RFR: 8274276: Cache normalizedBase URL in URLClassPath.FileLoader

Daniel Fuchs dfuchs at openjdk.java.net
Fri Sep 24 11:16:51 UTC 2021


On Fri, 24 Sep 2021 10:35:54 GMT, Сергей Цыпанов <github.com+10835776+stsypanov at openjdk.org> wrote:

> Currently on each invocation of `URLClassPath.FileLoader.getResource()` `normalizedBase` URL is recalculated using same final `baseURL` from parent class. This can be avoided giving significant improvement in memory consumption. Consider the benchmark:
> 
> @State(Scope.Benchmark)
> @BenchmarkMode(Mode.AverageTime)
> @OutputTimeUnit(TimeUnit.NANOSECONDS)
> @Fork(jvmArgsAppend = {"-Xms2g", "-Xmx2g"})
> public class URLClassPathBenchmark {
>     private final ClassLoader classLoader = getClass().getClassLoader();
> 
>     @Benchmark
>     public URL getResource() {
>         return classLoader.getResource("file:./config/application.properties");
>     }
> }
> 
> it demonstrates improvement brought in by this patch:
> 
> before
> Benchmark                                                           Mode  Cnt     Score     Error   Units
> URLClassPathBenchmark.getResource                                   avgt   50  2840,746 ±  22,206   ns/op
> URLClassPathBenchmark.getResource:·gc.alloc.rate                    avgt   50   645,313 ±   8,072  MB/sec
> URLClassPathBenchmark.getResource:·gc.alloc.rate.norm               avgt   50  2403,258 ±  17,639    B/op
> URLClassPathBenchmark.getResource:·gc.churn.G1_Eden_Space           avgt   50   656,624 ± 116,090  MB/sec
> URLClassPathBenchmark.getResource:·gc.churn.G1_Eden_Space.norm      avgt   50  2450,175 ± 440,011    B/op
> URLClassPathBenchmark.getResource:·gc.churn.G1_Survivor_Space       avgt   50     0,123 ±   0,149  MB/sec
> URLClassPathBenchmark.getResource:·gc.churn.G1_Survivor_Space.norm  avgt   50     0,459 ±   0,556    B/op
> URLClassPathBenchmark.getResource:·gc.count                         avgt   50    67,000            counts
> URLClassPathBenchmark.getResource:·gc.time                          avgt   50   117,000                ms
> 
> after
> Benchmark                                                           Mode  Cnt     Score    Error   Units
> URLClassPathBenchmark.getResource                                   avgt  100  2596,719 ±  9,786   ns/op
> URLClassPathBenchmark.getResource:·gc.alloc.rate                    avgt  100   448,780 ±  1,684  MB/sec
> URLClassPathBenchmark.getResource:·gc.alloc.rate.norm               avgt  100  1528,040 ±  0,005    B/op
> URLClassPathBenchmark.getResource:·gc.churn.G1_Eden_Space           avgt  100   479,905 ± 23,369  MB/sec
> URLClassPathBenchmark.getResource:·gc.churn.G1_Eden_Space.norm      avgt  100  1634,314 ± 79,821    B/op
> URLClassPathBenchmark.getResource:·gc.churn.G1_Survivor_Space       avgt  100     0,101 ±  0,097  MB/sec
> URLClassPathBenchmark.getResource:·gc.churn.G1_Survivor_Space.norm  avgt  100     0,345 ±  0,329    B/op
> URLClassPathBenchmark.getResource:·gc.count                         avgt  100    98,000           counts
> URLClassPathBenchmark.getResource:·gc.time                          avgt  100   218,000               ms

This is calling an overidable method within a constructor, which might have unforeseen side effects. Can you  make `Loader::getBaseURL` final? That would remove the concern.

-------------

PR: https://git.openjdk.java.net/jdk/pull/5677


More information about the core-libs-dev mailing list