Autocompletex

Autocompletex is a low-latency plug-and-play autocomplete tool using Redis sorted set. Written in pure Elixir, it focuses on rapid prototyping using your existing stack: simply start a redis instance, and start_link a GenServer.

Currently, it provides these implementation:

• Google-like query prediction based on popularity. E.G. ne -> netflix, new york times
• Lexicographic. Sorted in alphabetical order (faster)

There are two ways to run it:

• Use it as a standalone microservice with HTTP connection.
• GenServer

Installation

Add the :autocompletex to your mix.exs file:

def deps do
  [{:autocompletex, "~> 0.1.0"}]
end

Then add it to applications:

defp application() do
  [applications: [:logger, :autocompletex]]
end

Then, run mix deps.get in your shell to fetch the new dependency.

Usage

Overview

Currently, two types of autocompletion are supported:

• Lexicographic
• Predictive

If you want to suggest another scheme, please post an issue.

There are 3 ways to run it.

1. Standalone HTTP service
2. Using a GenServer
3. Supervision tree

Manual

To start a GenServer manually:

# For Lexicographic:
{:ok, conn} = Redix.start_link
db = "testdb"
{:ok, worker} = Autocompletex.Lexicographic.start_link(conn, db, Autocompletex.Lexicographic)

# For Predictive:
{:ok, conn} = Redix.start_link
db_prefix = "autocompletex"
{:ok, worker} = Autocompletex.Predictive.start_link(conn, db_prefix, Autocompletex.Predictive)

Alternatively, you can use it in a supervision tree.

Add this to config.exs:

config :autocompletex,
  redis_host: "localhost",
  redis_port: 6379,
  redis_string: nil,
  http_server: true,
  http_port: 3000,
  debug: false, # runs :observer.start if true
  type: :lexicographic #:predictive

Then call

Autocompletex.Lexicographic.upsert(Autocompletex.Lexicographic, ["test", "example"])

If http_server is set to true, two http endpoints will be accessible at the designated http_port(default: 3000).

upsert   -> /add?term=Elixir
complete -> /complete?term=te

API

There are two functions: upsert and complete.

upsert/2 means insert or update. For Lexicographic, if a query is already inserted, it will do nothing. For Predictive, it will increment the score of the query.

complete/3 returns a list of matched results. It takes 3 parameters: pid, prefix, rangelen. rangelen is the number of results to be returned. Defaults to 50.

:ok = Autocompletex.Lexicographic.upsert(worker, ["test", "example"])
{:ok, val} == complete(worker, "te") # assert val == ["test"]

Misc

Import file into Redis

If you have a list of user-generated search queries, you can use a mix task to index and provision the redis instance.

Simply do:

mix autocompletex.import --filename [path/to/file] [--predictive]

Internals

For predictive autocompletion, this tool will create keys [dbname]:[prefixes] as sorted sets. For example, for dbname autocompletex, word test:

autocompletex:t
autocompletex:te
autocompletex:tes

For lexicographic autocompletion, under sorted set [dbname].

Docs

To be updated. In the meantime, I'm happy to answer questions in issues.