In 6.6.3-9805 there is a panic in the query logs:
This panic is also present in 6.6.3-9806. In addition in 6.6.3-9806, in the ns_server error log there are errors indicating connection refused by query admin/stats endpoint. This results in the reported query throughput number being lower but in reality, there is no performance regression.
9805 - http://perf.jenkins.couchbase.com/job/iris-multi-client/10740/ - 126314.0
9806 - http://perf.jenkins.couchbase.com/job/iris-multi-client/10737/ - 83959.0
Not sure how long the query panic has been present, but can try to find the initial offending build. Looks like there is no relevant commit in 9806 to cause this issue. Last god run was 9804 so probably caused by this commit in 9805 and the admin/stats issue is spurious side effect: