-
Type: Bug
-
Resolution: Done
-
Priority: Major - P3
-
Affects Version/s: None
-
Component/s: None
-
None
We have noticed that some workloads are evicting internal pages when there are other candidates available.
A wtperf configuration that results in internal page eviction, when there is plenty of space in cache for the internal structure is:
conn_config="statistics=[all],statistics_log=(wait=10),cache_size=4G,eviction_trigger=85,eviction_target=70,eviction_dirty_trigger=85,eviction_dirty_target=60" table_config="type=file,allocation_size=4KB,internal_item_max=1433,internal_key_truncate=true,internal_page_max=16KB,leaf_item_max=1433,leaf_page_max=4KB,prefix_compression=false,split_pct=75" create=true key_sz=24 value_sz=900 icount=30000000 report_interval=5 run_time=1200 populate_threads=1 threads=((count=1,reads=1,throttle=5000))
Since the workload is read-only it's possible to rerun with create=false above, and not re-create the database between runs.
Investigating stats towards the end of such a run I see:
WT_TEST/WiredTigerStat.04.14:Dec 04 14:11:10 2728133 WT_TEST cache: internal pages evicted
- is depended on by
-
SERVER-21872 WiredTiger changes for 3.2.1
- Closed