rnewson / jlounge

WARNING
Very, very alpha.
Getting started;
1) mvn jetty:run
NOTES
shard management
The number of shards is preconfigured and unalterable (default is 1024).
all items are mapped to a shard by f(id % shard_count).
all shards are mapped to nodes via a consistent hash ring. The first N non-duplicate nodes from where the shard maps to 
are used to store the node.
However, writes are always permitted as long as at least one server is available. During failures or partitions, this 
might mean that an item is written to a server that does not ordinarily host the shard the item belongs to. In this 
case, a new replica of the shard is made there unilaterally and the item is stored. The node is responsible for sending 
this item back to one of the intended shard nodes as soon as it can.
read-write policies.
A read-write policy is the combination of two access policies;
Head policy - access is directed to the head of the replica list.
Tail policy - access is directed to the tail of the replica list.
Random policy - access is directed to a random replica.
Round Robin policy - access id directed to replicas in a cyclical order.
Useful combinations
read-tail,write-head - all reads represent fully replicated data.
read-head,write-head - low latency reads (data not yet replicated).
read-rr,write-head - high performance read, low chance of collision.
read-rr,write-rr - high performance read+write, high chance of collision.
Behavior by operation
function: get document
url form: GET https://localhost:8080/jlounge/<dbname>/<id>
behavior;
1) calculate shard = hash(id) % shard_count.
2) take next N non-duplicate hosts from shard map.
3) GET https://&lt;1st node>:5984/<dbname>-<shard>/<id>
4) If status line == 200, proxy full response to client.
5) if status line != 200 or no response, try the next host.
Any other error is returned to the client (a failure while writing to a server because it died, etc). The client must 
retry.
function: create document
url form: PUT https://localhost:8080/jlounge/&lt;dbname&gt;/&lt;id&gt; (with request body)
behavior;
1-2) as before
3) PUT https://&lt;1st node>:5984/<dbname>-<shard->/<id>
4) if status==201, we're good.
5) if status==409, return error to caller.
6) if status==404 or no response, try the next host.
NB: Must force db creation at some point.