See also T12071. Phacility environments, particularly repo hosts, do not define "E" in variables_order. This leads to an unnecessary subprocess execution during startup to repair it.

A somewhat-related issue is that neither XHProf or --trace can "see" this call. We probably can't do much about XHProf safely, but we could maintain a service call buffer of the last X calls and send them to any newly registered listener. This is indirectly visible in --trace when the first call starts at some positive number:

$ .bin/storage status --trace
>>> [7] (+0) <connect> 
...

The [7] means "six calls happened before --trace initialized".