When the cache is small we can allow internal pages to grow large enough that the cache becomes stuck.
The test/format configuration at the bottom of this report ended up with a cache containing 4 pages. Two pages for the metadata (that can't be evicted). Two internal pages in a tree:
0x98f790: row-store internal disk 0x98f710, entries 7, dirty, disk-alloc, write generation=2 0x7fa9880b5e70: row-store internal disk 0x7fa98806b000, entries 4981, dirty, disk-alloc, write generation=6
The configured cache size is 2MB and these 4 pages take up 2.5 MB. The second page with nearly 5000 entries is so large we can no longer make progress.
############################################ # RUN PARAMETERS ############################################ abort=0 auto_throttle=1 firstfit=0 bitcnt=8 bloom=1 bloom_bit_count=7 bloom_hash_count=14 bloom_oldest=0 cache=2 checkpoints=1 checksum=uncompressed chunk_size=3 compaction=0 compression=bzip-raw data_extend=0 data_source=table delete_pct=28 dictionary=0 evict_max=2 file_type=row-store backups=0 huffman_key=0 huffman_value=0 insert_pct=3 internal_key_truncation=1 internal_page_max=16 isolation=snapshot key_gap=3 key_max=72 key_min=28 leak_memory=0 leaf_page_max=13 logging=0 logging_archive=0 logging_prealloc=0 lsm_worker_threads=4 merge_max=20 mmap=1 ops=100000 prefix_compression=1 prefix_compression_min=6 repeat_data_pct=3 reverse=0 rows=100000 runs=100 split_pct=67 statistics=1 statistics_server=0 threads=16 timer=20 value_max=3805 value_min=16 wiredtiger_config= write_pct=59 ############################################