[IMPALA-5933] Compute incremental stats should always return a result set

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 2.6.0, Impala 2.7.0, Impala 2.8.0, Impala 2.9.0, Impala 2.10.0
• Fix Version/s: Impala 3.0
• Component/s: Backend, Frontend
• Epic Color: ghx-label-5
• Assignee: Zoltán Borók-Nagy
• Reporter: Alexander Behm
• Created: 13/Sep/17 23:08
• Updated: 30/Oct/18 21:14
• Resolved: 30/Oct/18 21:14

Description

Compute incremental stats should always return a result set that states which stats were computed/modified. Today, we have a shortcut in the code for compute incremental stats that does not return a result set when no new stats are computed because all partitions already have incremental stats.

The fact that the same command sometimes returns a result set and sometimes not depending on the state of a table is strange, and can confuse clients like JDBC/ODBC that might reasonably expect a result set for that statement.

The issue can be reproduced by running compute incremental stats twice in a row on the same table. The second run does not return a result set.

The culprit is in client-request-state.cc ClientRequestState::WaitInternal():

...
  if (catalog_op_type() == TCatalogOpType::DDL &&
      ddl_type() == TDdlType::COMPUTE_STATS && child_queries.size() > 0) {
    RETURN_IF_ERROR(UpdateTableAndColumnStats(child_queries));
  }
...

For a no-op incremental stats the number of child queries is 0, so we never set a result set or the result set metadata.

Issue Links

• relates to: IMPALA-5903 Inconsistent specification of result set and result set metadata