[ https://issues.apache.org/jira/browse/HIVE-15959?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15872423#comment-15872423 ]
Siddharth Seth commented on HIVE-15959: --------------------------------------- The code that has been moved from package.py to the ServiceDriver - package.py was heapSize*1.2. Java is using *2 Computing the headroom in the daemon. Didn't understand this part (will be 0 very often) {code} return Math.max(0, minHeadRoom - headRoom); {code} There's normally a 6+GB headroom between (xmx + cache) and container-size - to account for off heap allocations. This headroom was to carve out a part of the heap for llap data structures, and the cache. Think the simples would be to take the configured value (5%) and use that with JVM.maxAvailable. > LLAP: fix headroom calculation and move it to daemon > ---------------------------------------------------- > > Key: HIVE-15959 > URL: https://issues.apache.org/jira/browse/HIVE-15959 > Project: Hive > Issue Type: Bug > Reporter: Sergey Shelukhin > Assignee: Sergey Shelukhin > Attachments: HIVE-15959.patch > > -- This message was sent by Atlassian JIRA (v6.3.15#6346)