Tablet corruption after memory limit exceeded during rowset commit Hi all! We've been investigating a tablet corruption issue in our cluster and wanted to check if this is a known failure mode and whether it's been addressed in 4.0. What we observed: Before queries against the tablet started failing, we saw this exception in the BE logs:

W20251127 09:55:40.290861 280859927838336 tablet_updates.cpp:1037] apply tablet: 11165755 failed and retry later, status: Memory limit exceeded: _apply_rowset_commit error: apply rowset update state failed: Memory limit exceeded: Memory of update exceed limit. read and decompress page Backend: pocus-starrocks-be-2.pocus-starrocks-be-search.starrocks.svc.cluster.local, Used: 66048794146, Limit: 64537123921.

Followed later by:

W20251127 10:46:30.427786 281454309604992 tablet_updates.cpp:3934] get_applied_rowsets(version 362) failed tablet:11165755 #version:1 [363 363@0 363] #pending:0 cost (0/0/0)

It looks like the memory limit was exceeded during

_apply_rowset_commit

, and the tablet ended up in a corrupted/unrecoverable state afterward. Questions: 1. Is this a known issue where memory pressure during rowset commits can lead to tablet corruption? 2. Has this failure mode been addressed in 4.0? Appreciate any insights - thanks! EDIT:

current_version()

is

3.5.6-3ba6b8d