Databricks

kilaki · ‎01-24-2023

Noticed with a query based on inline select and joins fails to the client with 'Error occurred while deserializing arrow data' I.e the query succeeds on Databricks but client (DBeaver, AtScale) receives an error

The error is only noticed with Databricks SQL setup with preview as channel mode

The same query works fine with Databricks SQL, that has channel setup to current

The issue can be replicated with a query against samples.tpch.nation table

SELECT
  SUM(`t_36`.`noq_empty`) `ne36`,
  `t_36`.`nname` `nname`,
  SUM(`t_94`.`c0`) `c94`
FROM
  (
    SELECT  `t_59`.`noq_empty_gbakc13` `noq_empty`,
      `t_59`.`n_gbakc10` `nname`
    FROM
      (
        SELECT SUM(ntn1.n_nationkey) `noq_empty_gbakc13`,
         ntn1.n_name `n_gbakc10`
        FROM
          samples.tpch.nation ntn1
        WHERE
          true
        GROUP BY
          `n_gbakc10`
      ) `t_59`
  ) `t_36`
  JOIN (
    SELECT
      `t_135`.`c0` `c0`,
      `t_135`.`nname` `nname`
    FROM
      (
        SELECT
          `t_134`.`ci` `c0`,
          `t_134`.`nname` `nname`
        FROM
          (
            SELECT
              `t_133`.`ci_gbakc2` `ci`,
              `t_133`.`sono_gbakc3` `nname`
            FROM
              ( SELECT
                  SUM(ntn10.n_nationkey) `ci_gbakc2`,
                  ntn10.n_name  `sono_gbakc3`
                FROM
                  samples.tpch.nation ntn10
                WHERE
                  true
                GROUP BY
                  `sono_gbakc3`
                HAVING
                  SUM(`ntn10`.`n_nationkey`) > 1
              ) `t_133`
          ) `t_134`
        WHERE
          `t_134`.`ci` > 1
      ) `t_135`
    WHERE
      `t_135`.`c0` > 1
  ) `t_94` ON `t_94`.`nname` = `t_36`.`nname`
GROUP BY
  `t_36`.`nname`
HAVING
  SUM(`t_94`.`c0`) > 10

So would like to understand is there going to be a change in how the result-set object is being serialized in the upcoming releases of DBSQL?

OR Is this a bug?