Description
We have noticed some fairly significant throughput drops in Import Lines tests, of between 15-30%.
Two tests specifically are impacted:
- http://showfast.sc.couchbase.com/#/timeline/Linux/tools/import/all#tools_import_60M_json_lines_1s_1c_leto
- http://showfast.sc.couchbase.com/#/timeline/Linux/tools/import/all#tools_import_60M_json_lines_leto
We do see a history of throughput fluctuations in these tests, so more investigation is required from our end too to confirm what is going on, but with needing to sign off on RC1 of 7.1.1 I wanted to raise this just in case.
Examples of good/bad tests from test 2:
- Good (7.1.1-3135): http://perf.jenkins.couchbase.com/job/leto/24894/
- Bad (7.1.1-3166): http://perf.jenkins.couchbase.com/job/leto/25001/
Cbmonitor comparison: http://cbmonitor.sc.couchbase.com/reports/html/?snapshot=leto_711-3135_import_data_84d0&snapshot=leto_711-3166_import_data_b2df
I am rerunning the "good" test above to confirm whether the performance differences are environmental or product-related. The "bad" result is reproducable: http://showfast.sc.couchbase.com/#/runs/tools_import_60M_json_lines_leto/7.1.1-3166
Sorry for the quick-and-dirty initial report, I will be updating this tomorrow.