schlosna opened a new pull request, #347:
URL: https://github.com/apache/httpcomponents-core/pull/347
Check input colon count before performing IPv6 regex validation.
`org.apache.hc.core5.net.InetAddressUtils.isIPv6Address` is used when
constructing the `host` header for requests, and currently allocates a regex
matcher when checking if the provided address is an IPv6 address. This is
generates excess garbage and CPU cycles, especially in environments with mostly
IPv4 addresses. By checking the colon count first, we see a 50x improvement
checking `InetAddressUtils.isIPv6Address` for an IPv4 address.
The specific call backtrace seen looks like:
```
void java.util.regex.Matcher.<init>(Pattern, CharSequence)
Matcher java.util.regex.Pattern.matcher(CharSequence)
boolean
org.apache.hc.core5.net.InetAddressUtils.isIPv6HexCompressedAddress(String)
boolean org.apache.hc.core5.net.InetAddressUtils.isIPv6Address(String)
void org.apache.hc.core5.net.Host.format(StringBuilder, NamedEndpoint)
void org.apache.hc.core5.net.URIAuthority.format(StringBuilder,
URIAuthority)
String org.apache.hc.core5.net.URIAuthority.format(URIAuthority)
String org.apache.hc.core5.net.URIAuthority.toString()
String java.util.Objects.toString(Object, String)
void org.apache.hc.core5.http.message.BasicHeader.<init>(String,
Object, boolean)
void org.apache.hc.core5.http.message.BasicHeader.<init>(String,
Object)
void
org.apache.hc.core5.http.message.BasicHttpRequest.addHeader(String, Object)
void
org.apache.hc.core5.http.protocol.RequestTargetHost.process(HttpRequest,
EntityDetails, HttpContext)
void
org.apache.hc.core5.http.protocol.DefaultHttpProcessor.process(HttpRequest,
EntityDetails, HttpContext)
ClassicHttpResponse
org.apache.hc.client5.http.impl.classic.ProtocolExec.execute(ClassicHttpRequest,
ExecChain$Scope, ExecChain)
ClassicHttpResponse
org.apache.hc.client5.http.impl.classic.ExecChainElement.execute(ClassicHttpRequest,
ExecChain$Scope)
ClassicHttpResponse
org.apache.hc.client5.http.impl.classic.ExecChainElement$1.proceed(ClassicHttpRequest,
ExecChain$Scope)
ClassicHttpResponse
org.apache.hc.client5.http.impl.classic.RedirectExec.execute(ClassicHttpRequest,
ExecChain$Scope, ExecChain)
ClassicHttpResponse
org.apache.hc.client5.http.impl.classic.ExecChainElement.execute(ClassicHttpRequest,
ExecChain$Scope)
CloseableHttpResponse
org.apache.hc.client5.http.impl.classic.InternalHttpClient.doExecute(HttpHost,
ClassicHttpRequest, HttpContext)
CloseableHttpResponse
org.apache.hc.client5.http.impl.classic.CloseableHttpClient.execute(ClassicHttpRequest,
HttpContext)
CloseableHttpResponse
org.apache.hc.client5.http.impl.classic.CloseableHttpClient.execute(ClassicHttpRequest)
```
I wrote some
[Ipv6FormatBenchmarks](https://github.com/schlosna/httpclient-format-performance/blob/develop/src/jmh/java/com/github/schlosna/Ipv6FormatBenchmarks.java)
showing original and improved:
From `JDK 11.0.13, OpenJDK 64-Bit Server VM, 11.0.13+8-LTS` on `Intel(R)
Xeon(R) Platinum 8259CL CPU @ 2.50GHz`:
```
Benchmark
(address) Mode Cnt Score Error Units
Ipv6FormatBenchmarks.isIPv6Address_original
127.0.0.1 avgt 2 395.333 ns/op
Ipv6FormatBenchmarks.isIPv6Address_original:·gc.alloc.rate.norm
127.0.0.1 avgt 2 440.000 B/op
Ipv6FormatBenchmarks.isIPv6Address_original
0:0:0:0:0:0:0:1 avgt 2 425.793 ns/op
Ipv6FormatBenchmarks.isIPv6Address_original:·gc.alloc.rate.norm
0:0:0:0:0:0:0:1 avgt 2 208.000 B/op
Ipv6FormatBenchmarks.isIPv6Address_improved
127.0.0.1 avgt 2 15.780 ns/op
Ipv6FormatBenchmarks.isIPv6Address_improved:·gc.alloc.rate.norm
127.0.0.1 avgt 2 ≈ 10⁻⁶ B/op
Ipv6FormatBenchmarks.isIPv6Address_improved
0:0:0:0:0:0:0:1 avgt 2 430.078 ns/op
Ipv6FormatBenchmarks.isIPv6Address_improved:·gc.alloc.rate.norm
0:0:0:0:0:0:0:1 avgt 2 208.000 B/op
```
Seen in JFR profiles before changes:


--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]