[ https://issues.apache.org/jira/browse/OAK-4274?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15256042#comment-15256042 ]
Michael Dürig commented on OAK-4274: ------------------------------------ This might turn out to be worse than unmapping being deferred by JVM gc: client code holding a reference to a record "forever" will keep the containing tar file mapped, although the underlying segment might have been gc'ed already by revision gc. OTOH I don't think this is a problem re. memory consumption as the underlying OS will simply swap the actual memory content out. It might be a problem re. number of file handles. Definitely something to keep an eye on. Maybe [~jsedding]'s proposal with the phantom references gives us a way to better monitor this!? > Memory-mapped files can't be explicitly unmapped > ------------------------------------------------ > > Key: OAK-4274 > URL: https://issues.apache.org/jira/browse/OAK-4274 > Project: Jackrabbit Oak > Issue Type: Bug > Components: segment-next, segmentmk > Reporter: Francesco Mari > Labels: gc, resilience > Fix For: 1.6 > > > As described by [this JDK > bug|http://bugs.java.com/view_bug.do?bug_id=4724038], there is no way to > explicitly unmap memory mapped files. A memory mapped file is unmapped only > if the corresponding {{MappedByteBuffer}} is garbage collected by the JVM. -- This message was sent by Atlassian JIRA (v6.3.4#6332)