Sometimes coverage dips far below the threshold (90.x% instead of 98.x%). When this happens it is always the python integration test runs not being considered when it comes to computing
the coverage. Some more information on this can be found here.

I'm filing this issue because it just happened on a CI run following a merge to master:
https://app.circleci.com/pipelines/github/envoyproxy/nighthawk/1640/workflows/6823a1ea-bb7d-4bce-a42c-ae6c6b38de5d/jobs/12923/steps

TLDR; we probably need to wait for clang-10 to get used in the docker image we use to test, and we source that image from Envoy.
If that doesn't resolve it, more looking into this is warranted.