This issue is reproduced with the following configuration:
############################################ # RUN PARAMETERS ############################################ # bitcnt not applicable to this run cache=30 compression=none data_extend=0 data_source=table delete_pct=7 dictionary=1 file_type=row-store hot_backups=0 huffman_key=0 huffman_value=0 insert_pct=50 internal_key_truncation=0 internal_page_max=13 key_gap=16 key_max=64 key_min=12 leaf_page_max=12 ops=500000 prefix=1 repeat_data_pct=30 reverse=0 rows=100 runs=0 split_pct=52 threads=20 value_max=256 value_min=19 # wiredtiger_config not applicable to this run write_pct=45 ############################################
The configuration is very similar to that in issue WT-639, but the symptoms are different. I've seen two types of failure:
10: table, row-store t, file:wt.wt, session.checkpoint: illegal cell and page type combination: cell 4 on page at [write-check] is a value/ovfl,rm cell on a row-store leaf page t: session.checkpoint: WT_ERROR: non-specific WiredTiger error
and:
12: table, row-store t, file:wt.wt, session.verify: checkpoint ranges never verified: 15 t, file:wt.wt, session.verify: file ranges never verified: 669 t: session.verify: table:wt: post-ops verify: WT_ERROR: non-specific WiredTiger error
Running on a machine with SSDs, the failure generally happens after less than 20 iterations of test/format.
If I disable the in-memory page split code, I no longer see a failure.