RFR: 8353013: java.net.URI.create(String) may have low performance to scan the host/domain name from URI string when the hostname starts with number
Johannes Graham
duke at openjdk.org
Fri Apr 11 18:41:40 UTC 2025
On Fri, 28 Mar 2025 15:19:46 GMT, Rohitash <duke at openjdk.org> wrote:
> `scanByte` throws `NumberFormatException` for URIs that start with numbers, e.g., https://11111111.x.y/
> The current flow is `parseIPv4Address` → `scanIPv4Address` → `scanByte`. `parseIPv4Address` uses `NumberFormatException` for control flow, so it captures the exception, ignores it, and returns -1. This has been reported by AWS customer to cause low performance. Details: [JDK-8353013](https://bugs.openjdk.org/browse/JDK-8353013) & https://github.com/aws/aws-sdk-java-v2/issues/5933
>
> This PR avoids NumberFormatException by skipping calls to `Integer.parseInt` if the number of digits in the octet is > 3.
>
> I benchmarked on local machine for potential regressions.
> https://gist.github.com/rk-kmr/cb1a3d59225c17b180a29cc125ebf887
>
>
> Benchmark Mode Cnt Score Error Units
> URIBenchMark.newImplWithNormalUrl thrpt 10 0.004 ± 0.001 ops/ns
> URIBenchMark.newImplWithNumberlUrl thrpt 10 0.004 ± 0.001 ops/ns
> URIBenchMark.oldImplWithNormalUrl thrpt 10 0.004 ± 0.001 ops/ns
> URIBenchMark.oldImplWithNumUrl thrpt 10 0.001 ± 0.001 ops/ns
> URIBenchMark.newImplWithNormalUrl avgt 10 236.762 ± 8.700 ns/op
> URIBenchMark.newImplWithNumberlUrl avgt 10 264.017 ± 7.274 ns/op
> URIBenchMark.oldImplWithNormalUrl avgt 10 233.853 ± 6.539 ns/op
> URIBenchMark.oldImplWithNumUrl avgt 10 1183.572 ± 29.242 ns/op
>
>
> I ran following tests.
>
> make test-tier1
> make test-tier2
> make test TEST=jdk/java/net
src/java.base/share/classes/java/net/URI.java line 3438:
> 3436:
> 3437: // If no significant digits (all zeros), the value is 0
> 3438: if (significantDigitsNum == 0) return q;
Can avoid parseInt for short strings
Suggestion:
if (significantDigitsNum < 3) return q; // definitely < 255
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/24295#discussion_r2027378782
More information about the net-dev
mailing list