Regressions in Microsoft.Extensions.DependencyInjection.GetService

[DrewScoggins]

Run Information

Architecture	x64
OS	ubuntu 18.04
Baseline	a4b3772d1ada339220145843138d4c622134adeb
Compare	43b1f2c165257faae4d6a4aa886b701b63d2d5c0
Diff	Diff

Regressions in Microsoft.Extensions.DependencyInjection.GetService

Benchmark	Baseline	Test	Test/Base	Test Quality	Baseline IR	Compare IR	IR Ratio	Baseline ETL	Compare ETL
ServiceScopeProvider - Duration of single invocation	21.97 ns	38.43 ns	1.75	0.26
ServiceScope - Duration of single invocation	45.32 ns	60.61 ns	1.34	0.19

Historical Data in Reporting System

Repro

git clone https://github.com/dotnet/performance.git
python3 .\performance\scripts\benchmarks_ci.py -f netcoreapp5.0 --filter 'Microsoft.Extensions.DependencyInjection.GetService*'

Payloads

Baseline
Compare

Histogram

Microsoft.Extensions.DependencyInjection.GetService.ServiceScopeProvider

Microsoft.Extensions.DependencyInjection.GetService.ServiceScope

Docs

Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository

[ghost]

Tagging subscribers to this area: @eerhardt, @maryamariyan
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Run Information

Architecture	x64
OS	ubuntu 18.04
Baseline	a4b3772d1ada339220145843138d4c622134adeb
Compare	43b1f2c165257faae4d6a4aa886b701b63d2d5c0
Diff	Diff

Regressions in Microsoft.Extensions.DependencyInjection.GetService

Benchmark	Baseline	Test	Test/Base	Test Quality	Baseline IR	Compare IR	IR Ratio	Baseline ETL	Compare ETL
ServiceScopeProvider - Duration of single invocation	21.97 ns	38.43 ns	1.75	0.26
ServiceScope - Duration of single invocation	45.32 ns	60.61 ns	1.34	0.19

Historical Data in Reporting System

Repro

git clone https://github.com/dotnet/performance.git
python3 .\performance\scripts\benchmarks_ci.py -f netcoreapp5.0 --filter 'Microsoft.Extensions.DependencyInjection.GetService*'

Payloads

Baseline
Compare

Histogram

Microsoft.Extensions.DependencyInjection.GetService.ServiceScopeProvider

Microsoft.Extensions.DependencyInjection.GetService.ServiceScope

Docs

Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository

Author:	DrewScoggins
Assignees:	-
Labels:	arch-arm64, arch-x64, area-Extensions-DependencyInjection, os-linux, os-windows, tenet-performance, tenet-performance-benchmarks, untriaged
Milestone:	-

[eerhardt]

Looks like it is caused by #53325.

cc @maryamariyan @davidfowl @pakrym

[davidfowl]

bug fix > performance 😄

[pakrym]

bug fix > performance 😄

Sure, but did we expect a regression from that change? Is the regression caused by the if check? Can we make it cheaper?

[davidfowl]

I did yes, it's one extra branch. We should definitely see if we can make the check cheaper but I'm not sure it's worth redesigning anything over.

Maybe we can do a reference equals for the root scope check or stash a bool instead of service scope itself to avoid that comparison altogether

[pakrym]

We should try turning IsRootScope into a readonly field.

[davidfowl]

We should leave this open since the change didn't have much of an impact on this code path.

[davidfowl]

Looking at the actual benchmark(s) more deeply it seems like:

[Benchmark]
public IServiceScopeFactory ServiceScopeProvider() => _serviceScopeFactoryProvider.GetService<IServiceScopeFactory>();

[Benchmark]
public IServiceScope ServiceScope() => _serviceScope.CreateScope();

IServiceScopeFactory is a singleton so I wonder if this part of the change made thing worse. I would have thought that @pakrym's change here would have been a fine replacement for this but maybe avoiding the call to the runtime resolving when the value is cached is faster.

I don't get why CreateScope is slower.

[maryamariyan]

The latest charts show we gained back perf numbers after PR #54732:

So this issue can be safely closed.

[pakrym]

I think the regression is still there:

The avg is clearly higher.