[SERVER-17195] WiredTiger: highly contended write traffic may lock collection Created: 05/Feb/15 Updated: 15/May/15 Resolved: 18/Feb/15 |
|
| Status: | Closed |
| Project: | Core Server |
| Component/s: | WiredTiger |
| Affects Version/s: | 3.0.0-rc8 |
| Fix Version/s: | 3.0.0-rc9, 3.1.0 |
| Type: | Bug | Priority: | Critical - P2 |
| Reporter: | Rui Zhang (Inactive) | Assignee: | Susan LoVerso |
| Resolution: | Done | Votes: | 1 |
| Labels: | 28qa | ||
| Remaining Estimate: | Not Specified | ||
| Time Spent: | Not Specified | ||
| Original Estimate: | Not Specified | ||
| Description |
Test designwith wiredTiger, designed following test
ObservationWhen both update and read are to/from the same collection, very often, there will be pause in both read/write (close to 0, operation still gets done occasionally) Can reproduce this in regular basis Pause will last quite long for some case, for example, following pause lasts more than 3 minutes
When update and read are to/from different collections, there will still be pause for write, but the read is largely Ok and benefits from the reduced traffic from another collection.
Setup
|
| Comments |
| Comment by Rui Zhang (Inactive) [ 17/Feb/15 ] | |||||||||||
|
test rc9, run the the script for 120 second without any drop of throughput, and the overall throughput for update is higher than that of rc8. rc9 show around 28k throughput and without drop
rc8 peak at around 22k, and TPS drop to close to 0 after around 30 second
| |||||||||||
| Comment by Daniel Pasette (Inactive) [ 11/Feb/15 ] | |||||||||||
|
Sorry to be a bit pedantic with the jira hygiene, but we can't mark it "resolved" until we get a new drop from WT. Planned for tmrw morning. Then we'll resolve and backport. | |||||||||||
| Comment by Michael Cahill (Inactive) [ 10/Feb/15 ] | |||||||||||
|
Hi sallgeud, sorry for the confusion – I've updated the fix version, this will be addressed in 3.0. | |||||||||||
| Comment by Chad Kreimendahl [ 10/Feb/15 ] | |||||||||||
|
I see this as "fixed" and also labeled 3.1. Is that because the next wiredtiger iteration won't be included until 3.1 or will this update be included in 3.0? | |||||||||||
| Comment by Susan LoVerso [ 06/Feb/15 ] | |||||||||||
|
I am able to reproduce this. I have opened https://github.com/wiredtiger/wiredtiger/issues/1647 for this issue. Since we're updating the same thing over and over, WT ends up with a very long list of update structures that we're walking and the size of that list means the walks start taking a very long time. We'll evaluate the fix for inclusion in the next WT drop. |