Patrik's suggestions:

We should discuss the default values. Suggestion:
quarantine-systems-for = 5 d
retry-gate-closed-for = 2 s
maximum-retries-in-window = 1
retry-window = 2 s
gate-invalid-addresses-for = 20 s
resend-interval = 2 s
transport-failure-detector.acceptable-heartbeat-pause = 20 s
watch-failure-detector.acceptable-heartbeat-pause = 10 s

I remember that we had some interplay between gate-invalid-addresses-for, retry-window and maximum-retries-in-window when testing at EC2 with long DNS timeouts. Note that current default for retry-window is 60s.

I think retry-gate-closed-for=0s is trying too hard, and something we should consider to remove.