[ https://issues.apache.org/jira/browse/FLINK-5146?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15690104#comment-15690104 ]
ASF GitHub Bot commented on FLINK-5146: --------------------------------------- GitHub user StefanRRichter opened a pull request: https://github.com/apache/flink/pull/2857 [FLINK-5146] Improved resource cleanup in RocksDB keyed state backend Currently, the resources such as taken snapshots or iterators are not always cleaned up in the RocksDB state backend. In particular, not starting the runnable future will leave taken snapshots unreleased. This PR improves the behaviour of releasing resources obtained through the RocksDB JNI bridge. IMPORTANT NOTICE: This PR also upticks the current version of RocksDB to 4.11.2, as I found some instabilities in the old version that could lead to segfaults, in particular at GC time. You can merge this pull request into a Git repository by running: $ git pull https://github.com/StefanRRichter/flink rocks-release-snapshot Alternatively you can review and apply these changes as the patch at: https://github.com/apache/flink/pull/2857.patch To close this pull request, make a commit to your master/trunk branch with (at least) the following in the commit message: This closes #2857 ---- commit b041d87ee97e03e0442f151594e1e5b6f884a0f4 Author: Stefan Richter <s.rich...@data-artisans.com> Date: 2016-11-12T20:13:28Z Improved resource cleanup commit fc032f29d18f662d56a362a8198892cc9961ea72 Author: Stefan Richter <s.rich...@data-artisans.com> Date: 2016-11-12T22:26:48Z Unit Tests commit 97dfc6fc5c417415d344b7f8ce443eebebdddda4 Author: Stefan Richter <s.rich...@data-artisans.com> Date: 2016-11-13T04:04:03Z Uptick rocks version commit 4a988d55641a6971301c69b7b9ddf5b673b1c1c4 Author: Stefan Richter <s.rich...@data-artisans.com> Date: 2016-11-13T05:04:20Z testing something commit 90a90eb121f4a187fe8aaf909388a0c470767896 Author: Stefan Richter <s.rich...@data-artisans.com> Date: 2016-11-13T14:22:14Z ownership transfer ---- > Improved resource cleanup in RocksDB keyed state backend > -------------------------------------------------------- > > Key: FLINK-5146 > URL: https://issues.apache.org/jira/browse/FLINK-5146 > Project: Flink > Issue Type: Improvement > Components: State Backends, Checkpointing > Reporter: Stefan Richter > > Currently, the resources such as taken snapshots or iterators are not always > cleaned up in the RocksDB state backend. In particular, not starting the > runnable future will leave taken snapshots unreleased. > We should improve the releases of all resources allocated through the RocksDB > JNI bridge. -- This message was sent by Atlassian JIRA (v6.3.4#6332)