Ephemeral instances are only removed as whole resources, so there should
not be missing instances from an ephemeral resource. We can't error on
missing instances, because evaluation of checks may be pulling only a
single instance for evaluation before other sibling instances have been
opened.
There doesn't seem to be an objective reason for the 1 minute advance in
renew operation. If renewing requires some additional buffer, 1 minute
may not be correct and leaves provider guessing at terraform's behavior
based on experimentation. If additional time is needed around renewing
of ephemeral values, then the provider should take that into
consideration with their own expiration times.
Create a package for managing ephemeral resources and lifecycle.
Ephemeral resources have a very different lifecycle from managed
resources, and need to be managed more similarly to named values than
resources. The resources/ephemeral package is a separate database for
tracking ephemeral resource values, refreshing those resources as
required, and closing down the resources when they are no longer needed.
