Tim Armstrong created IMPALA-8804:
-------------------------------------
Summary: Memory based admission control is always disabled if pool
max mem is not set
Key: IMPALA-8804
URL: https://issues.apache.org/jira/browse/IMPALA-8804
Project: IMPALA
Issue Type: Improvement
Reporter: Tim Armstrong
Attachments: minicluster-fair-scheduler.xml, minicluster-llama-site.xml
Memory-based admission control doesn't kick in with the provided config files
where no max memory is configured for the pool. This is the documented
behaviour and not a bug - see
https://impala.apache.org/docs/build/html/topics/impala_admission.html.
However, it is inconvenient since you need to specify some max memory value
even if you don't want to limit the pool's share of the clusters resources (or
the cluster is variable in size).
This is unfriendly. It is also confusing since there is no explicit way to
enable memory-based admission control.
You can workaround by setting the pool max memory to a very high value.
To reproduce, start a minicluster with the provided configs. If you submit
multiple memory-intensive queries in parallel, they will never be queued.
{noformat}
start-impala-cluster.py
--impalad_args="-fair_scheduler_allocation_path=minicluster-fair-scheduler.xml
-llama_site_path=minicluster-llama-site.xml"
--impalad_args=-vmodule=admission-controller
{noformat}
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
