[jira] [Commented] (HADOOP-18218) Ensure S3A prefetching stream memory consumption scales
[ https://issues.apache.org/jira/browse/HADOOP-18218?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17742420#comment-17742420 ] Steve Loughran commented on HADOOP-18218: - think i will make memory consumption/release something in a gauge visible for a stream and for the s3afs as a whole -where I'd also like to count #of open streams > Ensure S3A prefetching stream memory consumption scales > --- > > Key: HADOOP-18218 > URL: https://issues.apache.org/jira/browse/HADOOP-18218 > Project: Hadoop Common > Issue Type: Sub-task > Components: fs/s3 >Affects Versions: 3.4.0 >Reporter: Steve Loughran >Priority: Major > > A recurrent problem in cloud store IO is running out of memory because blocks > are buffered in reads or writes. > We need to make sure that data/memory is managed in the prefetch code such > that it works in processes with many worker threads (hive, spark) and > does not exhaust resources. -- This message was sent by Atlassian Jira (v8.20.10#820010) - To unsubscribe, e-mail: common-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: common-issues-h...@hadoop.apache.org
[jira] [Commented] (HADOOP-18218) Ensure S3A prefetching stream memory consumption scales
[ https://issues.apache.org/jira/browse/HADOOP-18218?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17560980#comment-17560980 ] Daniel Carl Jones commented on HADOOP-18218: I'll scope this ticket down to focus on memory consumption. HADOOP-18291 highlights that there is no limits or clean-up for the disk cache. > Ensure S3A prefetching stream memory consumption scales > --- > > Key: HADOOP-18218 > URL: https://issues.apache.org/jira/browse/HADOOP-18218 > Project: Hadoop Common > Issue Type: Sub-task > Components: fs/s3 >Affects Versions: 3.3.9 >Reporter: Steve Loughran >Priority: Major > > A recurrent problem in cloud store IO is running out of memory because blocks > are buffered in reads or writes. > We need to make sure that data/memory is managed in the prefetch code such > that it works in processes with many worker threads (hive, spark) and > does not exhaust resources. -- This message was sent by Atlassian Jira (v8.20.10#820010) - To unsubscribe, e-mail: common-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: common-issues-h...@hadoop.apache.org