1. change service name, which changes consumer group name. All events processed twice. Seems hard, potential side effects
2. deploy to staging, produce test message in eqiad kafka prefixed with DC name staging uses
3. remove enabled rule from prod changeprop. Pretty drastic but no real risk if we're careful because events will be spooled in kafka. We think - we're not sure, could go badly.
My vote is for #2, less risk, less big changes. We also have a version on staging already. #1 seems quite manual, #3 offers too much risk.
Currently if we turn on changeprop the only thing it'll do is send varnish purge events on an mw purge. Unlikely and a duplicate is not a major issue. Deploying to staging seems safe and doesn't risk causing issues like rebalances (not certain as to whether this is a real risk).