We added (some) bazel build support to spark. It's not complete: it only builds the parts of spark we use and there are cases where the spark test infra, based on maven and including some integration testing, is not compatible with bazel ... at least not without significant changes, particularly given we need to support maven and bazel builds in parallel.
It's a moderately big code base (~5K java/scala files). I did it manually: it started out as an experiment, not being sure it could be completed. It took a few days, in particular going through the tests which include fairly complicated setups and when they fail, fail in complicated ways. But I thought (and still tend to think) it might be easier to do the conversion manually than spending time trying to debug the result of an automatic conversion.