Let's implement a proof-of-concept of sharding based on Virtual Buckets concept.

• Create vshard.router and vshard.storage module structure
• Use router.lua by @GeorgyKirichenko as a start point for vshard.router
• Implement replicaset configuration from New cluster configuration shard#38:

local router = require('vshard.router')
router.cfg { --will be moved to box.cfg() sometime
    sharding = {
           { uri = '10.0.0.21:3301', replicaset=1, master = true },
           { uri = '10.0.0.22:3301', replicaset=1 },

           { uri = '10.0.0.21:3302', replicaset=2, master = true },
           { uri = '10.0.0.22:3302', replicaset=2 },

           { uri = '10.0.0.21:3303', replicaset=3, master = true },
           { uri = '10.0.0.22:3303', replicaset=3 },
    },
}

• router: parse configuration into Lua hash { replicaset => [master, slave, ...] }, where master-slave are net.box objects. Use https://github.com/tarantool/shard/tree/dev for this code.
• storage: create table _vbucket to store bucked_id (the single field)
• storage: set box.cfg { read_only = false } if master is false
• router: on shard.cfg() connect to all servers and check for _vbucket table
• router: create routing table based on _vbucket contents from replicasets
• router: add a method shard.bootstrap() which will make initial redistribution of buckets over shards. The number of vbuckets can be hardcoded for now.
• storage: add a dispatcher function which will check bucket_id for every request and execute a Lua function with provided arguments

vshard.storage.dispatch = function(bucket_id, func, {arg1, arg2, ...}) end

• router: implement a function to call procedures by bucket_id:
  shard.call(key, func, {args}) => result
• Add tests
• Add CI