-
Type:
Bug
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
None
-
Storage Engines - Transactions
-
268.967
-
SE Transactions - 2026-09-25
-
3
__wt_verbose_dump_cache() walks every btree dhandle to report per-file page counts. That walk took a SIGSEGV, killing a process that was only trying to report diagnostics.
Identical crash frames in three separate cores:
#0 __wt_page_in_func bt_read.c:553 <- btree = S2BT(session) #1 __wt_page_swap_func btree_inline.h:2630 (held=0x0) #2 __tree_walk_internal bt_walk.c:337 #3 __wt_tree_walk bt_walk.c:431 #4 __verbose_dump_cache_single evict_verbose.c:63 #5 __verbose_dump_cache_apply evict_verbose.c:152 #6 __wt_verbose_dump_cache evict_verbose.c:193 #7 __conn_debug_info conn_api.c:1437
Observed effects: the dump truncated before cache dump: total found and the per-file dirty mass, and the intended clean __wt_abort() became a SIGSEGV, so Build Baron extracts a Segmentation Failure fault and the instances misclassify.
Seen in three BFGs, all wiredtiger @ 92445b64, format-stress-test-disagg-switch-data-validation-*:
- BFG-3495446 (ubuntu2004-stress-tests-arm64)
- BFG-3495728 (ubuntu2004-stress-tests)
- BFG-3496860 (ubuntu2004-release-stress-tests)
Reached in all three via format's watchdog calling conn->debug_info.
- is related to
-
WT-18260 test/format (disagg.mode=off) run more than 15 minutes past the maximum time: cache over dirty trigger, eviction workers idle
-
- Open
-