MIA (Modular Inference Accelerator) Rule Engine

MIA is a rule engine library for the Golang programming language. Inspired by the Drools for Java programming language, and MIA is done in a much lightweight manner for both I/O and process.

Like Drools, MIA has its own DSL for policy/rule drafting to use, and mainly focuses on compliance domain

Flow

• Input

• Output

DSL

Basic

• Scope represents the permission or access granted to the user

GRANT $scope AS live.stream:allow WHEN $subject.age_status = "adult";

• Action is an executable statement, which includes both call and parameters for upstream execution

EXEC $action AS account:ban WITH retention:180 WHEN $subject.age_status = "underage";

• Fact is the new status or attribute of the subject or object, which will be used in the next round of inference process

ADD $fact AS $subject.age_status:underage WHEN $subject.age < 14 AND ($subject.geo = "KR" OR $subject.geo = "TH" OR $subject.geo = "ID");

Extension

• Priority picks the decision with a higher priority if there are overlaps or conflicts

ADD $fact AS $subject.age_status:underage WHEN $subject.age < 14 AND ($subject.geo = "KR" OR $subject.geo = "TH" OR $subject.geo = "ID") PRIORITY 1;
ADD $fact AS $subject.age_status:minor WHEN $subject.age <18 PRIORITY 2;

• Default picks the decision to make when all conditions fail to meet

DEFAULT GRANT $scope AS account:normal;
GRANT $scope AS account:banned WHEN $subject.age_status = "underage";

AST

Take the following rules as the example to show the AST

DEFAULT account:normal
ADD $fact AS $subject.age_status:underage WHEN $subject.age < 14 AND ($subject.geo = "KR" OR $subject.geo = "TH" OR $subject.geo = "ID");
GRANT $scope AS account:banned WHEN $subject.age_status = "underage";

Engine

• ANTLR is used here as the lexer and parser to build the ASTs ( ANTLR 4 Documentation )
• Rete network is used here as the optimization algorithm for result inference ( Rete Algorithm )
• Drools-like implementation is adopted here for high performance of the engine ( Drools Rule Engine )

Rete Network

Take the following rules as the example to show the generated Rete network

ADD $fact AS $subject.age_status:underage WHEN $subject.age < 14 AND ($subject.geo = "KR" OR $subject.geo = "ID");
ADD $fact AS $subject.age_status:underage WHEN $subject.age < 13 AND $subject.geo = "US";
ADD $fact AS $subject.age_status:adult WHEN $subject.age >= 18;
GRANT $scope AS account:kids_mode WHEN $subject.age_status = "underage" AND $subject.geo = "US" PRIORITY 1;
GRANT $scope AS account:banned WHEN $subject.age_status = "underage" PRIORITY 2;

Usage

const POLICY = `
DEFAULT GRANT $scope AS account:normal; 
ADD $fact AS $subject.age_status:underage WHEN $subject.age < 14 AND ($subject.geo = "KR" OR $subject.geo = "ID");
ADD $fact AS $subject.age_status:underage WHEN $subject.age < 13 AND $subject.geo = "US";
ADD $fact AS $subject.age_status:adult WHEN $subject.age >= 18;
GRANT $scope AS account:kids_mode WHEN $subject.age_status = "underage" AND $subject.geo = "US" PRIORITY 1;
GRANT $scope AS account:banned WHEN $subject.age_status = "underage" PRIORITY 2;
`
e := engine.New(POLICY)
res := e.Eval("$subject.age", 13, "$subject.geo", "KR")
if res.Scopes["account"].Val() == "banned" {
    return false
}

Benchmark

For the case of 5 rules with 2 inference iterations (ex.)

goos: darwin
goarch: arm64
pkg: github.com/tiktok/mia-rule-engine/benchmark
cpu: Apple M1 Pro
BenchmarkEngine
BenchmarkEngine-10    	 1487048	       814.7 ns/op	    1216 B/op	      22 allocs/op
PASS