Optimize DefaultUserDestinationResolver.resolveDestination() [SPR-15602]

[spring-projects-issues]

Christoph Dreis opened SPR-15602 and commented

Hey,

I noticed in some of my stresstests that DefaultUserDestinationResolver.resolveDestination() was popping up quite often as a hot method and found some possible improvements.

The attached PR restructures the code of DefaultUserDestinationResolver.resolveDestination() or more correctly DefaultUserDestinationResolver.parse() in the following ways:

• Allow parse() to be inlined by extracting the logic of parsing subscription messages and normal messages to dedicated private methods.
  • (Previous) DefaultUserDestinationResolver::parse (458 bytes) hot method too big
  • (After) DefaultUserDestinationResolver::parse (95 bytes) inline (hot)
• Make the sourceDestination of a message part of the ParseResult in order to avoid getting it twice in the chain
• Move the call to fetch the Principal of a Message only to the parts where it is needed
• ...

Please find the results of the JMH benchmarks attached. In a nutshell, we can see no regression (on the contrary) for messages of type SUBSCRIBE/UNSUBSCRIBE/MESSAGE and almost doubled performance for HEARTBEAT messages.

Let me know what you think.

Cheers,
Christoph

---

Affects: 4.3.8

Attachments:

• benchmark_results.txt (5.58 kB)

Referenced from: pull request #1439, and commits d3e3365, 0ef1623, 6aeb8ef

[spring-projects-issues]

Christoph Dreis commented

Thanks for accepting the PR. Is a backport to 4.3.x possible? Would be highly appreciated.

[spring-projects-issues]

Juergen Hoeller commented

Sure, from my perspective. Rossen Stoyanchev, could you do that backport please?

[spring-projects-issues]

Rossen Stoyanchev commented

This is now in 4.3.x as well.

[spring-projects-issues]

Christoph Dreis commented

Thanks a lot! :)