[OTLP] Use ConditionalWeakTable for cache

[martincostello]

#7303 (comment)

Changes

Follow-up to #7303 to use ConditionalWeakTable<K, V> instead of ConcurrentDictionary<K, V>.

@cijothomas does this resolve your concern in the other PR?

Benchmarks

While this PR degrades compared to the original improvements, it is still significantly faster than before the caching was added.

Run with:

dotnet run --project .\test\Benchmarks\ --framework net10.0 --configuration Release -- --memory --filter "*ProtobufOtlpResourceSerializerBenchmarks*" --runtimes net10.0 net462

Copilot Summaries

This PR. vs main

gh-7326 is a mixed bag on .NET 10.0 but a clear regression on .NET Framework 4.6.2. Allocations are unchanged at 0 B in every benchmark, so allocation deltas are 0 B throughout and allocation ratios vs. baseline are n/a because the baseline is zero.

Duration vs. baseline (gh-7326 / main)

Runtime	AttributeCount	Baseline	gh-7326	Delta	Ratio
.NET 10.0	0	0.9662 ns	0.3002 ns	-0.6660 ns	0.3107x
.NET 10.0	4	11.6641 ns	7.4083 ns	-4.2558 ns	0.6351x
.NET 10.0	8	6.8801 ns	8.5426 ns	+1.6625 ns	1.2416x
.NET 10.0	16	13.1289 ns	13.3450 ns	+0.2161 ns	1.0165x
.NET 10.0	32	13.7314 ns	18.8867 ns	+5.1553 ns	1.3754x
.NET Framework 4.6.2	0	26.9956 ns	14.9621 ns	-12.0335 ns	0.5542x
.NET Framework 4.6.2	4	26.7247 ns	44.2335 ns	+17.5088 ns	1.6552x
.NET Framework 4.6.2	8	26.5985 ns	44.0871 ns	+17.4886 ns	1.6575x
.NET Framework 4.6.2	16	26.9786 ns	43.7270 ns	+16.7484 ns	1.6208x
.NET Framework 4.6.2	32	28.5463 ns	45.9108 ns	+17.3645 ns	1.6083x

High-level summary

Runtime	Overall summary
.NET 10.0	Mixed; geomean ratio 0.81x vs. baseline, with wins at attribute counts 0 and 4, near parity at 16, and regressions at 8 and 32
.NET Framework 4.6.2	Regression overall; geomean ratio 1.32x vs. baseline, with only attribute count 0 faster and all other cases ~1.61x-1.66x slower

Allocations

Runtime	AttributeCount(s)	Baseline	gh-7326	Delta	Ratio
Both runtimes	all	0 B	0 B	0 B	n/a

This PR vs. Before 7303

gh-7326 is substantially faster than baseline in 9 of 10 cases, with unchanged allocations. The only regression is .NET Framework 4.6.2, AttributeCount=0, which is 1.40x the baseline duration. All allocation measurements are 0 B vs 0 B, so allocation ratio is 1.00x across the board.

Duration ratio = gh-7326 / baseline; lower than 1.00x is better.

Runtime	AttributeCount	Baseline	gh-7326	Delta	Ratio vs baseline
.NET 10.0	0	1.494 ns	0.3002 ns	-1.1938 ns	0.20x
.NET 10.0	4	85.817 ns	7.4083 ns	-78.4087 ns	0.09x
.NET 10.0	8	165.621 ns	8.5426 ns	-157.0784 ns	0.05x
.NET 10.0	16	332.061 ns	13.3450 ns	-318.7160 ns	0.04x
.NET 10.0	32	665.643 ns	18.8867 ns	-646.7563 ns	0.03x
.NET Framework 4.6.2	0	10.665 ns	14.9621 ns	+4.2971 ns	1.40x
.NET Framework 4.6.2	4	354.221 ns	44.2335 ns	-309.9875 ns	0.12x
.NET Framework 4.6.2	8	630.657 ns	44.0871 ns	-586.5699 ns	0.07x
.NET Framework 4.6.2	16	1,217.009 ns	43.7270 ns	-1,173.2820 ns	0.04x
.NET Framework 4.6.2	32	2,487.241 ns	45.9108 ns	-2,441.3302 ns	0.02x

Allocation summary

Runtime	AttributeCount(s)	Baseline	gh-7326	Ratio vs baseline
.NET 10.0	0, 4, 8, 16, 32	0 B	0 B	1.00x
.NET Framework 4.6.2	0, 4, 8, 16, 32	0 B	0 B	1.00x

Overall: .NET 10.0 improved across every case, and .NET Framework 4.6.2 improved sharply once AttributeCount >= 4.

Before 7303 2435127

BenchmarkDotNet v0.15.8, Windows 11 (10.0.26200.8457/25H2/2025Update/HudsonValley2)
13th Gen Intel Core i7-13700H 2.90GHz, 1 CPU, 20 logical and 14 physical cores
.NET SDK 10.0.300
  [Host]               : .NET 10.0.8 (10.0.8, 10.0.826.23019), X64 RyuJIT x86-64-v3
  .NET 10.0            : .NET 10.0.8 (10.0.8, 10.0.826.23019), X64 RyuJIT x86-64-v3
  .NET Framework 4.6.2 : .NET Framework 4.8.1 (4.8.9325.0), X64 RyuJIT VectorSize=256

Method	Job	Runtime	AttributeCount	Mean	Error	StdDev	Ratio	RatioSD	Allocated	Alloc Ratio
WriteResource	.NET 10.0	.NET 10.0	0	1.494 ns	0.0597 ns	0.0776 ns	1.00	0.07	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	0	10.665 ns	0.1284 ns	0.1072 ns	7.16	0.36	-	NA
WriteResource	.NET 10.0	.NET 10.0	4	85.817 ns	1.7249 ns	2.0534 ns	1.00	0.03	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	4	354.221 ns	5.5426 ns	5.1846 ns	4.13	0.11	-	NA
WriteResource	.NET 10.0	.NET 10.0	8	165.621 ns	3.3070 ns	3.0933 ns	1.00	0.03	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	8	630.657 ns	10.6901 ns	8.3462 ns	3.81	0.08	-	NA
WriteResource	.NET 10.0	.NET 10.0	16	332.061 ns	6.4818 ns	9.2960 ns	1.00	0.04	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	16	1,217.009 ns	18.5761 ns	15.5119 ns	3.67	0.11	-	NA
WriteResource	.NET 10.0	.NET 10.0	32	665.643 ns	10.6954 ns	8.3503 ns	1.00	0.02	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	32	2,487.241 ns	19.5676 ns	15.2771 ns	3.74	0.05	-	NA

main (34dcd84)

BenchmarkDotNet v0.15.8, Windows 11 (10.0.26200.8457/25H2/2025Update/HudsonValley2)
13th Gen Intel Core i7-13700H 2.90GHz, 1 CPU, 20 logical and 14 physical cores
.NET SDK 10.0.300
  [Host]               : .NET 10.0.8 (10.0.8, 10.0.826.23019), X64 RyuJIT x86-64-v3
  .NET 10.0            : .NET 10.0.8 (10.0.8, 10.0.826.23019), X64 RyuJIT x86-64-v3
  .NET Framework 4.6.2 : .NET Framework 4.8.1 (4.8.9325.0), X64 RyuJIT VectorSize=256

Method	Job	Runtime	AttributeCount	Mean	Error	StdDev	Median	Ratio	RatioSD	Allocated	Alloc Ratio
WriteResource	.NET 10.0	.NET 10.0	0	0.9662 ns	0.1688 ns	0.4977 ns	1.207 ns	1.48	1.37	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	0	26.9956 ns	0.2246 ns	0.1754 ns	26.920 ns	41.47	28.25	-	NA
WriteResource	.NET 10.0	.NET 10.0	4	11.6641 ns	1.9201 ns	5.6613 ns	12.461 ns	1.34	1.03	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	4	26.7247 ns	0.6484 ns	1.7308 ns	26.159 ns	3.06	1.68	-	NA
WriteResource	.NET 10.0	.NET 10.0	8	6.8801 ns	0.1150 ns	0.0961 ns	6.882 ns	1.00	0.02	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	8	26.5985 ns	0.6073 ns	1.7227 ns	25.862 ns	3.87	0.25	-	NA
WriteResource	.NET 10.0	.NET 10.0	16	13.1289 ns	1.7294 ns	5.0993 ns	9.840 ns	1.13	0.57	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	16	26.9786 ns	0.5636 ns	0.5535 ns	26.822 ns	2.33	0.71	-	NA
WriteResource	.NET 10.0	.NET 10.0	32	13.7314 ns	0.2236 ns	0.1867 ns	13.822 ns	1.00	0.02	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	32	28.5463 ns	0.2539 ns	0.2120 ns	28.541 ns	2.08	0.03	-	NA

This PR

BenchmarkDotNet v0.15.8, Windows 11 (10.0.26200.8457/25H2/2025Update/HudsonValley2)
13th Gen Intel Core i7-13700H 2.90GHz, 1 CPU, 20 logical and 14 physical cores
.NET SDK 10.0.300
  [Host]               : .NET 10.0.8 (10.0.8, 10.0.826.23019), X64 RyuJIT x86-64-v3
  .NET 10.0            : .NET 10.0.8 (10.0.8, 10.0.826.23019), X64 RyuJIT x86-64-v3
  .NET Framework 4.6.2 : .NET Framework 4.8.1 (4.8.9325.0), X64 RyuJIT VectorSize=256

Method	Job	Runtime	AttributeCount	Mean	Error	StdDev	Median	Ratio	RatioSD	Allocated	Alloc Ratio
WriteResource	.NET 10.0	.NET 10.0	0	0.3002 ns	0.0262 ns	0.0218 ns	0.3074 ns	1.01	0.10	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	0	14.9621 ns	0.1515 ns	0.1265 ns	14.9673 ns	50.10	3.66	-	NA
WriteResource	.NET 10.0	.NET 10.0	4	7.4083 ns	0.1776 ns	0.2490 ns	7.3286 ns	1.00	0.05	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	4	44.2335 ns	0.8414 ns	0.6569 ns	44.2327 ns	5.98	0.21	-	NA
WriteResource	.NET 10.0	.NET 10.0	8	8.5426 ns	0.1897 ns	0.4043 ns	8.3700 ns	1.00	0.06	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	8	44.0871 ns	0.9005 ns	0.8844 ns	43.8193 ns	5.17	0.25	-	NA
WriteResource	.NET 10.0	.NET 10.0	16	13.3450 ns	0.2273 ns	0.1774 ns	13.3795 ns	1.00	0.02	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	16	43.7270 ns	0.3424 ns	0.2673 ns	43.6669 ns	3.28	0.05	-	NA
WriteResource	.NET 10.0	.NET 10.0	32	18.8867 ns	0.2964 ns	0.2475 ns	18.8287 ns	1.00	0.02	-	NA
WriteResource	.NET Framework 4.6.2	.NET Framework 4.6.2	32	45.9108 ns	0.9248 ns	0.7723 ns	45.7333 ns	2.43	0.05	-	NA

Merge requirement checklist

• CONTRIBUTING guidelines followed (license requirements, nullable enabled, static analysis, etc.)
• Unit tests added/updated
• Appropriate CHANGELOG.md files updated for non-trivial changes
• Changes in public API reviewed (if applicable)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.90%. Comparing base (576fcc0) to head (5c20c87).
⚠️ Report is 1 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #7326      +/-   ##
==========================================
+ Coverage   89.86%   89.90%   +0.03%     
==========================================
  Files         276      276              
  Lines       14006    14007       +1     
==========================================
+ Hits        12587    12593       +6     
+ Misses       1419     1414       -5     

Flag	Coverage Δ
unittests-Project-Experimental	89.83% <100.00%> (+<0.01%)	⬆️
unittests-Project-Stable	89.82% <100.00%> (+0.04%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
...ation/Serializer/ProtobufOtlpResourceSerializer.cs	86.04% <100.00%> (+0.33%)	⬆️

... and 4 files with indirect coverage changes

[Copilot]

Pull request overview

Updates the OTLP protobuf resource serialization cache to avoid retaining Resource instances indefinitely by switching from a ConcurrentDictionary to a ConditionalWeakTable, allowing cached entries to be reclaimed when the Resource key is no longer referenced.

Changes:

• Replaced the resource-bytes cache from ConcurrentDictionary<Resource, byte[]> to ConditionalWeakTable<Resource, byte[]>.
• Added framework-conditional cache access (GetOrAdd on net10.0+, GetValue otherwise).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[cijothomas]

Thanks. If possible, lets add a test to validate that we don't hold up Resource, if this is easy to test.