Leaderboard (rank table) implementation using ETS tables.

If available in Hex, the package can be installed by adding leaderboard to your list of dependencies in mix.exs:

def deps do
  [{:leaderboard, "~> 0.2"}]
end

First off, the leaderboard GenServer process must be started. Typically, it's started as a part of the supervision tree:

worker(Leaderboard, [Leaderboard.Test])

It requires table_name argument which is the name of the leaderboard. It must be an atom. The leaderboard tables shouldn't be started dynamically as the leaderboard names are atoms and they shouldn't be generated dynamically.

The leaderboard's API has functions for inserting, updating and deleting records. These are write functions and are serialised. Also, there are functions for reading records in specific order and/or with limited number of returned values:

Leaderboard.insert(Leaderboard.Test, 30, "foo")
Leaderboard.insert(Leaderboard.Test, 5, "bar")
Leaderboard.insert(Leaderboard.Test, 19, "baz")

# update "bar" to 10
Leaderboard.insert(Leaderboard.Test, 10, "bar")

# lookup the score of "bar"
Leaderboard.lookup(Leaderboard.Test, "bar")
#=> 10

# select top 2 records in ascending order
Leaderboard.select(Leaderboard.Test, :ascend, 2)
#=> [{10, "bar"}, {19, "baz"}]

# match all records that have score > 10 in descending order and return
# their keys
match_spec = [{{{:"$1",:"$2"}}, [{:">", :"$1", 10}], [:"$2"]}]
Leaderboard.match(Leaderboard.Test, match_spec, :descend)
#=> ["foo", "baz"]

# delete "foo" record
Leaderboard.delete(Leaderboard.Test, "foo")

The leaderboard is composed of a GenServer process and two ETS tables. The ETS key_table is of type :set:

The second ETS table called score_table is of type :ordered_set. It stores only keys without any values:

When a new record is inserted into the leaderboard, the record is inserted into both tables. All the writes are serialised via the GenServer process.

The ETS tables are :protected, so only the GenServer process that owns them can write. All the other processes are allowed just to read. Read operations are not serialised so they can be done in concurrent manner.

The score_table is :ordered_set. With the size of the table also increases the time needed for insert operation:

## LeaderboardBench
benchmark name                   iterations   average time
Insert to table of size 1000         200000   8.42 µs/op
Insert to table of size 10000        200000   9.37 µs/op
Insert to table of size 100000       200000   10.06 µs/op
Insert to table of size 1000000      100000   10.55 µs/op