[
https://issues.apache.org/jira/browse/IMPALA-13772?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jason Fehr updated IMPALA-13772:
--------------------------------
Description:
The insert into sys.impala_query_log DML that workload management executes is
cancelled after waiting for 10 seconds. For example, in the case where
admission control queues the query because the request pool is full or
executors are not available.
The short cancel timeframe is caused by the "FETCH_ROWS_TIMEOUT_MS" query
option not being set. Thus, the default of 10 seconds is used. With other
queries, the fetch rows timeout only starts after admission control has
admitted the query, but the workload management insert DMLs operate differently
and the fetch rows timeout starts immediately after planning.
Additionally, the workload management queries do not time out as expected based
on the value set in the query_log_write_timeout_s coordinator startup flag.
was:
The insert into sys.impala_query_log DML that workload management executes is
cancelled after waiting for 10 seconds. For example, in the case where
admission control queues the query because the request pool is full or
executors are not available.
The short cancel timeframe is caused by the "FETCH_ROWS_TIMEOUT_MS" query
option not being set. Thus, the default of 10 seconds is used.
Additionally, the workload management queries do not time out as expected based
on the value set in the query_log_write_timeout_s coordinator startup flag.
> Workload Management Insert DMLs Timeouts Incorrect
> --------------------------------------------------
>
> Key: IMPALA-13772
> URL: https://issues.apache.org/jira/browse/IMPALA-13772
> Project: IMPALA
> Issue Type: Bug
> Reporter: Jason Fehr
> Assignee: Jason Fehr
> Priority: Critical
> Fix For: Impala 5.0.0
>
>
> The insert into sys.impala_query_log DML that workload management executes is
> cancelled after waiting for 10 seconds. For example, in the case where
> admission control queues the query because the request pool is full or
> executors are not available.
> The short cancel timeframe is caused by the "FETCH_ROWS_TIMEOUT_MS" query
> option not being set. Thus, the default of 10 seconds is used. With other
> queries, the fetch rows timeout only starts after admission control has
> admitted the query, but the workload management insert DMLs operate
> differently and the fetch rows timeout starts immediately after planning.
> Additionally, the workload management queries do not time out as expected
> based on the value set in the query_log_write_timeout_s coordinator startup
> flag.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
