[SERVER-16182] Under WiredTiger actual leaf_page_max appears to be 75% of specified value Created: 17/Nov/14  Updated: 10/Dec/14  Resolved: 17/Nov/14

Status: Closed
Project: Core Server
Component/s: Diagnostics, Storage
Affects Version/s: 2.8.0-rc0
Fix Version/s: None

Type: Bug Priority: Major - P3
Reporter: Bruce Lucas (Inactive) Assignee: Unassigned
Resolution: Done Votes: 0
Labels: wiredtiger
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified

Operating System: ALL
Participants:

 Description   

With default leaf_page_max of 32 kB for collections, actual leaf pages appear never to be larger than 24 kB; 16 kB is default for indexes but actual leaf pages are never larger than 12 kB. Increase leaf_page_max to 64 kB and observed leaf page sizes are never larger than 48 kB.



 Comments   
Comment by Bruce Lucas (Inactive) [ 18/Nov/14 ]

Thanks agorrod. So does that mean for tables like MongoDB collections where (I believe) inserts are always at the end, a split_pct of 100 would be fine? Would there be any advantage to setting it to 100% - for example, would it save memory, or does leaf_page_max have no effect on the amount of memory allocated for a page?

Comment by Alex Gorrod [ 17/Nov/14 ]

This behavior is deliberate. There is a configuration for WiredTiger WT_SESSION::create that allows you to tune how full we make leaf pages called "split_pct". You can read about it in the documentation here:

http://source.wiredtiger.com/2.4.1/struct_w_t___s_e_s_s_i_o_n.html#a358ca4141d59c345f401c58501276bbb

The reason we don't fill pages up completely when they are first created is to allow space for future updates without needing to split into multiple pages.

Comment by Eliot Horowitz (Inactive) [ 17/Nov/14 ]

Is this actually a thing to change?
Seems like its just part of their math which is fine...

Generated at Thu Feb 08 03:40:13 UTC 2024 using Jira 9.7.1#970001-sha1:2222b88b221c4928ef0de3161136cc90c8356a66.