You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@impala.apache.org by "Matthew Mulder (JIRA)" <ji...@apache.org> on 2017/03/21 00:46:41 UTC
[jira] [Updated] (IMPALA-5104) admission control memory check
shouldn't fail queries with estimates equal to MEM_LIMIT
[ https://issues.apache.org/jira/browse/IMPALA-5104?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Matthew Mulder updated IMPALA-5104:
-----------------------------------
Description: I got the error{code}MainThread DEBUG:concurrent_select[853]:Error running query with id b647161dd5c826da:8f641d6100000000: Rejected query from pool root.default : request memory needed 256.00 GB per node is greater than process mem limit 256.00 GB.{code}which didn't make sense to me until I looked in the code and saw that the operation used was actually{code}schedule->GetPerHostMemoryEstimate() >= GetProcMemLimit(){code}It is a use case, at least during testing, to run a query through EXPLAIN to get the memory usage and then set MEM_LIMIT to the resulting estimate. To make this use case work, the above comparison should be changed to{code}schedule->GetPerHostMemoryEstimate() > GetProcMemLimit(){code}to avoid rejecting the query. The code is in admission-controller.cc (was: I got the error{code}MainThread DEBUG:concurrent_select[853]:Error running query with id b647161dd5c826da:8f641d6100000000: Rejected query from pool root.default : request memory needed 256.00 GB per node is greater than process mem limit 256.00 GB.{code}which didn't make sense to me until I looked in the code and saw that the operation used was actually{code}schedule->GetPerHostMemoryEstimate() >= GetProcMemLimit(){code}The error message should be fixed to write "greater than or equal to" to avoid this confusion. This is in admission-controller.cc)
Summary: admission control memory check shouldn't fail queries with estimates equal to MEM_LIMIT (was: REASON_REQ_OVER_NODE_MEM needs to write "greater than or equal to")
> admission control memory check shouldn't fail queries with estimates equal to MEM_LIMIT
> ---------------------------------------------------------------------------------------
>
> Key: IMPALA-5104
> URL: https://issues.apache.org/jira/browse/IMPALA-5104
> Project: IMPALA
> Issue Type: Bug
> Components: be
> Affects Versions: Impala 2.8.0
> Reporter: Matthew Mulder
> Priority: Trivial
>
> I got the error{code}MainThread DEBUG:concurrent_select[853]:Error running query with id b647161dd5c826da:8f641d6100000000: Rejected query from pool root.default : request memory needed 256.00 GB per node is greater than process mem limit 256.00 GB.{code}which didn't make sense to me until I looked in the code and saw that the operation used was actually{code}schedule->GetPerHostMemoryEstimate() >= GetProcMemLimit(){code}It is a use case, at least during testing, to run a query through EXPLAIN to get the memory usage and then set MEM_LIMIT to the resulting estimate. To make this use case work, the above comparison should be changed to{code}schedule->GetPerHostMemoryEstimate() > GetProcMemLimit(){code}to avoid rejecting the query. The code is in admission-controller.cc
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)