Details
-
Bug
-
Resolution: Fixed
-
Major
-
6.5.0
-
Untriaged
-
Yes
Description
We are observing a decrease in query throughput for many different benchmarks, all using Plasma.
The following tests all have similar issues:
- Avg. Query Throughput (queries/sec). CI19. Composite Filters (1000 -> 1). Covered. Plasma. request_plus
- Avg. Query Throughput (queries/sec). Q19. Composite Filters (1000 -> 1). Plasma. request_plus
- Avg. Query Throughput (queries/sec). CI19. Composite Filters (1000 -> 1). Covered. Plasma. not_bounded
- Avg. Query Throughput (queries/sec). Q19. Composite Filters (1000 -> 1). Plasma. not_bounded
- Avg. Query Throughput (queries/sec). AG6. Aggregation query. Plasma. not_bounded
- Avg. Query Throughput (queries/sec). AG1. Aggregation query. Plasma. not_bounded
- Avg. Query Throughput (queries/sec). AG5. Aggregation query. Plasma. not_bounded
- Avg. Query Throughput (queries/sec). AG9. Aggregation query. Plasma. not_bounded
- Avg. Query Throughput (queries/sec). Q20. Count (10K matches). 100M docs. Plasma. request_plus
- Avg. Query Throughput (queries/sec). Q20. Count (1K matches). 100M docs. Plasma. request_plus
- Avg. Query Throughput (queries/sec). Q20. Count (1K matches). 100M docs. Plasma. not_bounded
- Avg. Query Throughput (queries/sec). Q2. Singleton Unique Lookup. Plasma. DGM. request_plus
- Avg. Query Throughput (queries/sec). Q2. Singleton Unique Lookup. Plasma. DGM. not_bounded
- 90th percentile query latency (ms). Q2. Singleton Unique Lookup. 450 queries/sec. Plasma. request_plus
- Avg. Query Throughput (queries/sec). Q18. Range Scan with Limit and Offset (1K). Plasma. request_plus
- Avg. Query Throughput (queries/sec). Q18. Range Scan with Limit and Offset (1K). Plasma. not_bounded
- Avg. Query Throughput (queries/sec). CI18. Range Scan with Limit and Offset (1K). Covering. Plasma. not_bounded
- Avg. Query Throughput (queries/sec). CI18. Range Scan with Limit and Offset (1K). Covering. Plasma. request_plus
I will only provide data for one of the tests. If more data is require please let me know.
Test
Avg. Query Throughput (queries/sec). Q18. Range Scan with Limit and Offset (1K). Plasma. not_bounded
Index
CREATE INDEX by_capped_small ON `bucket-1`(capped_small);
Query
SELECT name, email FROM `bucket-1` WHERE capped_small > $1 ORDER BY capped_small OFFSET 1000 LIMIT 10;
Results
6.5.0-3123: 20338.0 q/s
6.5.0-3124: 11416.0 q/s
Report
Logs
6.5.0-3123
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14584/172.23.100.45.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14584/172.23.100.55.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14584/172.23.100.70.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14584/172.23.100.71.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14584/172.23.100.72.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14584/172.23.100.73.zip
6.5.0-3124
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14585/172.23.100.45.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14585/172.23.100.55.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14585/172.23.100.70.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14585/172.23.100.71.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14585/172.23.100.72.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-iris-14585/172.23.100.73.zip
Changelog
http://172.23.123.43:8000/getchangelog?product=couchbase-server&fromb=6.5.0-3123&tob=6.5.0-3124
PProf
6.5.0-3123
http://perf.jenkins.couchbase.com/job/iris/14584/artifact/172.23.100.55_n1ql_cpu_697116.pprof
6.5.0-3124
http://perf.jenkins.couchbase.com/job/iris/14585/artifact/172.23.100.55_n1ql_cpu_490cc3.pprof
Comment
I couldn't tell much from the pprof graphs, but if you look at the changelog it seems like the issue might be coming from some plasma commits.
Attachments
For Gerrit Dashboard: MB-34103 | ||||||
---|---|---|---|---|---|---|
# | Subject | Branch | Project | Status | CR | V |
109298,3 | MB-34103 Reset MVCCIterator.refreshCount on close | unstable | plasma | Status: MERGED | +2 | +1 |
110562,2 | MB-34103 Recycle wCtx on MVCCIterator.refresh() | unstable | plasma | Status: MERGED | +2 | +1 |
118565,2 | MB-34103 Revert "Recycle wCtx on MVCCIterator.refresh()" | unstable | plasma | Status: MERGED | +2 | +1 |