This ticket has been repurposed, to track a class of issues related to managing cache overflow (lookaside) and prepared updates. The prepared failures are described in comments below as well as in WT-4362 and WT-4364.
There is another smaller fix for the originally reported symptom regards unexpected page states in WT-4374.
LIMBO failure in format on the current develop branch:
t, file:wt.wt, WT_CURSOR.update: __wt_row_modify, 65: cbt->ref->state != WT_REF_LIMBO
I am unable to reproduce. Here's the CONFIG:
############################################ # RUN PARAMETERS ############################################ abort=0 alter=0 auto_throttle=1 backups=1 bitcnt=4 bloom=1 bloom_bit_count=46 bloom_hash_count=15 bloom_oldest=0 cache=31 cache_minimum=20 checkpoints=on checkpoint_log_size=135 checkpoint_wait=30 checksum=uncompressed chunk_size=10 compaction=0 compression=snappy data_extend=0 data_source=table delete_pct=11 dictionary=0 direct_io=0 encryption=rotn-7 evict_max=4 file_type=row-store firstfit=0 huffman_key=0 huffman_value=0 independent_thread_rng=1 in_memory=0 insert_pct=15 internal_key_truncation=1 internal_page_max=11 isolation=read-committed key_gap=6 key_max=30 key_min=14 leaf_page_max=16 leak_memory=0 logging=0 logging_archive=1 logging_compression=none logging_file_max=41093 logging_prealloc=0 long_running_txn=0 lsm_worker_threads=4 merge_max=9 mmap=1 modify_pct=0 ops=100000 prefix_compression=1 prefix_compression_min=6 quiet=1 read_pct=22 rebalance=1 repeat_data_pct=13 reverse=0 rows=100000 runs=1 salvage=1 split_pct=56 statistics=0 statistics_server=0 threads=31 timer=360 transaction_timestamps=1 transaction-frequency=4 value_max=3986 value_min=3 verify=1 wiredtiger_config= write_pct=52 ############################################
cc: sueloverso