[SERVER-16881] Pause due to sudden 4GB increase in cached bytes under WiredTiger Created: 15/Jan/15 Updated: 26/Jan/15 Resolved: 21/Jan/15 |
|
| Status: | Closed |
| Project: | Core Server |
| Component/s: | Storage |
| Affects Version/s: | 2.8.0-rc5 |
| Fix Version/s: | 3.0.0-rc6 |
| Type: | Bug | Priority: | Major - P3 |
| Reporter: | Bruce Lucas (Inactive) | Assignee: | Keith Bostic (Inactive) |
| Resolution: | Done | Votes: | 0 |
| Labels: | wiredtiger | ||
| Remaining Estimate: | Not Specified | ||
| Time Spent: | Not Specified | ||
| Original Estimate: | Not Specified | ||
| Attachments: |
|
| Backwards Compatibility: | Fully Compatible |
| Operating System: | ALL |
| Participants: |
| Description |
|
During one 10-minute run of a heavy mixed workload a period of several seconds of very low throughput was seen:
Here are the specific samples
The size of the sudden jump was: 7242022268 - 2941126278 = 3.98 GiB The size of the sudden drop was: 8594367123 - 4315898330 = 4.01 GiB Neither a sudden increase nor a sudden decrease of cache size by 4 GiB in 0.1 seconds due to real cached data is plausible. |
| Comments |
| Comment by Keith Bostic (Inactive) [ 26/Jan/15 ] |
|
This issue was caused by an accounting bug in tracking the memory currently held in cache, and I recall SERVER-16902 was a problem in selecting pages for splitting. (The accounting problem certainly might have exacerbated16902, though.) |
| Comment by Mark Callaghan [ 26/Jan/15 ] |
|
Is this caused by the intermittent use of huge, as in 4GB large), leaf pages? See comments from https://jira.mongodb.org/browse/SERVER-16902 |
| Comment by Keith Bostic (Inactive) [ 20/Jan/15 ] |
|
Thanks, bruce.lucas@10gen.com. If I instrument a WiredTiger branch to drop core when this happens, and build a binary on MCI, would that make it possible for you to give me a stack trace? (Not sure if MCI builds objects with symbol tables?) |