Anders Pearson

For internal communication. pass around HMACs so nodes can always verify that the node they are talking to is legit.

If a node has not responded successfully for a set period of time, assume they are dead and remove them completely from the neighbors list. Currently, they stay in the...

When we announce to a neighbor, our full list of neighbors should get sent to them and their full list of neighbors should come to us. Both of us should...

stashing to nodes should happen in parallel (scatter-gather) and a response can be returned to the client as soon as MinReplication nodes have responded positively. Other stashes can happen in...

Currently, when a node gets a request for an image it doesn’t have, it will go through all its neighbors looking for a copy. In the case of a ‘true’...

any nodes in the first N results of the read-ring that are the same location as the current node will be placed before nodes in other locations.

It would mark itself as read-only, then walk its store of images and make sure they are each at the desired replication rate on other nodes if possible.

For gossiper/verifier. Ability to set it to 0 for testing, etc.

Option to write out/serialize config file with updated neighbor list. The use-case is that the cluster has evolved quite a bit from the original configuration and it would be good...

Probably should default to them. Can we support both?