On Fri, 31 Jan 2020 at 14:49, Bjoern Rabenstein <[EMAIL PROTECTED]> wrote:
This is the sticking point for me. Should we accept something if it's not
unlikely that either a) we're only going to break it later, as we don't use
it and it's in the way of some new feature/optimization or b) we don't
break it, and development of Prometheus is slowed or stalled? How would
this actually play out in future?
This is not theoretical, we've already had such things get into the TSDB
codebase with no non-obvious-semantics, no unitests, that are in the way of
development, but it'd be a bit impolite to just remove it.
We have had a few places where code structure/factoring has been tweaked to
be more amenable to external usage and that seems to have worked out fine
so far. Most recently for example one of our importers requested that
PromQL not depend on the config package as part of adding a new feature. I
suspect there's a few trivial accessors around that we don't use ourselves