[Profiler] Add custom dl_iterate_phdr and use it in libunwind #5660

gleocadie · 2024-06-06T19:31:56Z

Summary of changes

Add a custom dl_iterate_phdr function which will be used by libunwind.

Reason for change

Currently, when unwind callstacks, libunwind uses dl_iterate_phdr to resolve instruction pointers into shared objects files.
We recently hit a deadlock case:

The thread to sample was allocating memory when interrupting by our sampling thread. This means that the malloc lock was acquired. When libunwind started to unwind the callstack, the execution got blocked on a lock in dl_iterate_phdr. This lock is shared amongst other functions (dlopen, dlclose, dladdr...)
At the same time, another thread acquired the dlopen & co lock, started loading a library (using dlopen) and was trying to allocate memory, but got blocked on the malloc lock.

Implementation details

Create a cache that will be filled by struct dl_phdr_info* objects. Update the cache if needed every time the sampling thread is about to interrupt a thread.
Wrap dlclose & dlopen to know if a change occured in the loaded libraries set.
Implement a custom dl_iterate_phdr function which will use the cache and instruct libunwind to use it.

Test coverage

Other details

datadog-ddstaging · 2024-06-06T19:49:08Z

Datadog Report

Branch report: gleocadie/add-custom-dl_iterate_phdr
Commit report: 4a50889
Test service: dd-trace-dotnet

✅ 0 Failed, 351666 Passed, 1929 Skipped, 30h 20m 17.74s Total Time
❄️ 2 New Flaky
⌛ 3 Performance Regressions

New Flaky Tests (2)

CheckSmokeForOldWayToStackWalk - Datadog.Profiler.SmokeTests.ExceptionGeneratorTest - Last Failure
Expand for error
```
 The test Samples.ExceptionGenerator is running for too long or was lost
```
NoExceptions - Datadog.Trace.ClrProfiler.IntegrationTests.SmokeTests.AssemblyLoadContextRedirectSmokeTest - Last Failure
Expand for error
```
 Expected no errors in smoke test: ptrace(ATTACH, 13745) FAILED Operation not permitted
 
 Expected: True
 Actual:   False
```

⌛ Performance Regressions vs Default Branch (3)

CallTarget+Inlining+NGEN - Samples.HttpMessageHandler.windows.net60.json.scenarios 962.79ms (+106.73ms, +12%) - Details
CallTarget+Inlining+NGEN - Samples.HttpMessageHandler.windows.netcoreapp31.json.scenarios 950.8ms (+80.82ms, +9%) - Details
Baseline - Samples.HttpMessageHandler.windows.net60.json.scenarios 290.25ms (+24.68ms, +9%) - Details

andrewlock · 2024-06-06T19:55:56Z

Execution-Time Benchmarks Report ⏱️

Execution-time results for samples comparing the following branches/commits:

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 shown 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).

gantt
    title Execution time (ms) FakeDbCommand (.NET Framework 4.6.2) 
    dateFormat  X
    axisFormat %s
    todayMarker off
    section Baseline
    This PR (5660) - mean (75ms)  : 63, 86
     .   : milestone, 75,
    master - mean (74ms)  : 65, 83
     .   : milestone, 74,

    section CallTarget+Inlining+NGEN
    This PR (5660) - mean (1,006ms)  : 964, 1048
     .   : milestone, 1006,
    master - mean (991ms)  : 973, 1009
     .   : milestone, 991,

gantt
    title Execution time (ms) FakeDbCommand (.NET Core 3.1) 
    dateFormat  X
    axisFormat %s
    todayMarker off
    section Baseline
    This PR (5660) - mean (111ms)  : 108, 114
     .   : milestone, 111,
    master - mean (110ms)  : 107, 113
     .   : milestone, 110,

    section CallTarget+Inlining+NGEN
    This PR (5660) - mean (734ms)  : 694, 775
     .   : milestone, 734,
    master - mean (698ms)  : 675, 722
     .   : milestone, 698,

gantt
    title Execution time (ms) FakeDbCommand (.NET 6) 
    dateFormat  X
    axisFormat %s
    todayMarker off
    section Baseline
    This PR (5660) - mean (99ms)  : 97, 101
     .   : milestone, 99,
    master - mean (93ms)  : 90, 96
     .   : milestone, 93,

    section CallTarget+Inlining+NGEN
    This PR (5660) - mean (678ms)  : 655, 701
     .   : milestone, 678,
    master - mean (657ms)  : 630, 683
     .   : milestone, 657,

gantt
    title Execution time (ms) HttpMessageHandler (.NET Framework 4.6.2) 
    dateFormat  X
    axisFormat %s
    todayMarker off
    section Baseline
    This PR (5660) - mean (192ms)  : 187, 197
     .   : milestone, 192,
    master - mean (192ms)  : 188, 196
     .   : milestone, 192,

    section CallTarget+Inlining+NGEN
    This PR (5660) - mean (1,089ms)  : 1060, 1118
     .   : milestone, 1089,
    master - mean (1,081ms)  : 1059, 1103
     .   : milestone, 1081,

gantt
    title Execution time (ms) HttpMessageHandler (.NET Core 3.1) 
    dateFormat  X
    axisFormat %s
    todayMarker off
    section Baseline
    This PR (5660) - mean (277ms)  : 272, 282
     .   : milestone, 277,
    master - mean (276ms)  : 271, 281
     .   : milestone, 276,

    section CallTarget+Inlining+NGEN
    This PR (5660) - mean (876ms)  : 852, 900
     .   : milestone, 876,
    master - mean (874ms)  : 849, 899
     .   : milestone, 874,

gantt
    title Execution time (ms) HttpMessageHandler (.NET 6) 
    dateFormat  X
    axisFormat %s
    todayMarker off
    section Baseline
    This PR (5660) - mean (267ms)  : 263, 271
     .   : milestone, 267,
    master - mean (269ms)  : 263, 275
     .   : milestone, 269,

    section CallTarget+Inlining+NGEN
    This PR (5660) - mean (863ms)  : 838, 887
     .   : milestone, 863,
    master - mean (867ms)  : 839, 894
     .   : milestone, 867,

andrewlock · 2024-06-06T20:18:39Z

Throughput/Crank Report:zap:

Throughput results for AspNetCoreSimpleController comparing the following branches/commits:

Cases where throughput results for the PR are worse than latest master (5% drop or greater), results are shown in red.

Note that these results are based on a single point-in-time result for each branch. For full results, see one of the many, many dashboards!

gantt
    title Throughput Linux x64 (Total requests) 
    dateFormat  X
    axisFormat %s
    section Baseline
    This PR (5660) (11.947M)   : 0, 11947217
    master (11.753M)   : 0, 11753412
    benchmarks/2.9.0 (11.542M)   : 0, 11542126

    section Automatic
    This PR (5660) (8.110M)   : 0, 8109706
    master (7.853M)   : 0, 7853279
    benchmarks/2.9.0 (8.263M)   : 0, 8262905

    section Trace stats
    master (8.259M)   : 0, 8258966

    section Manual
    This PR (5660) (10.287M)   : 0, 10287299
    master (9.899M)   : 0, 9899401

    section Manual + Automatic
    This PR (5660) (7.681M)   : 0, 7680864
    master (7.433M)   : 0, 7432897

    section Version Conflict
    master (6.731M)   : 0, 6730574

gantt
    title Throughput Linux arm64 (Total requests) 
    dateFormat  X
    axisFormat %s
    section Baseline
    This PR (5660) (9.343M)   : 0, 9342723
    master (9.574M)   : 0, 9574121
    benchmarks/2.9.0 (9.596M)   : 0, 9596140

    section Automatic
    This PR (5660) (6.500M)   : 0, 6499572
    master (6.671M)   : 0, 6671192

    section Trace stats
    master (6.990M)   : 0, 6990244

    section Manual
    This PR (5660) (8.187M)   : 0, 8186573
    master (8.307M)   : 0, 8306500

    section Manual + Automatic
    This PR (5660) (6.210M)   : 0, 6209609
    master (6.251M)   : 0, 6250704

    section Version Conflict
    master (5.762M)   : 0, 5761603

gantt
    title Throughput Windows x64 (Total requests) 
    dateFormat  X
    axisFormat %s
    section Baseline
    This PR (5660) (9.682M)   : 0, 9682346
    master (10.202M)   : 0, 10201759
    benchmarks/2.9.0 (10.213M)   : 0, 10213239

    section Automatic
    This PR (5660) (6.911M)   : 0, 6910573
    master (7.212M)   : 0, 7212394
    benchmarks/2.9.0 (7.482M)   : 0, 7482023

    section Trace stats
    master (7.549M)   : 0, 7549360

    section Manual
    This PR (5660) (8.799M)   : 0, 8798971
    master (9.013M)   : 0, 9012822

    section Manual + Automatic
    This PR (5660) (6.663M)   : crit ,0, 6663489
    master (7.018M)   : 0, 7017920

    section Version Conflict
    master (6.365M)   : 0, 6365432

andrewlock · 2024-06-06T20:20:38Z

Benchmarks Report for tracer 🐌

Benchmarks for #5660 compared to master:

2 benchmarks are faster, with geometric mean 1.136
1 benchmarks have fewer allocations

The following thresholds were used for comparing the benchmark speeds:

Mann–Whitney U test with statistical test for significance of 5%
Only results indicating a difference greater than 10% and 0.3 ns are considered.

Allocation changes below 0.5% are ignored.

Benchmark details

Benchmarks.Trace.ActivityBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Gen 1	Gen 2	Allocated
master	`StartStopWithChild`	net6.0	7.78μs	43.6ns	279ns	0.0154	0.00768	0	5.42 KB
master	`StartStopWithChild`	netcoreapp3.1	9.78μs	52.8ns	280ns	0.0241	0.00963	0	5.62 KB
master	`StartStopWithChild`	net472	16μs	49ns	190ns	1.01	0.293	0.0952	6.07 KB
#5660	`StartStopWithChild`	net6.0	7.72μs	43.7ns	293ns	0.0152	0.0076	0	5.42 KB
#5660	`StartStopWithChild`	netcoreapp3.1	10μs	56.4ns	374ns	0.0246	0.00983	0	5.61 KB
#5660	`StartStopWithChild`	net472	16.2μs	65ns	252ns	1.04	0.311	0.0956	6.07 KB

Benchmarks.Trace.AgentWriterBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Allocated
master	`WriteAndFlushEnrichedTraces`	net6.0	489μs	145ns	563ns	0	2.7 KB
master	`WriteAndFlushEnrichedTraces`	netcoreapp3.1	640μs	294ns	1.14μs	0	2.7 KB
master	`WriteAndFlushEnrichedTraces`	net472	838μs	306ns	1.19μs	0.414	3.3 KB
#5660	`WriteAndFlushEnrichedTraces`	net6.0	473μs	77ns	288ns	0	2.7 KB
#5660	`WriteAndFlushEnrichedTraces`	netcoreapp3.1	638μs	204ns	791ns	0	2.7 KB
#5660	`WriteAndFlushEnrichedTraces`	net472	830μs	155ns	560ns	0.414	3.3 KB

Benchmarks.Trace.AspNetCoreBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Allocated
master	`SendRequest`	net6.0	170μs	229ns	886ns	0.168	18.44 KB
master	`SendRequest`	netcoreapp3.1	191μs	299ns	1.16μs	0.192	20.6 KB
master	`SendRequest`	net472	0.000295ns	0.000179ns	0.000692ns	0	0 b
#5660	`SendRequest`	net6.0	169μs	173ns	669ns	0.169	18.44 KB
#5660	`SendRequest`	netcoreapp3.1	192μs	195ns	754ns	0.191	20.6 KB
#5660	`SendRequest`	net472	0.000353ns	0.000163ns	0.000609ns	0	0 b

Benchmarks.Trace.CIVisibilityProtocolWriterBenchmark - Same speed ✔️ Fewer allocations 🎉

Fewer allocations 🎉 in #5660

Benchmark	Base Allocated	Diff Allocated	Change	Change %
Benchmarks.Trace.CIVisibilityProtocolWriterBenchmark.WriteAndFlushEnrichedTraces‑netcoreapp3.1	41.89 KB	41.53 KB	-351 B	-0.84%

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Gen 1	Gen 2	Allocated
master	`WriteAndFlushEnrichedTraces`	net6.0	547μs	312ns	1.12μs	0.543	0	0	41.65 KB
master	`WriteAndFlushEnrichedTraces`	netcoreapp3.1	664μs	946ns	3.67μs	0.332	0	0	41.89 KB
master	`WriteAndFlushEnrichedTraces`	net472	852μs	3.28μs	12.7μs	8.36	2.64	0.44	53.25 KB
#5660	`WriteAndFlushEnrichedTraces`	net6.0	540μs	302ns	1.13μs	0.558	0	0	41.45 KB
#5660	`WriteAndFlushEnrichedTraces`	netcoreapp3.1	646μs	1.57μs	5.87μs	0.324	0	0	41.53 KB
#5660	`WriteAndFlushEnrichedTraces`	net472	862μs	4.3μs	18.3μs	8.04	2.23	0.446	53.23 KB

Benchmarks.Trace.DbCommandBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Allocated
master	`ExecuteNonQuery`	net6.0	1μs	0.463ns	1.79ns	0.011	808 B
master	`ExecuteNonQuery`	netcoreapp3.1	1.48μs	0.933ns	3.61ns	0.011	808 B
master	`ExecuteNonQuery`	net472	1.74μs	0.748ns	2.9ns	0.122	770 B
#5660	`ExecuteNonQuery`	net6.0	1.07μs	0.795ns	3.08ns	0.0112	808 B
#5660	`ExecuteNonQuery`	netcoreapp3.1	1.43μs	0.442ns	1.65ns	0.0108	808 B
#5660	`ExecuteNonQuery`	net472	1.74μs	0.752ns	2.71ns	0.122	770 B

Benchmarks.Trace.ElasticsearchBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Gen 1	Allocated
master	`CallElasticsearch`	net6.0	1.23μs	0.499ns	1.93ns	0.0141	0	976 B
master	`CallElasticsearch`	netcoreapp3.1	1.48μs	0.716ns	2.58ns	0.0134	0	976 B
master	`CallElasticsearch`	net472	2.43μs	1.49ns	5.77ns	0.157	0.00121	995 B
master	`CallElasticsearchAsync`	net6.0	1.23μs	1.16ns	4.34ns	0.0131	0	952 B
master	`CallElasticsearchAsync`	netcoreapp3.1	1.62μs	0.763ns	2.85ns	0.0138	0	1.02 KB
master	`CallElasticsearchAsync`	net472	2.65μs	1.89ns	7.33ns	0.167	0.00132	1.05 KB
#5660	`CallElasticsearch`	net6.0	1.27μs	1.53ns	5.91ns	0.0133	0	976 B
#5660	`CallElasticsearch`	netcoreapp3.1	1.56μs	1.55ns	5.79ns	0.0132	0	976 B
#5660	`CallElasticsearch`	net472	2.5μs	2.06ns	7.71ns	0.157	0.00125	995 B
#5660	`CallElasticsearchAsync`	net6.0	1.26μs	0.569ns	2.13ns	0.0132	0	952 B
#5660	`CallElasticsearchAsync`	netcoreapp3.1	1.58μs	1.63ns	5.87ns	0.0135	0	1.02 KB
#5660	`CallElasticsearchAsync`	net472	2.67μs	2ns	7.75ns	0.166	0.00134	1.05 KB

Benchmarks.Trace.GraphQLBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Allocated
master	`ExecuteAsync`	net6.0	1.22μs	0.474ns	1.84ns	0.0135	952 B
master	`ExecuteAsync`	netcoreapp3.1	1.68μs	2.5ns	9.69ns	0.0126	952 B
master	`ExecuteAsync`	net472	1.78μs	1.12ns	4.32ns	0.145	915 B
#5660	`ExecuteAsync`	net6.0	1.19μs	0.555ns	2.15ns	0.0132	952 B
#5660	`ExecuteAsync`	netcoreapp3.1	1.62μs	0.835ns	3.13ns	0.0128	952 B
#5660	`ExecuteAsync`	net472	1.78μs	0.898ns	3.48ns	0.145	915 B

Benchmarks.Trace.HttpClientBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Allocated
master	`SendAsync`	net6.0	4.22μs	2.32ns	8.99ns	0.0295	2.22 KB
master	`SendAsync`	netcoreapp3.1	5.12μs	2.73ns	10.2ns	0.0382	2.76 KB
master	`SendAsync`	net472	7.52μs	1.46ns	5.45ns	0.499	3.15 KB
#5660	`SendAsync`	net6.0	4.1μs	1.55ns	5.79ns	0.0308	2.22 KB
#5660	`SendAsync`	netcoreapp3.1	4.97μs	2.71ns	10.1ns	0.0374	2.76 KB
#5660	`SendAsync`	net472	7.5μs	1.66ns	6ns	0.499	3.15 KB

Benchmarks.Trace.ILoggerBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Allocated
master	`EnrichedLog`	net6.0	1.61μs	0.474ns	1.71ns	0.0226	1.64 KB
master	`EnrichedLog`	netcoreapp3.1	2.1μs	0.702ns	2.63ns	0.0221	1.64 KB
master	`EnrichedLog`	net472	2.62μs	1.87ns	7.01ns	0.25	1.57 KB
#5660	`EnrichedLog`	net6.0	1.52μs	0.837ns	3.13ns	0.0228	1.64 KB
#5660	`EnrichedLog`	netcoreapp3.1	2.07μs	1.17ns	4.53ns	0.0226	1.64 KB
#5660	`EnrichedLog`	net472	2.57μs	1.58ns	5.92ns	0.249	1.57 KB

Benchmarks.Trace.Log4netBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Gen 1	Allocated
master	`EnrichedLog`	net6.0	114μs	261ns	1.01μs	0.057	0	4.28 KB
master	`EnrichedLog`	netcoreapp3.1	119μs	188ns	726ns	0.0605	0	4.28 KB
master	`EnrichedLog`	net472	147μs	97.3ns	377ns	0.658	0.219	4.46 KB
#5660	`EnrichedLog`	net6.0	115μs	154ns	595ns	0.057	0	4.28 KB
#5660	`EnrichedLog`	netcoreapp3.1	117μs	158ns	611ns	0.059	0	4.28 KB
#5660	`EnrichedLog`	net472	147μs	87.3ns	338ns	0.658	0.219	4.46 KB

Benchmarks.Trace.NLogBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Allocated
master	`EnrichedLog`	net6.0	2.93μs	0.882ns	3.42ns	0.0304	2.2 KB
master	`EnrichedLog`	netcoreapp3.1	4.1μs	1.35ns	4.86ns	0.0288	2.2 KB
master	`EnrichedLog`	net472	4.87μs	5.57ns	19.3ns	0.32	2.02 KB
#5660	`EnrichedLog`	net6.0	3.14μs	0.923ns	3.57ns	0.0298	2.2 KB
#5660	`EnrichedLog`	netcoreapp3.1	4.03μs	0.997ns	3.86ns	0.0299	2.2 KB
#5660	`EnrichedLog`	net472	4.94μs	1.21ns	4.68ns	0.318	2.02 KB

Benchmarks.Trace.RedisBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Gen 1	Allocated
master	`SendReceive`	net6.0	1.32μs	0.73ns	2.73ns	0.0159	0	1.14 KB
master	`SendReceive`	netcoreapp3.1	1.74μs	1.63ns	6.32ns	0.0149	0	1.14 KB
master	`SendReceive`	net472	1.98μs	1.51ns	5.67ns	0.183	0.000988	1.16 KB
#5660	`SendReceive`	net6.0	1.44μs	1.18ns	4.56ns	0.0158	0	1.14 KB
#5660	`SendReceive`	netcoreapp3.1	1.82μs	0.589ns	2.12ns	0.0155	0	1.14 KB
#5660	`SendReceive`	net472	2.02μs	0.603ns	2.17ns	0.183	0	1.16 KB

Benchmarks.Trace.SerilogBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Allocated
master	`EnrichedLog`	net6.0	2.76μs	0.51ns	1.91ns	0.022	1.6 KB
master	`EnrichedLog`	netcoreapp3.1	3.83μs	2.96ns	11.1ns	0.0226	1.65 KB
master	`EnrichedLog`	net472	4.34μs	1.34ns	5.02ns	0.323	2.04 KB
#5660	`EnrichedLog`	net6.0	2.72μs	0.753ns	2.92ns	0.0218	1.6 KB
#5660	`EnrichedLog`	netcoreapp3.1	3.91μs	1.35ns	5.24ns	0.0216	1.65 KB
#5660	`EnrichedLog`	net472	4.49μs	3.5ns	12.6ns	0.322	2.04 KB

Benchmarks.Trace.SpanBenchmark - Faster 🎉 Same allocations ✔️

Faster 🎉 in #5660

Benchmark	base/diff	Base Median (ns)	Diff Median (ns)	Modality
Benchmarks.Trace.SpanBenchmark.StartFinishScope‑net6.0	1.148	550.64	479.44
Benchmarks.Trace.SpanBenchmark.StartFinishSpan‑net472	1.123	685.60	610.64

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Allocated
master	`StartFinishSpan`	net6.0	403ns	0.338ns	1.31ns	0.0081	576 B
master	`StartFinishSpan`	netcoreapp3.1	602ns	0.431ns	1.55ns	0.00789	576 B
master	`StartFinishSpan`	net472	686ns	0.62ns	2.4ns	0.0915	578 B
master	`StartFinishScope`	net6.0	551ns	0.815ns	3.16ns	0.00967	696 B
master	`StartFinishScope`	netcoreapp3.1	703ns	0.375ns	1.45ns	0.00948	696 B
master	`StartFinishScope`	net472	869ns	0.795ns	3.08ns	0.105	658 B
#5660	`StartFinishSpan`	net6.0	391ns	0.172ns	0.668ns	0.00806	576 B
#5660	`StartFinishSpan`	netcoreapp3.1	573ns	0.791ns	3.06ns	0.0079	576 B
#5660	`StartFinishSpan`	net472	610ns	0.523ns	2.03ns	0.0916	578 B
#5660	`StartFinishScope`	net6.0	479ns	0.142ns	0.55ns	0.00988	696 B
#5660	`StartFinishScope`	netcoreapp3.1	757ns	0.286ns	1.11ns	0.00942	696 B
#5660	`StartFinishScope`	net472	847ns	0.557ns	2.16ns	0.104	658 B

Benchmarks.Trace.TraceAnnotationsBenchmark - Same speed ✔️ Same allocations ✔️

Raw results

Branch	Method	Toolchain	Mean	StdError	StdDev	Gen 0	Allocated
master	`RunOnMethodBegin`	net6.0	607ns	0.164ns	0.636ns	0.00971	696 B
master	`RunOnMethodBegin`	netcoreapp3.1	882ns	0.563ns	2.1ns	0.0093	696 B
master	`RunOnMethodBegin`	net472	1.04μs	0.653ns	2.53ns	0.104	658 B
#5660	`RunOnMethodBegin`	net6.0	588ns	0.333ns	1.29ns	0.00966	696 B
#5660	`RunOnMethodBegin`	netcoreapp3.1	918ns	1.43ns	5.54ns	0.00928	696 B
#5660	`RunOnMethodBegin`	net472	1.1μs	1.08ns	4.19ns	0.104	658 B

gleocadie · 2024-06-07T15:31:09Z

profiler/src/ProfilerEngine/Datadog.Profiler.Native.Linux/LinuxStackFramesCollector.cpp

+        _phdr = std::make_unique<ElfW(Phdr)[]>(source->dlpi_phnum);
+        memcpy(_phdr.get(), source->dlpi_phdr, sizeof(ElfW(Phdr)) * source->dlpi_phnum);
+        destination.dlpi_phdr = _phdr.get();
+
+        // Those fields appeared in glibc 2.4 (with two others).
+        // Since we compile with glibc 2.17, those fields are present (size of struct dl_phdr_info contains those fields),
+        // so need to check the size/offset.
+        // We do not know how to copy dlpi_tls_data field and libunwind does not use them, we can nullify/zeroify them
+        destination.dlpi_tls_modid = 0;
+        destination.dlpi_tls_data = nullptr;


wonder what about alpine 🤔 . Will check

Ok it looks safe on musl libc: the oldest one I could find contains those 2 fields. The oldest musl libc used by our customers is 1.1.22 and dl_phdr_info contains those fields.

chrisnas

LGTM

profiler/src/ProfilerEngine/Datadog.Profiler.Native.Linux/LibrariesInfoCache.cpp

github-actions bot added the area:profiler Issues related to the continous-profiler label Jun 6, 2024

gleocadie commented Jun 7, 2024

View reviewed changes

gleocadie marked this pull request as ready for review June 10, 2024 09:08

gleocadie requested a review from a team as a code owner June 10, 2024 09:08

gleocadie force-pushed the gleocadie/add-custom-dl_iterate_phdr branch 4 times, most recently from 47ecbe4 to abe5f8d Compare June 10, 2024 10:22

chrisnas approved these changes Jun 10, 2024

View reviewed changes

gleocadie force-pushed the gleocadie/add-custom-dl_iterate_phdr branch 5 times, most recently from 0961502 to 95fbf54 Compare June 24, 2024 10:56

gleocadie force-pushed the gleocadie/add-custom-dl_iterate_phdr branch 2 times, most recently from 48712f4 to 2c0b163 Compare June 26, 2024 14:38

gleocadie added 10 commits June 27, 2024 09:47

Add custom dl_iterate_phdr and use it in libunwind

fbea5b5

wrap dl_phdr_info to deep-copy and own their lifetime

0240ceb

Clean up code

8c63466

Clean up code again

29f7ba6

Cache only when necessary

2c07401

WIP

a16adb7

Clean up

0cc3347

Have only one instance of libraries cache

e322cb6

Renaming

ad00a4c

Fix build

83931f8

gleocadie force-pushed the gleocadie/add-custom-dl_iterate_phdr branch from 2c0b163 to 83931f8 Compare June 27, 2024 07:59

kevingosse reviewed Jun 27, 2024

View reviewed changes

Address PR comment

4a50889

kevingosse approved these changes Jun 27, 2024

View reviewed changes

gleocadie merged commit 8cbbb4b into master Jun 28, 2024
62 of 65 checks passed

gleocadie deleted the gleocadie/add-custom-dl_iterate_phdr branch June 28, 2024 09:52

github-actions bot added this to the vNext-v2 milestone Jun 28, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Profiler] Add custom dl_iterate_phdr and use it in libunwind #5660

[Profiler] Add custom dl_iterate_phdr and use it in libunwind #5660

gleocadie commented Jun 6, 2024 •

edited

Loading

datadog-ddstaging bot commented Jun 6, 2024 •

edited

Loading

andrewlock commented Jun 6, 2024 •

edited

Loading

andrewlock commented Jun 6, 2024 •

edited

Loading

andrewlock commented Jun 6, 2024 •

edited

Loading

Raw results

Raw results

Raw results

Fewer allocations 🎉 in #5660

Raw results

Raw results

Raw results

Raw results

Raw results

Raw results

Raw results

Raw results

Raw results

Raw results

Faster 🎉 in #5660

Raw results

Raw results

gleocadie Jun 7, 2024

gleocadie Jun 7, 2024

chrisnas left a comment

[Profiler] Add custom dl_iterate_phdr and use it in libunwind #5660

[Profiler] Add custom dl_iterate_phdr and use it in libunwind #5660

Conversation

gleocadie commented Jun 6, 2024 • edited Loading

Summary of changes

Reason for change

Implementation details

Test coverage

Other details

datadog-ddstaging bot commented Jun 6, 2024 • edited Loading

Datadog Report

New Flaky Tests (2)

⌛ Performance Regressions vs Default Branch (3)

andrewlock commented Jun 6, 2024 • edited Loading

Execution-Time Benchmarks Report ⏱️

andrewlock commented Jun 6, 2024 • edited Loading

Throughput/Crank Report:zap:

andrewlock commented Jun 6, 2024 • edited Loading

Benchmarks Report for tracer 🐌

Benchmark details

Raw results

Raw results

Raw results

Fewer allocations 🎉 in #5660

Raw results

Raw results

Raw results

Raw results

Raw results

Raw results

Raw results

Raw results

Raw results

Raw results

Faster 🎉 in #5660

Raw results

Raw results

gleocadie Jun 7, 2024

Choose a reason for hiding this comment

gleocadie Jun 7, 2024

Choose a reason for hiding this comment

chrisnas left a comment

Choose a reason for hiding this comment

gleocadie commented Jun 6, 2024 •

edited

Loading

datadog-ddstaging bot commented Jun 6, 2024 •

edited

Loading

andrewlock commented Jun 6, 2024 •

edited

Loading

andrewlock commented Jun 6, 2024 •

edited

Loading

andrewlock commented Jun 6, 2024 •

edited

Loading