Performance regression between RC1 and latest snapshots with large amount of configuration properties

[dreis2211]

Hi,

after opening #21394 I thought I ran some benchmarks again on a project with a large amount of configuration properties. We've used https://github.com/martinvisser/config-properties-example in the past for this.

When doing so, I get the following results:

	RC1	Snapshot
	5.218	5.639
	5.737	5.917
	5.417	5.696
	5.439	5.739
	5.111	5.786
	5.133	5.854
	5.134	5.677
	5.335	6.680
	5.227	6.011
	5.250	5.778
Mean	5.300	5.878
Range	0.626	1.041

As you can see, the latest snapshot is about half a second slower than RC1. I'm not quite sure yet if this can be only traced back to the new hashCode calculation that was introduced in #20625, so I thought I better open an issue to investigate.

Cheers,
Christoph

[dreis2211]

After #21394 was merged I see roughly the same numbers, so it didn't really improve things.

[dreis2211]

Further investigation shows the following differences between RC1 and current snapshot builds:

This seems to be related to b4963d2 or the work done in #20625 or a combination of both. But I can't pinpoint it, yet.

[philwebb]

Thanks for spotting this.

[philwebb]

@dreis2211 I've pushed something that I think will make a difference. Could you try your sample app again and see if things have improved?

[dreis2211]

@philwebb We're good. It's even better than in 2.2.7 now. Thanks a lot.

	RC1	Snapshot
	5.641	4.242
	6.445	4.123
	5.592	4.270
	6.138	4.537
	5.471	4.185
	5.846	4.149
	6.517	4.206
	6.522	4.198
	5.442	4.187
	5.664	4.164
Mean	5.928	4.226
Range	1.080	0.414

	2.3.0	2.2.7
	4.314	5.196
	4.016	5.895
	4.195	6.808
	4.117	6.681
	4.229	6.490
	4.102	6.702
	4.080	6.766
	4.025	6.771
	4.084	6.563
	4.082	6.223
Mean	4.124	6.409
Range	0.298	1.612

[mbhave]

@dreis2211 Out of curiosity, did you do a local build to get the latest snapshot? Asking because I don't think repo.spring.io has this change yet due to a failed build.

[dreis2211]

I did, @mbhave

[mbhave]

thanks for confirming!