[SERVER-84549] Do not count fetch as read in SBE trial run Created: 04/Jan/24 Updated: 09/Jan/24 Resolved: 09/Jan/24 |
|
| Status: | Closed |
| Project: | Core Server |
| Component/s: | None |
| Affects Version/s: | None |
| Fix Version/s: | None |
| Type: | Task | Priority: | Major - P3 |
| Reporter: | Ivan Fefer | Assignee: | Ivan Fefer |
| Resolution: | Won't Fix | Votes: | 0 |
| Labels: | None | ||
| Remaining Estimate: | Not Specified | ||
| Time Spent: | Not Specified | ||
| Original Estimate: | Not Specified | ||
| Sprint: | QE 2024-01-22 |
| Participants: |
| Description |
|
SBE multi planner currently counts every read: index seek/scan and collection seek/scan as a "read" and the main measure of SBE plan in multi planner is reads per doc. Because of this, plan with fetch needs to be twice as efficient as plan without fetch to be selected. In classic, noFetch bonus is much smaller. To make reads more comparable with classic's works and avoid this problem, we should not count fetch stage in trial run tracker. |
| Comments |
| Comment by Ivan Fefer [ 09/Jan/24 ] |
|
Fetch is an expensive operation and having multi planner count it as a separate read is preferable. |