Improve performance of StompEncoder.encode() [SPR-14747]

[spring-projects-issues]

Christoph Dreis opened SPR-14747 and commented

Hey,

I just noticed a rather costly assertion in the websocket layer - more explicitly in StompEncoder.encode(). Sorry I didn't notice this earlier in order to bring this to 4.3.3.

It is very similar to #19191, but has a much bigger impact on websocket applications as its executed on basically every message apart from heartbeats. Apart from the JMH benchmarks below, which show a factor of ~1200, the Assert statement in question produced more than 60GB of heap pressure in just 20 seconds. Mostly coming from the string concatenation of course.

!stomp-encoder.jpg|thumbnail!

Benchmark	Mode	Cnt	Score	Error	Units
MyBenchmark.testNormal	thrpt	100	1639083,456	± 68418,140	ops/s
MyBenchmark.testEnhanced	thrpt	100	2087033509,372	± 21891731,313	ops/s

Much like the fix for #19191 I switched from Assert.notNull to a simple null check with an IllegalStateException.

While looking into StompEncoder I also moved one variable declaration a bit down, which was not needed in every-case. I hope you don't mind.

Best,
Christoph

---

Affects: 4.2.8, 4.3.3

Reference URL: #1185

Attachments:

• stomp-encoder.jpg (50.38 kB)

Issue Links:

• Improve performance of assertion in StompSubProtocolHandler [SPR-14624] #19191 Improve performance of assertion in StompSubProtocolHandler

Referenced from: commits f2e1e1b, 6c764f6, 02d83ce, 94753b5, a6b0b6e, 774e4c3, 6577faa

Backported to: 4.2.9

[spring-projects-issues]

Juergen Hoeller commented

Once again, a good catch! Merged now, to be backported to 4.3.4 and 4.2.9 tomorrow.

[spring-projects-issues]

Christoph Dreis commented

Thank you!