Yingjie Cao created FLINK-25960: ----------------------------------- Summary: Distribute the data read buffers more fairly among result partitions for sort-shuffle Key: FLINK-25960 URL: https://issues.apache.org/jira/browse/FLINK-25960 Project: Flink Issue Type: Sub-task Components: Runtime / Network Reporter: Yingjie Cao Fix For: 1.15.0
Currently, the data read buffers for sort-shuffle are allocated in a random way and some result partitions may occupy too many buffers which leads to the starvation of other result partitions. This ticket aims to improve the scenario by not reading data for those result partitions which already occupy more than the average number of read buffers per result partition. -- This message was sent by Atlassian Jira (v8.20.1#820001)