In clustered 2LC, TX1 on node1, TX2 on node2:

TX1 reads data from DB
TX2 updates data in DB (commits DB transaction and sends invalidation)
node1 receives invalidation, but there is nothing to be invalidated
TX1 PFERs data into cache

Result: stale data in cache.

Solution proposal: reads to DB are already preceded by registering pending put, which is invalidated if local remove occurs. We have to register callback in the cache for invalidations and apply the same policy.