[ https://issues.apache.org/jira/browse/KAFKA-7729?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16720114#comment-16720114 ]
Dimitris Mandalidis commented on KAFKA-7729: -------------------------------------------- Thanks [~ijuma], do you know if there's an issue for this? (googling it shows very old such bugs, and we 're using the latest 1.8 version) > Mysterious disk increase when switching to RedHat OpenJDK > --------------------------------------------------------- > > Key: KAFKA-7729 > URL: https://issues.apache.org/jira/browse/KAFKA-7729 > Project: Kafka > Issue Type: Bug > Affects Versions: 1.1.0 > Reporter: Dimitris Mandalidis > Priority: Minor > > (maybe it's not a bug actually, sort of a question) > We run a Kafka cluster 1.1.0 with 3 brokers running in RHEL 7.5 having their > logs over an LVM ext4 partition (fsck'd an hour ago). The cluster is using > Oracle JDK 8u144 and it uses 45G of disk space under normal circumstances. > We 've started switching to RedHat OpenJDK 1.8.0_191, one broker at a time. > Since a broker had to be restarted, we thought that disk space increase was > expected during rejoining the cluster, because it will first try to become > ISR and it will roll segments (and delete if needed) afterwards (correct me > if I 'm wrong). So the first OpenJDK broker ended up using 65G of disk space. > For the second broker, because we will be running short of disk space, we > decided to take him down and delete all the logs. So it also ended up using > 65G of space. > We let the two brokers to operate as such for 8 hours expecting that they > will eventually stabilize in the previous 45G but they didn't. The current > situation is that we have 2 OpenJDK brokers holding 65G of disk usage, and > one Oracle JDK broker which still holds 45G and waits to be migrated. > The actual question is that we cannot explain why we saw 20G disk space > increase just switching to OpenJDK. We had performed rolling upgrades in the > past (we 've started with 0.10) but we have never seen such a difference. Do > you have any idea? -- This message was sent by Atlassian JIRA (v7.6.3#76005)