[Bug] Some cache will be removed after FE restarted #56583
Description
Search before asking
- I had searched in the issues and found no similar issues.
Version
3.0.8
What's Wrong?
After FE restarted, FE will rebuild CloudTabletRebalancer.beToTabletsGlobal. If an BE send a report tablet rpc, FE will calcutate diff based on it. But, now it is empty. So, FE will drop many tablets in BE.
FE LOG:
2025-09-22 10:35:08,405 INFO (main|23) [MetaReader.read():136] finished to load image in 142476 ms
2025-09-22 10:35:09,776 INFO (cloud tablet rebalancer|23) [CloudTabletRebalancer.runAfterCatalogReady():238] cloud tablet rebalance begin
2025-09-22 10:35:09,776 INFO (cloud tablet rebalancer|23) [CloudTabletRebalancer.buildClusterToBackendMap():273] cluster to backends {GnCmF8H6=[1758091923812, 1758128301417, 1758091923813, 1758128301416, 1758091923814, 1758128301418]}
2025-09-22 10:35:12,489 INFO (cloud tablet rebalancer|244) [CloudTabletRebalancer.completeRouteInfo():633] collect to editlog route 0 infos, error num 0
2025-09-22 10:35:12,489 INFO (cloud tablet rebalancer|244) [CloudTabletRebalancer.checkInflightWarmUpCacheAsync():476] collect to editlog warmup before size=0 after size=0 infos
2025-09-22 10:35:15,761 INFO (report-thread|244) [CloudReportHandler.deleteFromBackend():79] delete tablet[1758097385790] from backend[1758128301416]
2025-09-22 10:35:28,648 INFO (cloud tablet rebalancer|244) [CloudTabletRebalancer.runAfterCatalogReady():257] finished to rebalancer. cost: 18872 ms
BE LOG:
be.INFO.log.20250920-144534:RuntimeLogger I20250922 10:35:16.936863 610025 task_worker_pool.cpp:425] successfully submit task|type=DROP|signature=1758097385790
What You Expected?
tabletReport should after beToTabletsGlobal was built.
How to Reproduce?
kill FE master.
Anything Else?
No response
Are you willing to submit PR?
- Yes I am willing to submit a PR!
Code of Conduct
- I agree to follow this project's Code of Conduct
