Cache MessagePack bytes rather than just the bytes by pierotibou · Pull Request #8175 · DataDog/dd-trace-dotnet

pierotibou · 2026-02-07T08:00:44Z

Summary of changes

In SpanMessagePackFormatter cache Serialized MessagePack bytes (thus with the header) instead of just the Bytes. Take it a step further and generate the byte arrays at compile time with source generators. Also remove AAS tags from MessagePackStringCache as they never change.

Reason for change

Mostly to make the code consistent with MessagePackStringCache where we were using that method. I chose that solution instead of impacting MessagePackStringCache as It's a micro optim.
Also the cache should contain only values that can change.

Implementation details

Moved AAS Tags from MessagePackStringCache to SpanMessagePackFormatter.
Use MessagePackSerializer.Serialize and WriteRaw instead of StringEncoding.UTF8.GetBytes and WriteStringBytes
Implement source generators.
There's one caveat that I duplicated the dotnet constant as TracerConstants is linked in dd-dotnet whereas sourcegenerators aren't. I didn't know if we want to have an almost empty shared project to share, or a extra dependency for dd_dotnet so I kept things simple for now.

Test coverage

Asked an agent to add a unit test that is testing all values to make sure I didn't forget one case. I didn't plan to commit it as I'm not sure it's valuable in other cases, but I did as a separate commit so it's easy to drop if we don't want to keep it.

Other details

Benchmark results (ran them 3 times as there was a good variance between runs, after was always better).

Before:

Method	Mean	Error	StdDev	Allocated
'Serialize simple span (minimal tags)'	293.7 ns	3.88 ns	3.44 ns	-
'Serialize span with some tags'	363.2 ns	4.36 ns	3.41 ns	-
'Serialize span with many tags'	942.5 ns	3.41 ns	3.02 ns	-

After:

Method	Mean	Error	StdDev	Allocated
'Serialize simple span (minimal tags)'	238.4 ns	3.69 ns	3.50 ns	-
'Serialize span with some tags'	316.0 ns	6.56 ns	6.43 ns	-
'Serialize span with many tags'	887.7 ns	17.71 ns	20.06 ns	-

ProcessTags are now computed lazily. They are also called in the relevant way from the settings.

…gePackFormatter Changed from StringEncoding.UTF8.GetBytes() + WriteStringBytes() to MessagePackSerializer.Serialize() + WriteRaw() for pre-encoded constant strings. Benchmark results show 4-16% performance improvement with zero allocations maintained: - Simple span: 267.9ns → 231.5ns (13.6% faster) - Span with tags: 353.6ns → 297.7ns (15.8% faster) - Span with many tags: 901.1ns → 863.4ns (4.2% faster) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 098a5dbbb6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

tracer/src/Datadog.Trace/Agent/MessagePack/SpanMessagePackFormatter.cs

pr-commenter · 2026-02-07T08:42:50Z

Benchmarks

Benchmark execution time: 2026-02-18 22:31:21

Comparing candidate commit c91512b in PR branch pierre/mpack with baseline commit e0e1029 in branch master.

Found 8 performance improvements and 16 performance regressions! Performance is the same for 156 metrics, 12 unstable metrics.

scenario:Benchmarks.Trace.AgentWriterBenchmark.WriteAndFlushEnrichedTraces net472

🟥 throughput [-43.260op/s; -41.587op/s] or [-7.372%; -7.087%]
🟩 execution_time [-12.705ms; -12.081ms] or [-5.824%; -5.538%]

scenario:Benchmarks.Trace.AgentWriterBenchmark.WriteAndFlushEnrichedTraces net6.0

🟩 execution_time [-28.906ms; -28.682ms] or [-24.021%; -23.835%]
🟩 throughput [+75.370op/s; +76.922op/s] or [+9.448%; +9.642%]

scenario:Benchmarks.Trace.AgentWriterBenchmark.WriteAndFlushEnrichedTraces netcoreapp3.1

🟥 execution_time [+85.544ms; +85.724ms] or [+77.566%; +77.729%]

scenario:Benchmarks.Trace.Asm.AppSecBodyBenchmark.AllCycleSimpleBody net6.0

🟩 execution_time [-27.805ms; -23.373ms] or [-12.686%; -10.664%]

scenario:Benchmarks.Trace.Asm.AppSecBodyBenchmark.AllCycleSimpleBody netcoreapp3.1

🟥 execution_time [+15.882ms; +21.799ms] or [+8.103%; +11.122%]

scenario:Benchmarks.Trace.Asm.AppSecEncoderBenchmark.EncodeLegacyArgs netcoreapp3.1

🟥 execution_time [+26.913ms; +27.744ms] or [+15.341%; +15.815%]

scenario:Benchmarks.Trace.CharSliceBenchmark.OptimizedCharSlice net6.0

🟩 execution_time [-138.963µs; -132.543µs] or [-8.827%; -8.419%]
🟩 throughput [+58.485op/s; +61.418op/s] or [+9.207%; +9.669%]

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearch net472

🟥 throughput [-43659.807op/s; -41776.156op/s] or [-13.704%; -13.113%]

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearch net6.0

🟥 throughput [-68560.593op/s; -62186.905op/s] or [-10.677%; -9.684%]

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearch netcoreapp3.1

🟥 throughput [-57584.030op/s; -49155.241op/s] or [-11.760%; -10.039%]

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearchAsync net472

🟥 throughput [-51330.243op/s; -48957.432op/s] or [-16.523%; -15.759%]

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearchAsync net6.0

🟥 throughput [-67280.039op/s; -55905.826op/s] or [-10.978%; -9.122%]

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearchAsync netcoreapp3.1

🟥 throughput [-42805.116op/s; -33440.044op/s] or [-9.036%; -7.059%]

scenario:Benchmarks.Trace.HttpClientBenchmark.SendAsync net472

🟥 throughput [-6302.197op/s; -5919.911op/s] or [-7.093%; -6.662%]

scenario:Benchmarks.Trace.HttpClientBenchmark.SendAsync netcoreapp3.1

🟥 throughput [-7505.753op/s; -6103.404op/s] or [-6.167%; -5.015%]

scenario:Benchmarks.Trace.RedisBenchmark.SendReceive net472

🟥 throughput [-36728.495op/s; -34317.708op/s] or [-10.384%; -9.703%]

scenario:Benchmarks.Trace.RedisBenchmark.SendReceive net6.0

🟥 throughput [-66277.662op/s; -54127.746op/s] or [-11.762%; -9.606%]

scenario:Benchmarks.Trace.RedisBenchmark.SendReceive netcoreapp3.1

🟥 throughput [-38908.410op/s; -29885.287op/s] or [-9.218%; -7.080%]

scenario:Benchmarks.Trace.SingleSpanAspNetCoreBenchmark.SingleSpanAspNetCore net6.0

🟩 execution_time [-36.215ms; -33.226ms] or [-18.335%; -16.823%]

scenario:Benchmarks.Trace.SpanBenchmark.StartFinishScope net6.0

🟥 execution_time [+15.020ms; +20.904ms] or [+7.665%; +10.668%]

scenario:Benchmarks.Trace.SpanBenchmark.StartFinishSpan netcoreapp3.1

🟩 execution_time [-16.085ms; -12.114ms] or [-7.484%; -5.637%]

dd-trace-dotnet-ci-bot · 2026-02-07T08:45:45Z

Execution-Time Benchmarks Report ⏱️

Execution-time results for samples comparing This PR (8175) and master.

⚠️ Potential regressions detected

HttpMessageHandler

Metric	Master (Mean ± 95% CI)	Current (Mean ± 95% CI)	Change	Status
.NET Framework 4.8 - CallTarget+Inlining+NGEN
duration	1155.35 ± (1155.24 - 1161.34) ms	1231.76 ± (1231.89 - 1238.73) ms	+6.6%	❌⬆️

Full Metrics Comparison

FakeDbCommand

Metric	Master (Mean ± 95% CI)	Current (Mean ± 95% CI)	Change	Status
.NET Framework 4.8 - Baseline
duration	68.20 ± (68.19 - 68.40) ms	68.19 ± (68.17 - 68.46) ms	-0.0%	✅
.NET Framework 4.8 - Bailout
duration	71.97 ± (71.88 - 72.15) ms	71.83 ± (71.76 - 72.03) ms	-0.2%	✅
.NET Framework 4.8 - CallTarget+Inlining+NGEN
duration	1029.67 ± (1032.55 - 1041.04) ms	1022.70 ± (1025.29 - 1033.24) ms	-0.7%	✅
.NET Core 3.1 - Baseline
process.internal_duration_ms	22.43 ± (22.38 - 22.48) ms	22.52 ± (22.49 - 22.55) ms	+0.4%	✅⬆️
process.time_to_main_ms	86.38 ± (86.19 - 86.56) ms	86.26 ± (86.11 - 86.41) ms	-0.1%	✅
runtime.dotnet.exceptions.count	0 ± (0 - 0)	0 ± (0 - 0)	+0.0%	✅
runtime.dotnet.mem.committed	15.50 ± (15.49 - 15.50) MB	15.51 ± (15.51 - 15.52) MB	+0.1%	✅⬆️
runtime.dotnet.threads.count	12 ± (12 - 12)	12 ± (12 - 12)	+0.0%	✅
.NET Core 3.1 - Bailout
process.internal_duration_ms	22.24 ± (22.22 - 22.27) ms	22.39 ± (22.37 - 22.42) ms	+0.7%	✅⬆️
process.time_to_main_ms	87.39 ± (87.29 - 87.48) ms	87.55 ± (87.43 - 87.68) ms	+0.2%	✅⬆️
runtime.dotnet.exceptions.count	0 ± (0 - 0)	0 ± (0 - 0)	+0.0%	✅
runtime.dotnet.mem.committed	15.53 ± (15.53 - 15.54) MB	15.56 ± (15.55 - 15.56) MB	+0.2%	✅⬆️
runtime.dotnet.threads.count	13 ± (13 - 13)	13 ± (13 - 13)	+0.0%	✅
.NET Core 3.1 - CallTarget+Inlining+NGEN
process.internal_duration_ms	258.30 ± (255.23 - 261.37) ms	245.67 ± (244.19 - 247.15) ms	-4.9%	✅
process.time_to_main_ms	504.18 ± (503.58 - 504.79) ms	524.97 ± (524.21 - 525.72) ms	+4.1%	✅⬆️
runtime.dotnet.exceptions.count	0 ± (0 - 0)	0 ± (0 - 0)	+0.0%	✅
runtime.dotnet.mem.committed	53.28 ± (53.26 - 53.31) MB	53.65 ± (53.62 - 53.68) MB	+0.7%	✅⬆️
runtime.dotnet.threads.count	28 ± (28 - 28)	28 ± (28 - 28)	+1.8%	✅⬆️
.NET 6 - Baseline
process.internal_duration_ms	20.99 ± (20.97 - 21.02) ms	20.94 ± (20.91 - 20.96) ms	-0.3%	✅
process.time_to_main_ms	74.34 ± (74.24 - 74.45) ms	74.42 ± (74.29 - 74.54) ms	+0.1%	✅⬆️
runtime.dotnet.exceptions.count	0 ± (0 - 0)	0 ± (0 - 0)	+0.0%	✅
runtime.dotnet.mem.committed	15.22 ± (15.22 - 15.22) MB	15.23 ± (15.23 - 15.24) MB	+0.1%	✅⬆️
runtime.dotnet.threads.count	10 ± (10 - 10)	10 ± (10 - 10)	+0.0%	✅
.NET 6 - Bailout
process.internal_duration_ms	20.95 ± (20.93 - 20.97) ms	20.89 ± (20.87 - 20.92) ms	-0.3%	✅
process.time_to_main_ms	75.40 ± (75.33 - 75.47) ms	75.68 ± (75.62 - 75.74) ms	+0.4%	✅⬆️
runtime.dotnet.exceptions.count	0 ± (0 - 0)	0 ± (0 - 0)	+0.0%	✅
runtime.dotnet.mem.committed	15.33 ± (15.33 - 15.33) MB	15.33 ± (15.33 - 15.34) MB	+0.0%	✅⬆️
runtime.dotnet.threads.count	11 ± (11 - 11)	11 ± (11 - 11)	+0.0%	✅
.NET 6 - CallTarget+Inlining+NGEN
process.internal_duration_ms	253.58 ± (252.73 - 254.44) ms	241.98 ± (240.84 - 243.11) ms	-4.6%	✅
process.time_to_main_ms	477.69 ± (477.11 - 478.27) ms	496.49 ± (496.00 - 496.98) ms	+3.9%	✅⬆️
runtime.dotnet.exceptions.count	0 ± (0 - 0)	0 ± (0 - 0)	+0.0%	✅
runtime.dotnet.mem.committed	53.97 ± (53.94 - 54.00) MB	54.52 ± (54.49 - 54.55) MB	+1.0%	✅⬆️
runtime.dotnet.threads.count	28 ± (28 - 28)	28 ± (28 - 28)	+0.3%	✅⬆️
.NET 8 - Baseline
process.internal_duration_ms	19.23 ± (19.20 - 19.25) ms	19.25 ± (19.22 - 19.28) ms	+0.1%	✅⬆️
process.time_to_main_ms	73.68 ± (73.57 - 73.78) ms	73.78 ± (73.68 - 73.89) ms	+0.1%	✅⬆️
runtime.dotnet.exceptions.count	0 ± (0 - 0)	0 ± (0 - 0)	+0.0%	✅
runtime.dotnet.mem.committed	12.27 ± (12.26 - 12.27) MB	12.26 ± (12.25 - 12.26) MB	-0.1%	✅
runtime.dotnet.threads.count	10 ± (10 - 10)	10 ± (10 - 10)	+0.0%	✅
.NET 8 - Bailout
process.internal_duration_ms	19.20 ± (19.17 - 19.22) ms	19.16 ± (19.14 - 19.18) ms	-0.2%	✅
process.time_to_main_ms	75.02 ± (74.95 - 75.09) ms	74.88 ± (74.81 - 74.94) ms	-0.2%	✅
runtime.dotnet.exceptions.count	0 ± (0 - 0)	0 ± (0 - 0)	+0.0%	✅
runtime.dotnet.mem.committed	12.31 ± (12.30 - 12.32) MB	12.30 ± (12.29 - 12.31) MB	-0.1%	✅
runtime.dotnet.threads.count	11 ± (11 - 11)	11 ± (11 - 11)	+0.0%	✅
.NET 8 - CallTarget+Inlining+NGEN
process.internal_duration_ms	183.35 ± (182.57 - 184.13) ms	168.30 ± (167.42 - 169.19) ms	-8.2%	✅
process.time_to_main_ms	462.07 ± (461.57 - 462.58) ms	478.12 ± (477.54 - 478.71) ms	+3.5%	✅⬆️
runtime.dotnet.exceptions.count	0 ± (0 - 0)	0 ± (0 - 0)	+0.0%	✅
runtime.dotnet.mem.committed	41.23 ± (41.21 - 41.26) MB	41.71 ± (41.67 - 41.74) MB	+1.2%	✅⬆️
runtime.dotnet.threads.count	27 ± (26 - 27)	27 ± (27 - 27)	+0.5%	✅⬆️

HttpMessageHandler

Metric	Master (Mean ± 95% CI)	Current (Mean ± 95% CI)	Change	Status
.NET Framework 4.8 - Baseline
duration	196.14 ± (195.88 - 196.85) ms	197.00 ± (196.95 - 197.83) ms	+0.4%	✅⬆️
.NET Framework 4.8 - Bailout
duration	208.26 ± (211.40 - 215.30) ms	199.96 ± (199.54 - 200.48) ms	-4.0%	✅
.NET Framework 4.8 - CallTarget+Inlining+NGEN
duration	1155.35 ± (1155.24 - 1161.34) ms	1231.76 ± (1231.89 - 1238.73) ms	+6.6%	❌⬆️
.NET Core 3.1 - Baseline
process.internal_duration_ms	195.88 ± (195.49 - 196.27) ms	195.31 ± (194.89 - 195.73) ms	-0.3%	✅
process.time_to_main_ms	90.88 ± (90.59 - 91.18) ms	90.59 ± (90.33 - 90.85) ms	-0.3%	✅
runtime.dotnet.exceptions.count	3 ± (3 - 3)	3 ± (3 - 3)	+0.0%	✅
runtime.dotnet.mem.committed	20.80 ± (20.78 - 20.82) MB	20.67 ± (20.65 - 20.69) MB	-0.6%	✅
runtime.dotnet.threads.count	20 ± (20 - 20)	20 ± (20 - 20)	+0.1%	✅⬆️
.NET Core 3.1 - Bailout
process.internal_duration_ms	195.65 ± (195.20 - 196.11) ms	194.26 ± (193.80 - 194.71) ms	-0.7%	✅
process.time_to_main_ms	92.04 ± (91.78 - 92.31) ms	91.75 ± (91.55 - 91.95) ms	-0.3%	✅
runtime.dotnet.exceptions.count	3 ± (3 - 3)	3 ± (3 - 3)	+0.0%	✅
runtime.dotnet.mem.committed	20.72 ± (20.69 - 20.74) MB	20.77 ± (20.75 - 20.80) MB	+0.3%	✅⬆️
runtime.dotnet.threads.count	21 ± (21 - 21)	21 ± (21 - 21)	-1.0%	✅
.NET Core 3.1 - CallTarget+Inlining+NGEN
process.internal_duration_ms	446.29 ± (443.81 - 448.76) ms	426.13 ± (424.24 - 428.02) ms	-4.5%	✅
process.time_to_main_ms	516.56 ± (515.83 - 517.30) ms	535.00 ± (534.13 - 535.86) ms	+3.6%	✅⬆️
runtime.dotnet.exceptions.count	3 ± (3 - 3)	3 ± (3 - 3)	+0.0%	✅
runtime.dotnet.mem.committed	63.45 ± (63.34 - 63.57) MB	63.33 ± (63.23 - 63.43) MB	-0.2%	✅
runtime.dotnet.threads.count	29 ± (29 - 29)	29 ± (29 - 29)	+0.2%	✅⬆️
.NET 6 - Baseline
process.internal_duration_ms	199.60 ± (199.09 - 200.11) ms	199.87 ± (199.39 - 200.36) ms	+0.1%	✅⬆️
process.time_to_main_ms	78.99 ± (78.73 - 79.24) ms	78.30 ± (78.14 - 78.45) ms	-0.9%	✅
runtime.dotnet.exceptions.count	4 ± (4 - 4)	4 ± (4 - 4)	+0.0%	✅
runtime.dotnet.mem.committed	20.95 ± (20.93 - 20.97) MB	20.87 ± (20.85 - 20.89) MB	-0.4%	✅
runtime.dotnet.threads.count	19 ± (19 - 19)	19 ± (19 - 19)	-0.1%	✅
.NET 6 - Bailout
process.internal_duration_ms	198.97 ± (198.53 - 199.40) ms	198.91 ± (198.45 - 199.37) ms	-0.0%	✅
process.time_to_main_ms	79.55 ± (79.35 - 79.76) ms	79.73 ± (79.56 - 79.90) ms	+0.2%	✅⬆️
runtime.dotnet.exceptions.count	4 ± (4 - 4)	4 ± (4 - 4)	+0.0%	✅
runtime.dotnet.mem.committed	21.00 ± (20.98 - 21.03) MB	21.02 ± (20.99 - 21.05) MB	+0.1%	✅⬆️
runtime.dotnet.threads.count	20 ± (20 - 21)	20 ± (20 - 20)	-0.1%	✅
.NET 6 - CallTarget+Inlining+NGEN
process.internal_duration_ms	466.46 ± (464.76 - 468.17) ms	479.50 ± (472.41 - 486.59) ms	+2.8%	✅⬆️
process.time_to_main_ms	491.33 ± (490.65 - 492.00) ms	508.99 ± (508.24 - 509.74) ms	+3.6%	✅⬆️
runtime.dotnet.exceptions.count	4 ± (4 - 4)	4 ± (4 - 4)	+0.0%	✅
runtime.dotnet.mem.committed	63.14 ± (63.04 - 63.25) MB	63.44 ± (63.34 - 63.54) MB	+0.5%	✅⬆️
runtime.dotnet.threads.count	30 ± (30 - 30)	30 ± (30 - 30)	+0.0%	✅
.NET 8 - Baseline
process.internal_duration_ms	197.37 ± (196.90 - 197.84) ms	195.75 ± (195.31 - 196.19) ms	-0.8%	✅
process.time_to_main_ms	77.92 ± (77.73 - 78.11) ms	77.61 ± (77.40 - 77.81) ms	-0.4%	✅
runtime.dotnet.exceptions.count	4 ± (4 - 4)	4 ± (4 - 4)	+0.0%	✅
runtime.dotnet.mem.committed	16.32 ± (16.30 - 16.34) MB	16.34 ± (16.31 - 16.36) MB	+0.1%	✅⬆️
runtime.dotnet.threads.count	19 ± (19 - 19)	19 ± (19 - 19)	-0.7%	✅
.NET 8 - Bailout
process.internal_duration_ms	196.62 ± (196.14 - 197.10) ms	196.73 ± (196.26 - 197.20) ms	+0.1%	✅⬆️
process.time_to_main_ms	79.21 ± (79.01 - 79.41) ms	78.70 ± (78.51 - 78.90) ms	-0.6%	✅
runtime.dotnet.exceptions.count	4 ± (4 - 4)	4 ± (4 - 4)	+0.0%	✅
runtime.dotnet.mem.committed	16.40 ± (16.38 - 16.42) MB	16.35 ± (16.33 - 16.38) MB	-0.3%	✅
runtime.dotnet.threads.count	20 ± (20 - 20)	20 ± (20 - 20)	-0.2%	✅
.NET 8 - CallTarget+Inlining+NGEN
process.internal_duration_ms	374.82 ± (373.35 - 376.28) ms	459.07 ± (457.30 - 460.84) ms	+22.5%	✅⬆️
process.time_to_main_ms	475.48 ± (474.87 - 476.09) ms	489.04 ± (488.08 - 490.00) ms	+2.9%	✅⬆️
runtime.dotnet.exceptions.count	4 ± (4 - 4)	4 ± (4 - 4)	+0.0%	✅
runtime.dotnet.mem.committed	52.99 ± (52.95 - 53.03) MB	55.34 ± (55.31 - 55.38) MB	+4.4%	✅⬆️
runtime.dotnet.threads.count	29 ± (29 - 29)	29 ± (29 - 29)	+1.3%	✅⬆️

Comparison explanation

Execution-time benchmarks measure the whole time it takes to execute a program, and are intended to measure the one-off costs. Cases where the execution time results for the PR are worse than latest master results are highlighted in **red**. The following thresholds were used for comparing the execution times:

Welch test with statistical test for significance of 5%
Only results indicating a difference greater than 5% and 5 ms are considered.

Note that these results are based on a single point-in-time result for each branch. For full results, see the dashboard.

Graphs show the p99 interval based on the mean and StdDev of the test run, as well as the mean value of the run (shown as a diamond below the graph).

Duration charts

FakeDbCommand (.NET Framework 4.8)

gantt
    title Execution time (ms) FakeDbCommand (.NET Framework 4.8)
    dateFormat  x
    axisFormat %Q
    todayMarker off
    section Baseline
    This PR (8175) - mean (68ms)  : 67, 70
    master - mean (68ms)  : 67, 70

    section Bailout
    This PR (8175) - mean (72ms)  : 71, 73
    master - mean (72ms)  : 71, 73

    section CallTarget+Inlining+NGEN
    This PR (8175) - mean (1,029ms)  : 971, 1088
    master - mean (1,037ms)  : 976, 1098

FakeDbCommand (.NET Core 3.1)

gantt
    title Execution time (ms) FakeDbCommand (.NET Core 3.1)
    dateFormat  x
    axisFormat %Q
    todayMarker off
    section Baseline
    This PR (8175) - mean (115ms)  : 111, 119
    master - mean (115ms)  : 112, 117

    section Bailout
    This PR (8175) - mean (116ms)  : 114, 117
    master - mean (115ms)  : 114, 116

    section CallTarget+Inlining+NGEN
    This PR (8175) - mean (806ms)  : 775, 838
    master - mean (796ms)  : 746, 846

FakeDbCommand (.NET 6)

gantt
    title Execution time (ms) FakeDbCommand (.NET 6)
    dateFormat  x
    axisFormat %Q
    todayMarker off
    section Baseline
    This PR (8175) - mean (101ms)  : 99, 103
    master - mean (101ms)  : 99, 103

    section Bailout
    This PR (8175) - mean (102ms)  : 101, 103
    master - mean (101ms)  : 100, 103

    section CallTarget+Inlining+NGEN
    This PR (8175) - mean (773ms)  : 741, 805
    master - mean (770ms)  : 744, 795

FakeDbCommand (.NET 8)

gantt
    title Execution time (ms) FakeDbCommand (.NET 8)
    dateFormat  x
    axisFormat %Q
    todayMarker off
    section Baseline
    This PR (8175) - mean (100ms)  : 98, 102
    master - mean (99ms)  : 97, 102

    section Bailout
    This PR (8175) - mean (100ms)  : 99, 102
    master - mean (101ms)  : 100, 102

    section CallTarget+Inlining+NGEN
    This PR (8175) - mean (679ms)  : 657, 700
    master - mean (681ms)  : 657, 704

HttpMessageHandler (.NET Framework 4.8)

gantt
    title Execution time (ms) HttpMessageHandler (.NET Framework 4.8)
    dateFormat  x
    axisFormat %Q
    todayMarker off
    section Baseline
    This PR (8175) - mean (197ms)  : 192, 202
    master - mean (196ms)  : 191, 202

    section Bailout
    This PR (8175) - mean (200ms)  : 195, 205
    master - mean (213ms)  : 183, 244

    section CallTarget+Inlining+NGEN
    This PR (8175) - mean (1,235ms)  : crit, 1186, 1284
    master - mean (1,158ms)  : 1114, 1203

HttpMessageHandler (.NET Core 3.1)

gantt
    title Execution time (ms) HttpMessageHandler (.NET Core 3.1)
    dateFormat  x
    axisFormat %Q
    todayMarker off
    section Baseline
    This PR (8175) - mean (295ms)  : 288, 302
    master - mean (296ms)  : 290, 302

    section Bailout
    This PR (8175) - mean (296ms)  : 290, 302
    master - mean (297ms)  : 290, 304

    section CallTarget+Inlining+NGEN
    This PR (8175) - mean (1,001ms)  : 961, 1040
    master - mean (1,001ms)  : 953, 1049

HttpMessageHandler (.NET 6)

gantt
    title Execution time (ms) HttpMessageHandler (.NET 6)
    dateFormat  x
    axisFormat %Q
    todayMarker off
    section Baseline
    This PR (8175) - mean (287ms)  : 279, 295
    master - mean (288ms)  : 278, 299

    section Bailout
    This PR (8175) - mean (288ms)  : 280, 296
    master - mean (288ms)  : 282, 293

    section CallTarget+Inlining+NGEN
    This PR (8175) - mean (1,026ms)  : 912, 1140
    master - mean (997ms)  : 948, 1046

HttpMessageHandler (.NET 8)

gantt
    title Execution time (ms) HttpMessageHandler (.NET 8)
    dateFormat  x
    axisFormat %Q
    todayMarker off
    section Baseline
    This PR (8175) - mean (284ms)  : 279, 290
    master - mean (287ms)  : 278, 296

    section Bailout
    This PR (8175) - mean (286ms)  : 281, 292
    master - mean (286ms)  : 279, 294

    section CallTarget+Inlining+NGEN
    This PR (8175) - mean (982ms)  : crit, 957, 1007
    master - mean (889ms)  : 859, 919

andrewlock

Seems reasonable to me, thanks!

andrewlock · 2026-02-09T09:24:24Z

tracer/src/Datadog.Trace/Agent/MessagePack/SpanMessagePackFormatter.cs

-        private readonly byte[] _parentIdBytes = StringEncoding.UTF8.GetBytes("parent_id");
-        private readonly byte[] _errorBytes = StringEncoding.UTF8.GetBytes("error");
-        private readonly byte[] _metaStructBytes = StringEncoding.UTF8.GetBytes("meta_struct");
+        private readonly byte[] _traceIdBytes = MessagePackSerializer.Serialize("trace_id");


There's no reason we can't do this at compile time, right, very similar to what we do for TagList bytes, by using a small source generator 🤔 Would remove a small amount of startup time theoretically... Also theoretically we could remove some of those allocations in .NET Core+ by using ReadOnlySpan<byte> too, but I'm not sure the messagepackserializer.WriteRaw would support that, so maybe not worth it...

I agree, I thought about it after reading a related comment on SLack. But I've shied away from it. Can give it a try through in a next PR though.

Done in 34f7afe

tracer/src/Datadog.Trace/Agent/MessagePack/SpanMessagePackFormatter.cs

tracer/test/Datadog.Trace.Tests/Agent/MessagePack/SpanMessagePackFormatterTests.cs

andrewlock · 2026-02-09T09:31:34Z

tracer/test/Datadog.Trace.Tests/Agent/MessagePack/SpanMessagePackFormatterTests.cs

+        span.Metrics[Datadog.Trace.Metrics.TopLevelSpan].Should().Be(1.0, "root span should have _dd.top_level=1.0");
+
+        // ===== Verify span fields that use cached names =====
+        span.Service.Should().NotBeNullOrEmpty("service name should be present");


tiny nit: probably over-asserting a bit in this test IMO - this test is liable to break in the future if we change/refactor things, but it's not too bad.

Ok I'll simplify it a bit before merging then. Thanks.

done in c91512b but I haven't simplified so much, I felt the rest was relevant.

pierotibou · 2026-02-09T10:53:09Z

I was surprised to see the degradation on the benchmark. After looking, IIUC those bench run the Sample app each time so it is actually possible to have a degradation of startup time and overall time (even though given the improvement I'm still surprised of). So TLDR, I can ignore this particular feedback even though it advocates for pushing this work at compile time. Right @andrewlock ?

andrewlock · 2026-02-09T12:01:53Z

I was surprised to see the degradation on the benchmark. After looking, IIUC those bench run the Sample app each time so it is actually possible to have a degradation of startup time and overall time (even though given the improvement I'm still surprised of). So TLDR, I can ignore this particular feedback even though it advocates for pushing this work at compile time. Right @andrewlock ?

Yeah, those are explicitly trying to look at the impact on startup time, by running a relatively short-lived app repeatedly. There is naturally some variance though (which has recently got worse even in the baseline scenario) and so given that the other .NET Framework app and all the other TFMs didn't see a significant hit, I would not be too worried about it I think.

To be safe, you could try re-running the execution time benchmarks to see if you get the same result. If so, maybe there's something else going on there we need to look into 🤷‍♂️

pierotibou · 2026-02-09T13:00:14Z

To be safe, you could try re-running the execution time benchmarks to see if you get the same result. If so, maybe there's something else going on there we need to look into 🤷‍♂️

It ran twice as I forced push. Not sure it was the same test, but IIRC it was indeed raising an issue on NET FX the first time as well. Anyway, this PR sits on top of the ProcessTags one, so maybe I'll have time to investigate compilation changes by the time I can merge this one

- Replicated what was done for TagsList but it seems weird to have the #if NETCOREAPP in a .net 6 generated file - There are still twice the type field, it's iso with what we had but I need to remove it as it's useless.

pierotibou and others added 4 commits February 6, 2026 21:36

Handle Service Name override in process tags

2ae61d9

Lazy init and other fixes

4aa5210

ProcessTags are now computed lazily. They are also called in the relevant way from the settings.

fix benchmarks as well

6167f88

pierotibou requested a review from a team as a code owner February 7, 2026 08:00

chatgpt-codex-connector bot reviewed Feb 7, 2026

View reviewed changes

tracer/src/Datadog.Trace/Agent/MessagePack/SpanMessagePackFormatter.cs Outdated Show resolved Hide resolved

pierotibou changed the title ~~Cache MessagePack strings rather than bytes~~ Cache MessagePack bytes rather than just the bytes Feb 7, 2026

pierotibou added 3 commits February 7, 2026 10:24

Remove from the cache the values that are actually immutable

ea9cc6b

revert some unrelated changes

d1378b6

add test

769c3e9

pierotibou force-pushed the pierre/mpack branch from 098a5db to 769c3e9 Compare February 7, 2026 09:35

pierotibou added area:tracer The core tracer library (Datadog.Trace, does not include OpenTracing, native code, or integrations) AI Generated Largely based on code generated by an AI or LLM. This label is the same across all dd-trace-* repos labels Feb 7, 2026

andrewlock approved these changes Feb 9, 2026

View reviewed changes

pierotibou force-pushed the pierre/tags branch from 48dd387 to 5d9a366 Compare February 18, 2026 09:20

pierotibou requested review from a team as code owners February 18, 2026 09:20

pierotibou added 2 commits February 18, 2026 22:33

Switch to compile time

34f7afe

- Replicated what was done for TagsList but it seems weird to have the #if NETCOREAPP in a .net 6 generated file - There are still twice the type field, it's iso with what we had but I need to remove it as it's useless.

simplify a tad the new utest

c91512b

This was referenced Feb 18, 2026

[DSM] Generate MessagePack values at compile time #8220

Open

[CIVIz] Generate MessagePack values at compile time #8221

Draft

Comments

Conversation

pierotibou commented Feb 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary of changes

Reason for change

Implementation details

Test coverage

Other details

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

pr-commenter bot commented Feb 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmarks

scenario:Benchmarks.Trace.AgentWriterBenchmark.WriteAndFlushEnrichedTraces net472

scenario:Benchmarks.Trace.AgentWriterBenchmark.WriteAndFlushEnrichedTraces net6.0

scenario:Benchmarks.Trace.AgentWriterBenchmark.WriteAndFlushEnrichedTraces netcoreapp3.1

scenario:Benchmarks.Trace.Asm.AppSecBodyBenchmark.AllCycleSimpleBody net6.0

scenario:Benchmarks.Trace.Asm.AppSecBodyBenchmark.AllCycleSimpleBody netcoreapp3.1

scenario:Benchmarks.Trace.Asm.AppSecEncoderBenchmark.EncodeLegacyArgs netcoreapp3.1

scenario:Benchmarks.Trace.CharSliceBenchmark.OptimizedCharSlice net6.0

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearch net472

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearch net6.0

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearch netcoreapp3.1

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearchAsync net472

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearchAsync net6.0

scenario:Benchmarks.Trace.ElasticsearchBenchmark.CallElasticsearchAsync netcoreapp3.1

scenario:Benchmarks.Trace.HttpClientBenchmark.SendAsync net472

scenario:Benchmarks.Trace.HttpClientBenchmark.SendAsync netcoreapp3.1

scenario:Benchmarks.Trace.RedisBenchmark.SendReceive net472

scenario:Benchmarks.Trace.RedisBenchmark.SendReceive net6.0

scenario:Benchmarks.Trace.RedisBenchmark.SendReceive netcoreapp3.1

scenario:Benchmarks.Trace.SingleSpanAspNetCoreBenchmark.SingleSpanAspNetCore net6.0

scenario:Benchmarks.Trace.SpanBenchmark.StartFinishScope net6.0

scenario:Benchmarks.Trace.SpanBenchmark.StartFinishSpan netcoreapp3.1

Uh oh!

dd-trace-dotnet-ci-bot bot commented Feb 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Execution-Time Benchmarks Report ⏱️

⚠️ Potential regressions detected

HttpMessageHandler

FakeDbCommand

HttpMessageHandler

Uh oh!

andrewlock left a comment

Choose a reason for hiding this comment

Uh oh!

andrewlock Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

pierotibou Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

pierotibou Feb 18, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

andrewlock Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

pierotibou Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

pierotibou Feb 18, 2026

Choose a reason for hiding this comment

Uh oh!

pierotibou commented Feb 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

andrewlock commented Feb 9, 2026

Uh oh!

pierotibou commented Feb 9, 2026

Uh oh!

Reviewers

Assignees

pierotibou commented Feb 7, 2026 •

edited

Loading

pr-commenter bot commented Feb 7, 2026 •

edited

Loading

dd-trace-dotnet-ci-bot bot commented Feb 7, 2026 •

edited

Loading

pierotibou commented Feb 9, 2026 •

edited

Loading