搜索排序插件语言的选择

es中支持用groovy定制开发排序，但实际使用时发现性能较差。

动态语言的好处是写代码时不用做类型转换，但用非jvm的脚本(js,lua)需要把很多对象复制、重构一遍，代价太大。用scala看起来可以对性能和易用性做个折中(还可能要做类型转换，但不需要声明类型，不需要写return，比Java稍好，性能相当)。以下为测试:

    private static void testGroovy() {
        GroovyScriptEngineService service = new GroovyScriptEngineService();
        String groovyScript = "priceLimit = doc['priceLimit'];\n" +
                "if (priceLimit == 0.0) { return 1; }\n" +
                "score = (doc['parValue'] + C.toFloat()) / priceLimit;\n" +
                "if (score < 1) { return score; } else { return 1; }";
        ;
        CompiledScript script = new CompiledScript(ScriptType.FILE, "groovy", "GROOVY", service.compile(groovyScript, null));
        Map<String, Object> params = new HashMap<String, Object>();
        Map<String, Object> docMap = new HashMap<String, Object>();
        params.put("doc", docMap);
        params.put("C", "13");

        // warmup
        System.err.println("Warming up...");
        for (int k = 0; k < 1000; ++k) {
            for (int i = 0; i < 30000; ++i) {
                params.put("_doc", i);
                docMap.put("priceLimit", Math.random() * 10000 - 5000);
                docMap.put("parValue", Math.random() * 10000 - 5000);
                LeafScoreScript scoreScript = service.executable(script, null, params);
                scoreScript.runAsDouble();
            }
        }

        // test
        System.err.println("Begin to test...");
        long old = System.currentTimeMillis();
        for (int k = 0; k < 1000; ++k) {
            for (int i = 0; i < 30000; ++i) {
                params.put("_doc", i);
                docMap.put("priceLimit", Math.random() * 10000 - 5000);
                docMap.put("parValue", Math.random() * 10000 - 5000);
                LeafScoreScript scoreScript = service.executable(script, null, params);
                scoreScript.runAsDouble();
            }
        }
        System.err.println("Groovy Elapsed Ticks: " + (System.currentTimeMillis() - old) + " ms.");
    }

    private static double calcScore(Object doc, Object C, Object _doc) {
        double priceLimit = (double)(((Map<String, Object>)doc).get("priceLimit"));
        if (priceLimit == 0.0) {
            return 1;
        }
        double score = ((double)((Map<String, Object>)doc).get("parValue") + Double.parseDouble(C.toString()) ) / priceLimit;
        if (score < 1) {
            return score;
        } else {
            return 1;
        }
    }

    private static void testJava() {
        Map<String, Object> params = new HashMap<String, Object>();
        Map<String, Object> docMap = new HashMap<String, Object>();
        params.put("doc", docMap);
        String C = "13";

        // warmup
        System.err.println("Warming up...");
        for (int k = 0; k < 1000; ++k) {
            for (int i = 0; i < 30000; ++i) {
                Integer _doc = i;
                docMap.put("priceLimit", Math.random() * 10000 - 5000);
                docMap.put("parValue", Math.random() * 10000 - 5000);
                calcScore(docMap, C, _doc);
            }
        }

        // test
        System.err.println("Begin to test...");
        long old = System.currentTimeMillis();
        for (int k = 0; k < 1000; ++k) {
            for (int i = 0; i < 30000; ++i) {
                Integer _doc = i;
                docMap.put("priceLimit", Math.random() * 10000 - 5000);
                docMap.put("parValue", Math.random() * 10000 - 5000);
                calcScore(docMap, C, _doc);
            }
        }
        System.err.println("Java Elapsed Ticks: " + (System.currentTimeMillis() - old) + " ms.");
    }

    private static void testScala() throws javax.script.ScriptException {
        Map<String, Object> params = new HashMap<String, Object>();
        Map<String, Object> docMap = new HashMap<String, Object>();
        params.put("doc", docMap);
        params.put("C", "13");
        ScriptEngine scriptEngine = new ScriptEngineManager().getEngineByName("scala");
        ((MutableSettings.BooleanSetting)(((IMain)scriptEngine).settings()
                .usejavacp())).value_$eq(true);
        scriptEngine.put("_doc", docMap);
        scriptEngine.put("_C", "13");
        String script = "val doc = _doc.asInstanceOf[java.util.Map[String, Object]]\n" +
                "val C = _C.toString\n" +

                "val priceLimit = doc.get(\"priceLimit\").asInstanceOf[Double]\n" +
                "if (priceLimit == 0.0) { 1 }\n" +
                "val score = (doc.get(\"parValue\").asInstanceOf[Double] + java.lang.Double.parseDouble(C.toString)) / priceLimit\n" +
                "Math.min(score, 1)";
        System.err.println(script);
        final javax.script.CompiledScript compiled = ((Compilable)scriptEngine).compile(script);

        // warmup
        System.err.println("Warming up...");
        for (int k = 0; k < 1000; ++k) {
            for (int i = 0; i < 30000; ++i) {
                params.put("_doc", i);
                docMap.put("priceLimit", Math.random() * 10000 - 5000);
                docMap.put("parValue", Math.random() * 10000 - 5000);
                compiled.eval();
            }
        }

        // test
        System.err.println("Begin to test...");
        long old = System.currentTimeMillis();
        for (int k = 0; k < 1000; ++k) {
            for (int i = 0; i < 30000; ++i) {
                params.put("_doc", i);
                docMap.put("priceLimit", Math.random() * 10000 - 5000);
                docMap.put("parValue", Math.random() * 10000 - 5000);
                compiled.eval();
            }
        }
        System.err.println("Scala Elapsed Ticks: " + (System.currentTimeMillis() - old) + " ms.");
    }

    public static void main(String[] args) throws javax.script.ScriptException {
        testScala();
        testJava();
        testGroovy();
    }

测试结果如下:

scala script比原生Java函数调用多20%的overhead,groovy则慢10倍。

用scala script engine需要增加依赖:

<dependency>
    <groupId>org.scala-lang</groupId>
    <artifactId>scala-library</artifactId>
    <version>2.11.7</version>
</dependency>

<dependency>
    <groupId>org.apache.clerezza.scala</groupId>
    <artifactId>script-engine</artifactId>
    <version>1.0.0</version>
</dependency>