[KUDU-1300] INSERT...SELECT from Kudu table into partitioned Kudu table gives error for BOOLEAN column

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Component/s: impala
• Assignee: David Alves
• Reporter: John Russell
• Created: 19/Jan/16 20:03
• Updated: 04/Mar/16 10:08
• Resolved: 04/Mar/16 10:08

Description

This bug is a follow-on from KUDU-1299.

Although I could INSERT...SELECT from my Impala table into an unpartitioned Kudu table (after doing away with a TIMESTAMP column that caused an error), when I tried the exact same operation where the destination was a partitioned Kudu table, then I got another conversion error saying the Impala BOOLEAN type wasn't supported. Again, the Kudu docs say the BOOLEAN type should work:

create TABLE log_ingest_docs_kudu2
(
id bigint
, ip string
, f2 string
, f3 string
, the_date string
, method string
, path string
, status smallint
, size bigint
, referer string
, agent string
, is_search_term boolean
, search_term string
, is_doc_page boolean
, doc_page string) DISTRIBUTE BY HASH (id) INTO 10 BUCKETS
TBLPROPERTIES
(
'kudu.master_addresses'='yadayada:7051'
, 'kudu.key_columns'='id'
, 'kudu.table_name'='log_ingest_docs_kudu2'
, 'storage_handler'='com.cloudera.kudu.hive.KuduStorageHandler'
);

insert into log_ingest_docs_kudu2 select * from log_ingest_docs_kudu;
WARNINGS:
Impala type BOOLEAN is not available in Kudu.



Impala type BOOLEAN is not available in Kudu.



select count(*) from log_ingest_docs_kudu2;
+----------+
| count(*) |
+----------+
| 0        |
+----------+

select distinct is_search_term, is_doc_page from log_ingest_docs_kudu;
WARNINGS:
Impala type BOOLEAN is not available in Kudu.



Impala type BOOLEAN is not available in Kudu.

Notice a couple of things:

• In this case, the source table is another Kudu table, not the original Impala table. (How does Kudu know at that point that the BOOLEAN data inside a Kudu table is the Impala BOOLEAN type?)
• The only difference between table log_ingest_docs_kudu (where the INSERT...SELECT worked) and log_ingest_docs_kudu2 (where it failed) is the DISTRIBUTE BY HASH added to the latter.
• The BOOLEAN message is stated to be a warning, yet it caused the operation to fail, since no rows are inserted into the destination table.
• If I go back to log_ingest_docs_kudu where the data (including the BOOLEAN column) was successfully transferred, I find that I can't do any query that references the BOOLEAN column. So the problem manifests itself consistently in the SELECT statement, and inconsistently in the INSERT statement (depends on whether the table has a DISTRIBUTE BY clause or not).