|
Hello watchers of this ticket,
This has come up more often of late, both in this ticket and anecdotally, so there's clearly a general need. There's also some need for discussion to make sure we come up with something usable and maintainable. So please do chime in with any design thoughts here.
For starters I want to test three assertions:
-
there's value to reducing what facts are collected in at least two arenas: a) resource consumption on the agent, and b) resource consumption on the master (pdb, etc). I'm calling this out to highlight that a solution where, say, we collected all the facts but blacklisted what we report, would not be a good solution. We want to not even collect unwanted facts.
-
we'd prefer a blacklist to a whitelist (all comments here have been for a blacklist, but one idea on the puppet-users thread mentioned a whitelist/blacklist hybrid).
-
we could blacklist top-level structured facts (e.g. ec2, partitions, blockdevices) and don't need more fine-grained blacklisting (and a top-level structured fact being blacklisted should imply that any associated legacy flat facts are also blacklisted). I think this will be more practical to implement, esp if we want to honor the use case that we keep resource consumption on the agent low.
|