danpi opened a new issue, #4554: URL: https://github.com/apache/bookkeeper/issues/4554
**BUG REPORT** ***Describe the bug*** After upgrading from Pulsar 2.8 to LTS 3.0+, with the corresponding BookKeeper version upgraded from 4.14.4 to 4.16.6, the system initially runs well. However, when the throughput increases to a certain level, such as 100MB of writes per node, and after running for a period of time, I observe that the system's physical memory usage increases linearly. Meanwhile, the BookKeeper JVM itself does not show significant memory changes, and there are a large number of SST files in BookKeeper's locations that remain uncleared. ***To Reproduce*** Steps to reproduce the behavior: 1. Deploy a minimal Pulsar cluster. 2. Configure the message retention time in the Pulsar broker for the namespace, setting both retention and TTL to 6 hours. 3. Use the Pulsar perf tool to write data, maintaining a throughput of 30MB/s or higher per BookKeeper node. The higher the throughput, the easier it is to reproduce the issue. 4. After running continuously for a few days, you will observe that the earliest *.log files in BookKeeper are from 6 hours ago, but the earliest files in the locations directory are still from several days ago and cannot be cleared. ***Expected behavior*** The SST files and entry logs should both be retained for only 6 hours, as configured. There should not be a large accumulation of SST files, which can negatively impact query performance, increase disk storage pressure, and lead to physical memory usage issues. ***Screenshots*** As shown in the figure, the entryLog files are retained for a maximum of 6 hours. <img width="917" alt="Image" src="https://github.com/user-attachments/assets/fb4eea99-0370-4275-acb7-0756cf74a831"; /> However, the oldest SST files in the locations directory remain uncleared. <img width="989" alt="Image" src="https://github.com/user-attachments/assets/76bf40b7-b540-42b1-a8bd-d1227cbddab1"; /> ***Additional context*** -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
