[IMPALA-10585] retry_failed_queries=true should not apply to DMLs

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 4.0.0
• Component/s: Backend
• Epic Color: ghx-label-5
• Assignee: Csaba Ringhofer
• Reporter: Joe McDonnell
• Created: 15/Mar/21 20:14
• Updated: 10/Mar/24 16:22
• Resolved: 10/Mar/24 16:22

Description

I noticed that retry_failed_queries=true will retry insert statements:

[localhost:21050] joetest> insert into retrytest select count(*) from functional.alltypes where bool_col = sleep(50);
Query: insert into retrytest select count(*) from functional.alltypes where bool_col = sleep(50)
Query submitted at: 2021-03-15 10:23:32 (Coordinator: http://joemcdonnell:25000)
Query progress can be monitored at: http://joemcdonnell:25000/query_plan?query_id=5f4b8c0224faa31a:4a585cf700000000
...
Failed due to unreachable impalad(s): joemcdonnell:27002                                                        00000000
Retried query link: http://joemcdonnell:25000/query_plan?query_id=824b6b103ea68ea3:bc804b4Failed due to unreachable impalad(s): joemcdonnell:27002
...
Query has been retried using query id: 824b6b103ea68ea3:bc804b4500000000                  Retried query link: http://joemcdonnell:25000/query_plan?query_id=824b6b103ea68ea3:bc804b4500000000
Modified 1 row(s) in 47.71s

I don't think this was intended to work, because https://issues.apache.org/jira/browse/IMPALA-9734 was closed saying that we don't do retries for write statements. There also aren't any tests for these cases.

I think we intended to exempt DML statements from retry_failed_queries=true. We should implement that and add tests to make sure DMLs don't get retried.

Issue Links

• relates to: IMPALA-9124 Transparently retry queries that fail due to cluster membership changes