Optimize UserAgents

[schlosna]

Before this PR

Similar to #864 , JFRs showing significant allocations and CPU consumed due to com.palantir.conjure.java.api.config.service.UserAgents#isValidName and com.palantir.conjure.java.api.config.service.UserAgents#format during Dialogue request processing

After this PR

Some initial benchmark results on 16 core Intel(R) Xeon(R) Platinum 8259CL CPU @ 2.50GHz where regex is the previous implementation and parser is our hand rolled non-allocating version.

JDK 11.0.13, OpenJDK 64-Bit Server VM, 11.0.13+8-LTS
Benchmark                                                      (name)  Mode  Cnt    Score      Error   Units
UserAgentsBenchmark.parser                                    service  avgt    3    0.021 ±    0.002   us/op
UserAgentsBenchmark.parser:·gc.alloc.rate.norm                service  avgt    3   ≈ 10⁻⁵               B/op
UserAgentsBenchmark.parser                     valid-service-name-123  avgt    3    0.073 ±    0.005   us/op
UserAgentsBenchmark.parser:·gc.alloc.rate.norm valid-service-name-123  avgt    3   ≈ 10⁻⁵               B/op
UserAgentsBenchmark.regex                                     service  avgt    3    0.092 ±    0.055   us/op
UserAgentsBenchmark.regex:·gc.alloc.rate.norm                 service  avgt    3  128.000 ±    0.001    B/op
UserAgentsBenchmark.regex                      valid-service-name-123  avgt    3    0.299 ±    0.005   us/op
UserAgentsBenchmark.regex:·gc.alloc.rate.norm  valid-service-name-123  avgt    3  200.000 ±    0.001    B/op

JDK 17.0.4, OpenJDK 64-Bit Server VM, 17.0.4+8-LTS
Benchmark                                                      (name)  Mode  Cnt     Score      Error   Units
UserAgentsBenchmark.parser                                    service  avgt    3     0.015 ±    0.002   us/op
UserAgentsBenchmark.parser:·gc.alloc.rate.norm                service  avgt    3    ≈ 10⁻⁵               B/op
UserAgentsBenchmark.parser                     valid-service-name-123  avgt    3     0.030 ±    0.001   us/op
UserAgentsBenchmark.parser:·gc.alloc.rate.norm valid-service-name-123  avgt    3    ≈ 10⁻⁵               B/op
UserAgentsBenchmark.regex                                     service  avgt    3     0.078 ±    0.021   us/op
UserAgentsBenchmark.regex:·gc.alloc.rate.norm                 service  avgt    3   128.008 ±    0.043    B/op
UserAgentsBenchmark.regex                      valid-service-name-123  avgt    3     0.251 ±    0.017   us/op
UserAgentsBenchmark.regex:·gc.alloc.rate.norm  valid-service-name-123  avgt    3   200.013 ±    0.001    B/op

==COMMIT_MSG==
Optimize UserAgents

Optimize UserAgents.isValidName to use a hand-rolled implementation to validate user agent name to avoid
allocation and regex overhead.

Increase default UserAgents.format string builder buffer size and ensure capacity as we append user agents.
==COMMIT_MSG==

Possible downsides?

Complexity of maintaining is higher than simple regex

[changelog-app]

Generate changelog in changelog/@unreleased

Type
See change types. Select one:

• Feature
• Improvement
• Fix
• Break
• Deprecation
• Manual task
• Migration

Description

Optimize UserAgents

Optimize UserAgents.isValidName to use a hand-rolled implementation to validate user agent name to avoid
allocation and regex overhead.

Increase default UserAgents.format string builder buffer size and ensure capacity as we append user agents.

Check the box to generate changelog(s)

• Generate changelog entry

[iamdanfox]

Optimized code remains very readable. I'm always surprised at how expensive regexes are in java!

[schlosna]

Tweaked the implementation a bit more exactly follow regex and improved a bit more:

JDK 17.0.4, OpenJDK 64-Bit Server VM, 17.0.4+8-LTS
Benchmark                                                      (name)  Mode  Cnt     Score     Error   Units
UserAgentsBenchmark.parser                                    service  avgt    3     0.008 ±   0.001   us/op
UserAgentsBenchmark.parser:·gc.alloc.rate.norm                service  avgt    3    ≈ 10⁻⁵              B/op
UserAgentsBenchmark.parser                     valid-service-name-123  avgt    3     0.018 ±   0.002   us/op
UserAgentsBenchmark.parser:·gc.alloc.rate.norm valid-service-name-123  avgt    3    ≈ 10⁻⁵              B/op
UserAgentsBenchmark.regex                                     service  avgt    3     0.077 ±   0.006   us/op
UserAgentsBenchmark.regex:·gc.alloc.rate.norm                 service  avgt    3   128.009 ±   0.028    B/op
UserAgentsBenchmark.regex                      valid-service-name-123  avgt    3     0.251 ±   0.148   us/op
UserAgentsBenchmark.regex:·gc.alloc.rate.norm  valid-service-name-123  avgt    3   200.013 ±   0.008    B/op

[carterkozak]

Looks great, thanks!

[svc-autorelease]

Released 2.27.0