This copies metaData into pRequest->parseMeta unconditionally, but request teardown (doDestroyRequest) does not free parseMeta via catalogFreeMetaData. This introduces a persistent leak for every stmt2 bind/parse. Either only store parseMeta when parseOnly is enabled (matching doAsyncQueryFromAnalyse), or ensure parseMeta is freed during request destruction/reset before overwriting it.

catalogGetAllMeta no longer initializes/clears *pRsp. If catalogAsyncGetAllMeta returns an error (or tsem_init fails), the function returns without touching pRsp, leaving any existing pointers/values in caller memory intact and risking later misuse/free. Consider zeroing *pRsp at entry (or documenting/guaranteeing that callers must pass a zeroed SMetaData).

This loop allocates/initializes a TAOS_STMT2 but never closes it. In the rest of this test file, statements are consistently released via taos_stmt2_close; skipping it here will leak resources and can destabilize later tests. Ensure each stmt created in the loop is closed (and that any associated result state is released) before the next iteration/end of test.

The async test path is currently unreachable: the loop starts at i=1, but the wait is guarded by if (i == 0). As written, the semaphore/AsyncArgs setup is unused and the async option isn’t exercised. Either iterate over both options (0 and 1) and wait in the async case, or remove the async scaffolding if only the sync path is intended.

Same issue as above: the loop starts at i=1 so the async option (i==0) is never executed, and the semaphore wait block is dead code. This reduces test coverage of the async stmt2 execution path and leaves unnecessary synchronization setup in the test.

Removing the fallback handling here means translate will now hard-fail when VStbRefDbs is missing from the meta cache (getVStbRefDbsFromCache returns TSDB_CODE_PAR_TABLE_NOT_EXIST when pMetaCache/pVStbRefDbs is NULL). This breaks sync parse flows that do not populate meta cache (e.g., qParseSql/parseSqlIntoAst passes NULL meta cache) and stmt1 query parsing (qStmtParseQuerySql called with NULL metadata). Either restore the previous fallback to allow an empty vgroup list, or add a synchronous catalog fetch path for VStbRefDbs when it’s not present in cache.

On failure, this cleanup only destroys pMetaData->pVStbRefDbs, but catalogGetAllMeta can populate many other fields in SMetaData (pDbVgroup, pTableMeta, pUser, etc.). This will leak memory (and possibly leave partially-initialized pointers) on error. Prefer calling catalogFreeMetaData(pMetaData) (or the appropriate catalog cleanup API) for complete cleanup.

The parameter binding for the prepared statement is incorrect. The query ... where dataname in (?,?) ... has two parameter markers, but the TAOS_STMT2_BINDV is configured to bind only one parameter (num_params is 1). This means the test doesn't correctly cover the scenario with multiple parameters in an IN clause.

    TAOS_STMT2_BIND* param_ptrs[] = {&params[0], &params[1]};
    TAOS_STMT2_BINDV bindv = {2, NULL, NULL, param_ptrs};
    code = taos_stmt2_bind_param(stmt, &bindv, -1);

Similar to the previous test case, the parameter binding here is incorrect. The query uses two parameter markers (?,?), but the binding is set up for only a single parameter. This should be corrected to bind both parameters.

      TAOS_STMT2_BIND* param_ptrs[] = {&params[0], &params[1]};
      TAOS_STMT2_BINDV bindv = {2, NULL, NULL, param_ptrs};
      code = taos_stmt2_bind_param(stmt, &bindv, -1);

The parameter binding is incorrect for this test case as well. The query has two parameter markers, but the binding is configured for only one. Please adjust the TAOS_STMT2_BINDV to bind both parameters.

      TAOS_STMT2_BIND* param_ptrs[] = {&params[0], &params[1]};
      TAOS_STMT2_BINDV bindv = {2, NULL, NULL, param_ptrs};
      code = taos_stmt2_bind_param(stmt, &bindv, -1);

This cleanup logic is duplicated in stmtBindBatch2 at lines 2031-2035. To improve maintainability and reduce code duplication, consider extracting this logic into a static helper function.

The function collectMetaKey is also declared in the public header include/libs/parser/parser.h. This declaration in an internal header is redundant and should be removed to avoid potential inconsistencies and improve maintainability.