Hit this on a patch, not on mainline on format-stress-test-disagg-multi-snapshot-read-1 }}{{amazon2023-stress-tests-arm64 (linked in comment) aborted with a SIGABRT and core dump.

stack

#3  __wt_abort                    src/os_common/os_abort.c:32
#4  __split_multi_inmem           src/btree/bt_split.c:1698
#5  __wt_split_rewrite            src/btree/bt_split.c:2576
#6  __evict_page_dirty_update     src/evict/evict_page.c:899
#7  __wt_evict                    src/evict/evict_page.c:706
#8  __wti_evict_page              src/evict/evict_dispatch.c:254
#9  __wti_evict_lru_pages         src/evict/evict_queue.c:140
#10 __evict_pass                  src/evict/evict_thread.c:696
#11 __evict_server                src/evict/evict_thread.c:198

• bt_split.c:1698 on develop is WT_ASSERT(session, prev_onpage->next == tmp). the CI log only has the SIGABRT and stack, not the assertion text, so that's read from the source line. 
• it's the "restore this update chain" check: after walking from the onpage update, the next update should be the one we decided to restore.
• seeds: random.data_seed=6531864, random.extra_seed=3147232

command

./t -c ../../../test/format/CONFIG.disagg disagg.multi=1 disagg.multi_validation=1 disagg.snapshot_read=1 runs.predictable_replay=1 runs.rows=50000:200000 runs.ops=500000:2000000

resolved config: rows=80000, ops=2000000, 3 tables, 19 threads, leader mode, palite page log.

things in the config that look relevant

• stress.failpoint_eviction_split=1 }}(fails 1% of eviction splits with {{{}EBUSY, see __split_multi_lock)
• ops.prepare=1, preserve_prepared=0, precise_checkpoint=1
• tiny memory_page_max: table1=4, table3=1
• table1 is delete heavy (pct.delete=13)

not reproduced yet

• ran the full resolved CONFIG locally (macOS, Debug+diagnostic) 7 times on the patch branch and 6 times on plain develop. all passed, so no local repro. likely needs linux/arm64 or just more iterations, it only failed once in 10 here.

not the victim cache

• the patch was WT-18808 (palite victim cache sizing), but the victim cache was off in this run. no victim_cache option is set anywhere in the CONFIG or the task definition, both palite bounds default to 0, and the diff doesn't touch bt_split.c or evict.
• I couldnt find pre-existing BF/BFG.