Stefania commented on CASSANDRA-12794:

The idea of CASSANDRA-7792 is that we set NULL to something other than the 
empty string, something that is not present in the csv, e.g. {{NULL='\-'}}. The 
problem is that if we do this in this example, then it fails to convert the 
empty strings of the missing {{bigint}} values when passing them to prepared 
statements. We could bypass this by disabling prepared statements with 
{{PREPAREDSTATEMENTS = False}}, but then it throws another parsing error. The 
only workaround at the moment is to insert zeros for the missing {{bigint}} 
values, and then it can be imported with something like {{NULL='-'}}. You can 
also copy only the columns that are present into the csv by specifying the 
column names in both the export and import commands.

This specific example with empty strings in composite partition keys of type 
text, and empty {{bigint}} fields is not something that we haven't tested so 
far, and we will need a patch for it. 

> COPY FROM with NULL='' fails when inserting empty row in primary key 
> ---------------------------------------------------------------------
>                 Key: CASSANDRA-12794
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-12794
>             Project: Cassandra
>          Issue Type: Bug
>          Components: CQL
>         Environment: Tested using C* 2.1.15
>            Reporter: Sucwinder Bassi
> Using this table:
> CREATE TABLE testtab (  a_id text,  b_id text,  c_id text,  d_id text,  
> order_id uuid,  acc_id bigint,  bucket bigint,  r_id text,  ts bigint,  
> PRIMARY KEY ((a_id, b_id, c_id, d_id), order_id));
> insert one row:
> INSERT INTO testtab (a_id, b_id , c_id , d_id , order_id, r_id ) VALUES ( '', 
> '', '', 'a1', 645e7d3c-aef7-4e3c-b834-24b792cf2e55, 'r1');
> Use COPY to dump the row to temp.csv:
> copy testtab TO 'temp.csv';
> Which creates this file:
> $ cat temp.csv 
> ,,,a1,645e7d3c-aef7-4e3c-b834-24b792cf2e55,,,r1,
> Truncate the testtab table and then use copy from with NULL='' to insert the 
> row:
> cqlsh:sbkeyspace> COPY testtab FROM 'temp.csv' with NULL='';
> Using 1 child processes
> Starting copy of sbkeyspace.testtab with columns ['a_id', 'b_id', 'c_id', 
> 'd_id', 'order_id', 'acc_id', 'bucket', 'r_id', 'ts'].
> Failed to import 1 rows: ParseError - Cannot insert null value for primary 
> key column 'a_id'. If you want to insert empty strings, consider using the 
> WITH NULL=<marker> option for COPY.,  given up without retries
> Failed to process 1 rows; failed rows written to import_sbkeyspace_testtab.err
> Processed: 1 rows; Rate:       2 rows/s; Avg. rate:       3 rows/s
> 1 rows imported from 1 files in 0.398 seconds (0 skipped).
> It shows 1 rows inserted, but the table is empty:
> select * from testtab ;
>  a_id | b_id | c_id | d_id | order_id | acc_id | bucket | r_id | ts
> ------+------+------+------+----------+--------+--------+------+----
> (0 rows)
> The same error is returned even without the with NULL=''. Is it actually 
> possible for copy from to insert an empty row into the primary key? The 
> insert command shown above inserts the empty row for the primary key without 
> any problems.
> Is this related to https://issues.apache.org/jira/browse/CASSANDRA-7792?

This message was sent by Atlassian JIRA

Reply via email to