I see from mach raptor-test --help that there are some options for tweaking gecko profiler runs:

  --gecko-profile       Profile the run and output the results in
                        $MOZ_UPLOAD_DIR. After talos is finished,
                        profiler.firefox.com will be launched in Firefox so
                        you can analyze the local profiles. To disable auto-
                        launching of profiler.firefox.com set the
                        DISABLE_PROFILE_LAUNCH=1 env var.
  --gecko-profile-interval GECKO_PROFILE_INTERVAL
                        How frequently to take samples (milliseconds)
  --gecko-profile-entries GECKO_PROFILE_ENTRIES
                        How many samples to take with the profiler

However, what I don't see is the ability to manipulate the list of threads being recorded, nor changing the feature set selected.

This enhancement comes from me wishing I could use raptor-test along side the experimental JSTracer option, but as far as I can tell there's no way for me to ask for the JSTracer option in a raptor test.

Yes currently the gecko profiling options are hard-coded in the Raptor webext [0]. The best option at this point is to manipulate that file and push to try. We could add more --gecko-profile command line options (i.e. threads to profile) however that cannot be passed on to try runs easily at all - that requires adding additional support to ./mach try fuzzy etc. but could be supported with some work down the road.

[0] https://searchfox.org/mozilla-central/rev/0376cbf447efa16922c550da3bfd783b916e35d3/testing/raptor/webext/raptor/runner.js#323

From a personal standpoint, I'm actually most interested in local runs, but I take your point about try runs.

Btw. is JSTracer a thread or a feature?

JSTracer is a feature https://searchfox.org/mozilla-central/rev/1b2636e8517aa48422ed516affe4d28cb7fa220a/tools/profiler/public/GeckoProfiler.h#163

Thanks. I filed bug 1545386 for this particular addition. Looks like this bug turned more into a meta bug.

Matthew, I think I discovered the underlying problem why it is that slow, and will work on a patch on bug 1550702.