[herf]

Their description of Lacquer sounds problematic to me.

The usual failure case for this is that a cache node goes down for awhile, misses some invalidation messages, then comes up and starts serving (stale) traffic again.

This case isn't so bad if you keep your TTL low, and have cheap page revalidation.

[rantfoil]

Ideally invalidation happens on an offline queue with retry support. Server down? Reschedule job and retry later.

We use Delayed Job at Posterous for things like this. Resque works too.