Biao Geng created FLINK-37518: --------------------------------- Summary: Optimize number of files created by flink history server Key: FLINK-37518 URL: https://issues.apache.org/jira/browse/FLINK-37518 Project: Flink Issue Type: Improvement Reporter: Biao Geng Attachments: image-2025-03-20-11-41-19-610.png
When using flink session cluster to run large-scale batch jobs, we may use flink history server to record the finished job status. However, it can easily use up all inodes of the filesystem as it will create lots of small files. For example, as the following picture shows, for a job with parallelism=640, each stage will have 640 records. If we run lots of jobs like running TPC-DS tests, the newly created small files would drain the filesystem's inode. !image-2025-03-20-11-41-19-610.png! -- This message was sent by Atlassian Jira (v8.20.10#820010)