[IMPALA-9071] When metastore.warehouse.dir != metastore.warehouse.external.dir, Impala writes to the wrong location for external tables

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: Impala 3.4.0
• Fix Version/s: Impala 3.4.0
• Component/s: Frontend
• Labels: broken-build
• Epic Color: ghx-label-12
• Assignee: Quanlong Huang
• Reporter: Joe McDonnell
• Created: 19/Oct/19 20:10
• Updated: 14/Apr/20 23:01
• Resolved: 08/Nov/19 07:53

Description

Hive introduced a translation layer that can convert a normal table to an external table. When doing so without a specified location, the translated external table uses metastore.warehouse.external.dir as the location rather than metastore.warehouse.dir. Impala does not know about this distinction, so it writes to the location it thinks the table should be (under metastore.warehouse.dir). This means I can do the following:

[localhost:21000] joetest> select count(*) from functional.alltypes;
Query: select count(*) from functional.alltypes
Query submitted at: 2019-10-19 13:08:24 (Coordinator: http://joemcdonnell:25000)
Query progress can be monitored at: http://joemcdonnell:25000/query_plan?query_id=68434b05e2badd50:a18a2e3000000000
+----------+
| count(*) |
+----------+
| 7300     |
+----------+
Fetched 1 row(s) in 0.14s
[localhost:21000] joetest> create table testtable as select * from functional.alltypes;
Query: create table testtable as select * from functional.alltypes
Query submitted at: 2019-10-19 13:08:36 (Coordinator: http://joemcdonnell:25000)
Query progress can be monitored at: http://joemcdonnell:25000/query_plan?query_id=794b92fb68f36ab0:910d036400000000
+----------------------+
| summary              |
+----------------------+
| Inserted 7300 row(s) |
+----------------------+
Fetched 1 row(s) in 0.50s
[localhost:21000] joetest> select count(*) from testtable;
Query: select count(*) from testtable
Query submitted at: 2019-10-19 13:08:43 (Coordinator: http://joemcdonnell:25000)
Query progress can be monitored at: http://joemcdonnell:25000/query_plan?query_id=66423abf016e65af:8362460900000000
+----------+
| count(*) |
+----------+
| 0        |
+----------+
Fetched 1 row(s) in 0.13s

We inserted 7300 rows, but we can't select them back because they were written to the wrong location.