+  ASSERT(0);
+}
+
+SArray *metaGetSmaTbUids(SMeta *pMeta, bool isDup) {
+  // TODO
+  ASSERT(0);
+  return NULL;
+}
+
+SMSmaCursor *metaOpenSmaCursor(SMeta *pMeta, tb_uid_t uid) {
+  // TODO
+  ASSERT(0);
+  return NULL;
+}
+
+static int metaEncodeSchema(void **buf, SSchemaWrapper *pSW) {
+  int      tlen = 0;
+  SSchema *pSchema;
+
+  tlen += taosEncodeFixedU32(buf, pSW->nCols);
+  for (int i = 0; i < pSW->nCols; i++) {
+    pSchema = pSW->pSchema + i;
+    tlen += taosEncodeFixedI8(buf, pSchema->type);
+    tlen += taosEncodeFixedI16(buf, pSchema->colId);
+    tlen += taosEncodeFixedI32(buf, pSchema->bytes);
+    tlen += taosEncodeString(buf, pSchema->name);
+  }
+
+  return tlen;
+}
+
+static void *metaDecodeSchema(void *buf, SSchemaWrapper *pSW) {
+  SSchema *pSchema;
+
+  buf = taosDecodeFixedU32(buf, &pSW->nCols);
+  pSW->pSchema = (SSchema *)taosMemoryMalloc(sizeof(SSchema) * pSW->nCols);
+  for (int i = 0; i < pSW->nCols; i++) {
+    pSchema = pSW->pSchema + i;
+    buf = taosDecodeFixedI8(buf, &pSchema->type);
+    buf = taosDecodeFixedI16(buf, &pSchema->colId);
+    buf = taosDecodeFixedI32(buf, &pSchema->bytes);
+    buf = taosDecodeStringTo(buf, pSchema->name);
+  }
+
+  return buf;
+}
+
+static int metaEncodeTbInfo(void **buf, STbCfg *pTbCfg) {
+  int tsize = 0;
+
+  tsize += taosEncodeString(buf, pTbCfg->name);
+  tsize += taosEncodeFixedU32(buf, pTbCfg->ttl);
+  tsize += taosEncodeFixedU32(buf, pTbCfg->keep);
+  tsize += taosEncodeFixedU8(buf, pTbCfg->info);
+
+  if (pTbCfg->type == META_SUPER_TABLE) {
+    SSchemaWrapper sw = {.nCols = pTbCfg->stbCfg.nTagCols, .pSchema = pTbCfg->stbCfg.pTagSchema};
+    tsize += metaEncodeSchema(buf, &sw);
+  } else if (pTbCfg->type == META_CHILD_TABLE) {
+    tsize += taosEncodeFixedU64(buf, pTbCfg->ctbCfg.suid);
+    tsize += tdEncodeKVRow(buf, pTbCfg->ctbCfg.pTag);
+  } else if (pTbCfg->type == META_NORMAL_TABLE) {
+    // TODO
+  } else {
+    ASSERT(0);
+  }
+
+  return tsize;
+}
+
+static void *metaDecodeTbInfo(void *buf, STbCfg *pTbCfg) {
+  buf = taosDecodeString(buf, &(pTbCfg->name));
+  buf = taosDecodeFixedU32(buf, &(pTbCfg->ttl));
+  buf = taosDecodeFixedU32(buf, &(pTbCfg->keep));
+  buf = taosDecodeFixedU8(buf, &(pTbCfg->info));
+
+  if (pTbCfg->type == META_SUPER_TABLE) {
+    SSchemaWrapper sw;
+    buf = metaDecodeSchema(buf, &sw);
+    pTbCfg->stbCfg.nTagCols = sw.nCols;
+    pTbCfg->stbCfg.pTagSchema = sw.pSchema;
+  } else if (pTbCfg->type == META_CHILD_TABLE) {
+    buf = taosDecodeFixedU64(buf, &(pTbCfg->ctbCfg.suid));
+    buf = tdDecodeKVRow(buf, &(pTbCfg->ctbCfg.pTag));
+  } else if (pTbCfg->type == META_NORMAL_TABLE) {
+    // TODO
+  } else {
+    ASSERT(0);
+  }
+  return buf;
+}

source/libs/tdb/CMakeLists.txt

@@ -9,9 +9,8 @@ target_sources(tdb
     "src/db/tdbDb.c"
     "src/db/tdbEnv.c"
     "src/db/tdbTxn.c"
+    "src/db/tdbPage.c"
     "src/db/tdbOs.c"
-    "src/page/tdbPage.c"
-    "src/page/tdbPageL.c"
 )
 
 target_include_directories(

source/libs/tdb/src/db/tdbBtree.c

@@ -30,7 +30,6 @@ struct SBTree {
   int            valLen;
   SPager        *pPager;
   FKeyComparator kcmpr;
-  u8             fanout;
   int            pageSize;
   int            maxLocal;
   int            minLocal;
@@ -85,6 +84,8 @@ int tdbBtreeOpen(int keyLen, int valLen, SPager *pPager, FKeyComparator kcmpr, S
   SBTree *pBt;
   int     ret;
 
+  ASSERT(keyLen != 0);
+
   *ppBt = NULL;
 
   pBt = (SBTree *)tdbOsCalloc(1, sizeof(*pBt));
@@ -100,21 +101,14 @@ int tdbBtreeOpen(int keyLen, int valLen, SPager *pPager, FKeyComparator kcmpr, S
   pBt->pPager = pPager;
   // pBt->kcmpr
   pBt->kcmpr = kcmpr ? kcmpr : tdbDefaultKeyCmprFn;
-  // pBt->fanout
-  if (keyLen == TDB_VARIANT_LEN) {
-    pBt->fanout = TDB_DEFAULT_FANOUT;
-  } else {
-    ASSERT(0);
-    // TODO: pBt->fanout = 0;
-  }
   // pBt->pageSize
   pBt->pageSize = tdbPagerGetPageSize(pPager);
   // pBt->maxLocal
-  pBt->maxLocal = (pBt->pageSize - 14) / pBt->fanout;
+  pBt->maxLocal = tdbPageCapacity(pBt->pageSize, sizeof(SIntHdr)) / 4;
   // pBt->minLocal: Should not be allowed smaller than 15, which is [nPayload][nKey][nData]
-  pBt->minLocal = (pBt->pageSize - 14) / pBt->fanout / 2;
+  pBt->minLocal = pBt->maxLocal / 2;
   // pBt->maxLeaf
-  pBt->maxLeaf = pBt->pageSize - 14;
+  pBt->maxLeaf = tdbPageCapacity(pBt->pageSize, sizeof(SLeafHdr));
   // pBt->minLeaf
   pBt->minLeaf = pBt->minLocal;
 
@@ -226,6 +220,43 @@ int tdbBtreeGet(SBTree *pBt, const void *pKey, int kLen, void **ppVal, int *vLen
   return 0;
 }
 
+int tdbBtreePGet(SBTree *pBt, const void *pKey, int kLen, void **ppKey, int *pkLen, void **ppVal, int *vLen) {
+  SBTC         btc;
+  SCell       *pCell;
+  int          cret;
+  void        *pTKey;
+  void        *pTVal;
+  SCellDecoder cd;
+
+  tdbBtcOpen(&btc, pBt);
+
+  tdbBtCursorMoveTo(&btc, pKey, kLen, &cret);
+  if (cret) {
+    return cret;
+  }
+
+  pCell = tdbPageGetCell(btc.pPage, btc.idx);
+  tdbBtreeDecodeCell(btc.pPage, pCell, &cd);
+
+  pTKey = TDB_REALLOC(*ppKey, cd.kLen);
+  pTVal = TDB_REALLOC(*ppVal, cd.vLen);
+
+  if (pTKey == NULL || pTVal == NULL) {
+    TDB_FREE(pTKey);
+    TDB_FREE(pTVal);
+  }
+
+  *ppKey = pTKey;
+  *ppVal = pTVal;
+  *pkLen = cd.kLen;
+  *vLen = cd.vLen;
+
+  memcpy(*ppKey, cd.pKey, cd.kLen);
+  memcpy(*ppVal, cd.pVal, cd.vLen);
+
+  return 0;
+}
+
 static int tdbBtCursorMoveTo(SBTC *pBtc, const void *pKey, int kLen, int *pCRst) {
   int     ret;
   SBTree *pBt;
@@ -929,7 +960,7 @@ static int tdbBtreeEncodeCell(SPage *pPage, const void *pKey, int kLen, const vo
   }
 
   // 2. Encode payload part
-  if (leaf) {
+  if (leaf && vLen > 0) {
     ret = tdbBtreeEncodePayload(pPage, pCell + nHeader, pKey, kLen, pVal, vLen, &nPayload);
   } else {
     ret = tdbBtreeEncodePayload(pPage, pCell + nHeader, pKey, kLen, NULL, 0, &nPayload);
@@ -968,6 +999,7 @@ static int tdbBtreeDecodePayload(SPage *pPage, const u8 *pPayload, SCellDecoder
   return 0;
 }
 
+// TODO: here has problem
 static int tdbBtreeDecodeCell(SPage *pPage, const SCell *pCell, SCellDecoder *pDecoder) {
   u8  flags;
   u8  leaf;

source/libs/tdb/src/db/tdbDb.c

@@ -96,6 +96,10 @@ int tdbDbGet(TDB *pDb, const void *pKey, int kLen, void **ppVal, int *vLen) {
   return tdbBtreeGet(pDb->pBt, pKey, kLen, ppVal, vLen);
 }
 
+int tdbDbPGet(TDB *pDb, const void *pKey, int kLen, void **ppKey, int *pkLen, void **ppVal, int *vLen) {
+  return tdbBtreePGet(pDb->pBt, pKey, kLen, ppKey, pkLen, ppVal, vLen);
+}
+
 int tdbDbcOpen(TDB *pDb, TDBC **ppDbc) {
   int   ret;
   TDBC *pDbc = NULL;
@@ -131,3 +135,9 @@ int tdbDbcClose(TDBC *pDbc) {
 
   return 0;
 }
+
+int tdbDbcInsert(TDBC *pDbc, const void *pKey, int keyLen, const void *pVal, int valLen) {
+  // TODO
+  ASSERT(0);
+  return 0;
+}

source/libs/tdb/src/db/tdbPCache.c

@@ -122,7 +122,7 @@ static void tdbPCacheClearLock(SPCache *pCache) { tdbMutexDestroy(&(pCache->mute
 
 static void tdbPCacheLock(SPCache *pCache) { tdbMutexLock(&(pCache->mutex)); }
 
-static void tdbPCacheUnlock(SPCache *pCache) { tdbMutexDestroy(&(pCache->mutex)); }
+static void tdbPCacheUnlock(SPCache *pCache) { tdbMutexUnlock(&(pCache->mutex)); }
 
 static bool tdbPCacheLocked(SPCache *pCache) {
   assert(0);

source/libs/tdb/src/db/tdbPage.c

@@ -242,6 +242,18 @@ void tdbPageCopy(SPage *pFromPage, SPage *pToPage) {
   pToPage->nOverflow = pFromPage->nOverflow;
 }
 
+int tdbPageCapacity(int pageSize, int amHdrSize) {
+  int szPageHdr;
+
+  if (pageSize < 65536) {
+    szPageHdr = pageMethods.szPageHdr;
+  } else {
+    szPageHdr = pageLargeMethods.szPageHdr;
+  }
+
+  return pageSize - szPageHdr - amHdrSize;
+}
+
 static int tdbPageAllocate(SPage *pPage, int szCell, SCell **ppCell) {
   SCell *pFreeCell;
   u8    *pOffset;
@@ -504,3 +516,80 @@ SPageMethods pageMethods = {
     getPageFreeCellInfo,  // getFreeCellInfo
     setPageFreeCellInfo   // setFreeCellInfo
 };
+
+typedef struct __attribute__((__packed__)) {
+  u8 cellNum[3];
+  u8 cellBody[3];
+  u8 cellFree[3];
+  u8 nFree[3];
+} SPageHdrL;
+
+typedef struct __attribute__((__packed__)) {
+  u8 szCell[3];
+  u8 nxOffset[3];
+} SFreeCellL;
+
+// cellNum
+static inline int  getLPageCellNum(SPage *pPage) { return TDB_GET_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellNum); }
+static inline void setLPageCellNum(SPage *pPage, int cellNum) {
+  TDB_PUT_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellNum, cellNum);
+}
+
+// cellBody
+static inline int  getLPageCellBody(SPage *pPage) { return TDB_GET_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellBody); }
+static inline void setLPageCellBody(SPage *pPage, int cellBody) {
+  TDB_PUT_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellBody, cellBody);
+}
+
+// cellFree
+static inline int  getLPageCellFree(SPage *pPage) { return TDB_GET_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellFree); }
+static inline void setLPageCellFree(SPage *pPage, int cellFree) {
+  TDB_PUT_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellFree, cellFree);
+}
+
+// nFree
+static inline int  getLPageNFree(SPage *pPage) { return TDB_GET_U24(((SPageHdrL *)(pPage->pPageHdr))[0].nFree); }
+static inline void setLPageNFree(SPage *pPage, int nFree) {
+  TDB_PUT_U24(((SPageHdrL *)(pPage->pPageHdr))[0].nFree, nFree);
+}
+
+// cell offset
+static inline int getLPageCellOffset(SPage *pPage, int idx) {
+  ASSERT(idx >= 0 && idx < getPageCellNum(pPage));
+  return TDB_GET_U24(pPage->pCellIdx + 3 * idx);
+}
+
+static inline void setLPageCellOffset(SPage *pPage, int idx, int offset) {
+  TDB_PUT_U24(pPage->pCellIdx + 3 * idx, offset);
+}
+
+// free cell info
+static inline void getLPageFreeCellInfo(SCell *pCell, int *szCell, int *nxOffset) {
+  SFreeCellL *pFreeCell = (SFreeCellL *)pCell;
+  *szCell = TDB_GET_U24(pFreeCell->szCell);
+  *nxOffset = TDB_GET_U24(pFreeCell->nxOffset);
+}
+
+static inline void setLPageFreeCellInfo(SCell *pCell, int szCell, int nxOffset) {
+  SFreeCellL *pFreeCell = (SFreeCellL *)pCell;
+  TDB_PUT_U24(pFreeCell->szCell, szCell);
+  TDB_PUT_U24(pFreeCell->nxOffset, nxOffset);
+}
+
+SPageMethods pageLargeMethods = {
+    3,                     // szOffset
+    sizeof(SPageHdrL),     // szPageHdr
+    sizeof(SFreeCellL),    // szFreeCell
+    getLPageCellNum,       // getCellNum
+    setLPageCellNum,       // setCellNum
+    getLPageCellBody,      // getCellBody
+    setLPageCellBody,      // setCellBody
+    getLPageCellFree,      // getCellFree
+    setLPageCellFree,      // setCellFree
+    getLPageNFree,         // getFreeBytes
+    setLPageNFree,         // setFreeBytes
+    getLPageCellOffset,    // getCellOffset
+    setLPageCellOffset,    // setCellOffset
+    getLPageFreeCellInfo,  // getFreeCellInfo
+    setLPageFreeCellInfo   // setFreeCellInfo
+};

source/libs/tdb/src/inc/tdbBtree.h

@@ -43,6 +43,7 @@ int tdbBtreeOpen(int keyLen, int valLen, SPager *pFile, FKeyComparator kcmpr, SB
 int tdbBtreeClose(SBTree *pBt);
 int tdbBtCursorInsert(SBTC *pCur, const void *pKey, int kLen, const void *pVal, int vLen);
 int tdbBtreeGet(SBTree *pBt, const void *pKey, int kLen, void **ppVal, int *vLen);
+int tdbBtreePGet(SBTree *pBt, const void *pKey, int kLen, void **ppKey, int *pkLen, void **ppVal, int *vLen);
 
 // SBTC
 int tdbBtcOpen(SBTC *pCur, SBTree *pBt);

source/libs/tdb/src/inc/tdbDb.h

@@ -29,11 +29,13 @@ int tdbDbClose(TDB *pDb);
 int tdbDbDrop(TDB *pDb);
 int tdbDbInsert(TDB *pDb, const void *pKey, int keyLen, const void *pVal, int valLen);
 int tdbDbGet(TDB *pDb, const void *pKey, int kLen, void **ppVal, int *vLen);
+int tdbDbPGet(TDB *pDb, const void *pKey, int kLen, void **ppKey, int *pkLen, void **ppVal, int *vLen);
 
 // TDBC
 int tdbDbcOpen(TDB *pDb, TDBC **ppDbc);
 int tdbDbNext(TDBC *pDbc, void **ppKey, int *kLen, void **ppVal, int *vLen);
 int tdbDbcClose(TDBC *pDbc);
+int tdbDbcInsert(TDBC *pDbc, const void *pKey, int keyLen, const void *pVal, int valLen);
 
 #ifdef __cplusplus
 }

source/libs/tdb/src/inc/tdbPage.h

@@ -111,6 +111,7 @@ void tdbPageInit(SPage *pPage, u8 szAmHdr, int (*xCellSize)(const SPage *, SCell
 int  tdbPageInsertCell(SPage *pPage, int idx, SCell *pCell, int szCell, u8 asOvfl);
 int  tdbPageDropCell(SPage *pPage, int idx);
 void tdbPageCopy(SPage *pFromPage, SPage *pToPage);
+int  tdbPageCapacity(int pageSize, int amHdrSize);
 
 static inline SCell *tdbPageGetCell(SPage *pPage, int idx) {
   SCell *pCell;

source/libs/tdb/src/inc/tdbTxn.h

@@ -20,7 +20,7 @@
 extern "C" {
 #endif
 
-typedef struct STxn STXN;
+typedef struct STxn TXN;
 
 struct STxn {
   u64 txnId;

source/libs/tdb/src/page/tdbPageL.c

@@ -1,93 +0,0 @@
-/*
- * 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 "tdbInt.h"
-
-typedef struct __attribute__((__packed__)) {
-  u8 cellNum[3];
-  u8 cellBody[3];
-  u8 cellFree[3];
-  u8 nFree[3];
-} SPageHdrL;
-
-typedef struct __attribute__((__packed__)) {
-  u8 szCell[3];
-  u8 nxOffset[3];
-} SFreeCellL;
-
-// cellNum
-static inline int  getPageCellNum(SPage *pPage) { return TDB_GET_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellNum); }
-static inline void setPageCellNum(SPage *pPage, int cellNum) {
-  TDB_PUT_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellNum, cellNum);
-}
-
-// cellBody
-static inline int  getPageCellBody(SPage *pPage) { return TDB_GET_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellBody); }
-static inline void setPageCellBody(SPage *pPage, int cellBody) {
-  TDB_PUT_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellBody, cellBody);
-}
-
-// cellFree
-static inline int  getPageCellFree(SPage *pPage) { return TDB_GET_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellFree); }
-static inline void setPageCellFree(SPage *pPage, int cellFree) {
-  TDB_PUT_U24(((SPageHdrL *)(pPage->pPageHdr))[0].cellFree, cellFree);
-}
-
-// nFree
-static inline int  getPageNFree(SPage *pPage) { return TDB_GET_U24(((SPageHdrL *)(pPage->pPageHdr))[0].nFree); }
-static inline void setPageNFree(SPage *pPage, int nFree) {
-  TDB_PUT_U24(((SPageHdrL *)(pPage->pPageHdr))[0].nFree, nFree);
-}
-
-// cell offset
-static inline int getPageCellOffset(SPage *pPage, int idx) {
-  ASSERT(idx >= 0 && idx < getPageCellNum(pPage));
-  return TDB_GET_U24(pPage->pCellIdx + 3 * idx);
-}
-
-static inline void setPageCellOffset(SPage *pPage, int idx, int offset) {
-  TDB_PUT_U24(pPage->pCellIdx + 3 * idx, offset);
-}
-
-// free cell info
-static inline void getPageFreeCellInfo(SCell *pCell, int *szCell, int *nxOffset) {
-  SFreeCellL *pFreeCell = (SFreeCellL *)pCell;
-  *szCell = TDB_GET_U24(pFreeCell->szCell);
-  *nxOffset = TDB_GET_U24(pFreeCell->nxOffset);
-}
-
-static inline void setPageFreeCellInfo(SCell *pCell, int szCell, int nxOffset) {
-  SFreeCellL *pFreeCell = (SFreeCellL *)pCell;
-  TDB_PUT_U24(pFreeCell->szCell, szCell);
-  TDB_PUT_U24(pFreeCell->nxOffset, nxOffset);
-}
-
-SPageMethods pageLargeMethods = {
-    3,                    // szOffset
-    sizeof(SPageHdrL),    // szPageHdr
-    sizeof(SFreeCellL),   // szFreeCell
-    getPageCellNum,       // getCellNum
-    setPageCellNum,       // setCellNum
-    getPageCellBody,      // getCellBody
-    setPageCellBody,      // setCellBody
-    getPageCellFree,      // getCellFree
-    setPageCellFree,      // setCellFree
-    getPageNFree,         // getFreeBytes
-    setPageNFree,         // setFreeBytes
-    getPageCellOffset,    // getCellOffset
-    setPageCellOffset,    // setCellOffset
-    getPageFreeCellInfo,  // getFreeCellInfo
-    setPageFreeCellInfo   // setFreeCellInfo
-};