Aitozi created FLINK-24124: ------------------------------ Summary: Use explicit pool size to construct thread pool instead of Hardware.getNumberCPUCores() Key: FLINK-24124 URL: https://issues.apache.org/jira/browse/FLINK-24124 Project: Flink Issue Type: Improvement Components: Runtime / Coordination Reporter: Aitozi
As described in the [comments|https://issues.apache.org/jira/browse/FLINK-21685?focusedCommentId=17303866&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-17303866]. If we calculate the pool size based on hardware information in containerize environment, it will cause the value too small. But there are some code still use the {{Hardware.getNumberCPUCores()}} as the thread pool size like in {{JobManagerSharedServices#scheduledExecutorService}} . This pool will be used to do like : * dispose checkpoint (which is a I/O work) * connection to resourceManager * ... But we often only set one core for jobManger. This will lead to all the above action can only be executed one by one. So this ticket is meant to use explicit pool size to avoid these problems in containerize environment. -- This message was sent by Atlassian Jira (v8.3.4#803005)