If your Databricks data contains non-display characters Line Feeds (LF or '0A'x) or Carriage Returns (CR '0D'x), BULKUNLOAD from Databricks might result in one or more of the following errors...

ERROR: Invalid data format in BULKUNLOAD stream. If the data contains the delimiter character you can use BL_USE_ESCAPE=YES to fix this error.

ERROR: read_and_convert_data_file(1): get_column_data failed, rc = 0xfffffc1a

ERROR: Message from TK: An invalid number was found in buffer.

ERROR: read_and_convert_data_file(1): tkzsu8ToDouble failed, rc = 0x803fe810

In addition, the following might occur:

• More rows are returned than expected.
• Wrong values occur in the result set.
• The position of the problem-column in the result set might affect the results that are returned.
• BL_USE_ESCAPE does not address the problem.

Workaround

To work around this issue, do any of the following:

• Do not include the columns that have values that contain the '0A'x and/or '0D'x non-display characters.
• Clean the source data to remove the embedded non-display characters.
• Create a Databricks view of the Databricks table that transforms those non-display characters to a character. Here is an example: 
  
  proc sql ;     
     connect to spark (connection-options) ;
     execute by spark (CREATE OR REPLACE VIEW schema.view-name AS
                        SELECT other-columns,
                               CASE                          
                                 WHEN problem-column RLIKE '[\r\n]' THEN regexp_replace(problem-column, '[\r\n]+', ' ')
                                 ELSE problem-column
                               END AS problem-column,
                               other-columns
                          FROM schema.table-name; ) ;
   quit ;  
  

Resolution

With the hot fix for this issue, you will be able to use undocumented parameters to BL_OPTIONS. (That is, BL_OPTIONS='BL_ROW_DELIMITER=CR'.)