libdatadog 29.0.0: integrate libdatadog sample types by r1viollet · Pull Request #504 · DataDog/ddprof

r1viollet · 2026-03-05T17:35:02Z

What does this PR do?

Upgrades libdatadog to v29.0.0 and integrates the new ddog_prof_SampleType enum for all pprof sample type declarations.

Previously, ddprof passed sample type names as free-form strings directly to libdatadog. Starting with v28, libdatadog introduced a ddog_prof_SampleType enum to replace those strings, with the string mapping handled internally by the library.

Motivation

Allow to release ddprof

How to test the change?

Run with --preset cpu_live_heap --debug_pprof_prefix /tmp/test\ and inspect with pprof -raw`: all 6 expected columns.

Run with tracepoints:

DDPROF=/home/r1viollet/dd/ddprof/build_gcc_alpine-linux-1.2.5_AlpRel/ddprof && sudo $DDPROF -l notice -S test-r1 -e "event=tlb:tlb_flush" --debug_pprof_prefix /tmp/test /usr/local/bin/BadBoggleSolver_run 10

API changes from v26.0.0: - ddog_prof_Endpoint_agent/agentless gained a use_system_resolver bool. ddprof is an out-of-process profiler that does not fork during export, so use_system_resolver=true (system /etc/resolv.conf) is safe and correct. - ddog_prof_ValueType (name+unit strings) replaced by ddog_prof_SampleType enum; ddog_prof_Period.type_ renamed to .sample_type accordingly. The ValueType->SampleType mapping is expressed as a constexpr lookup table (k_ddog_sample_types) whose rows directly mirror PROFILE_TYPE_TABLE in perf_watcher.hpp. A static_assert on DDPROF_PWT_LENGTH enforces that adding a new profile type requires updating the mapping explicitly. Also add Datadog_LOCAL_ROOT cmake override to Findlibdatadog.cmake to allow testing local libdatadog builds without a GitHub release (skips the fetch script when set).

pr-commenter · 2026-03-05T17:43:25Z

Benchmark results for collatz

Parameters

	Baseline	Candidate
config	baseline	candidate
profiler-version	ddprof 0.23.0+0c700352.103448873	ddprof 0.23.0+18c47295.103465578

Summary

Found 0 performance improvements and 0 performance regressions! Performance is the same for 1 metrics, 0 unstable metrics.

See unchanged results

scenario	Δ mean execution_time
scenario:ddprof -S bench-collatz --preset cpu_only collatz_runner.sh	same

pr-commenter · 2026-03-05T17:43:28Z

Benchmark results for BadBoggleSolver_run

Parameters

	Baseline	Candidate
config	baseline	candidate
profiler-version	ddprof 0.23.0+0c700352.103448873	ddprof 0.23.0+18c47295.103465578

Summary

Found 0 performance improvements and 0 performance regressions! Performance is the same for 1 metrics, 0 unstable metrics.

See unchanged results

scenario	Δ mean execution_time
scenario:ddprof -S bench-bad-boggle-solver BadBoggleSolver_run work 1000	unsure [-16.575ms; -3.960ms] or [-0.830%; -0.198%]

r1viollet · 2026-03-05T17:43:49Z

include/perf_watcher.hpp

+
+// Predefined sample type mappings — defined in perf_watcher.cc.
+// Generic sample counting (hardware events, misc software events).
+extern const WatcherSampleTypes k_stype_tracepoint;


slightly strange, adjusting

…mapping Remove the intermediate DDPROF_PWT_* profile type taxonomy and k_ddog_sample_types lookup table, replacing them with WatcherSampleTypes structs that directly store ddog_prof_SampleType integer values per aggregation mode. Key changes: - Remove PROFILE_TYPE_TABLE macro and DDPROF_SAMPLE_TYPES enum - Remove 5 helper functions (sample_type_name_from_idx, etc.) - Add WatcherSampleTypes{sample_types[], count_types[]} to PerfWatcher - Add pprof_active bool to replace the sDUM special-casing - Replace ProfSampleTypes/get_active_ids/compute_pprof_values with a simple SlotRegistry that deduplicates ddog_prof_SampleType slots - Use uint32_t storage to avoid libdatadog header dependency in perf_watcher.cc; static_asserts in ddprof_pprof.cc verify values All 53 tests pass. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

r1viollet · 2026-03-05T18:06:34Z

tracepoints do not work... fixing this

src/pprof/ddprof_pprof.cc

- Fix tracepoint count sentinel: k_stype_tracepoint count_types now use k_stype_val_sample (sentinel = no companion) instead of k_stype_val_tracepoint. Previously both pprof_index and pprof_count_index resolved to the same slot, causing the count to overwrite the primary value in pprof_aggregate. - Fix watcher_has_tracepoint to check k_stype_val_tracepoint instead of k_stype_val_sample - Fix debug pprof filename extension from .pprof.lz4 to .pprof.zst (the serialized bytes from libdatadog are zstd-compressed) - Add static_assert for DDOG_PROF_SAMPLE_TYPE_TRACEPOINT == 38 - Use named k_stype_val_* constants in sample_type_name() switch cases - Add static_assert per sample type string to catch future libdatadog renames at compile time Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Replace the outdated DDPROF_PWT_* table with the new WatcherSampleTypes model. Document all three watcher kinds (cpu, alloc, tracepoint), the primary/count companion pattern, the sentinel mechanism, and the compile-time static_assert safety net. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/perf_watcher.cc

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

include/watcher_sample_types.hpp

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

include/watcher_sample_types.hpp

This avoids forward declaring profile types

The documentation can be regenated from source if needed

r1viollet · 2026-03-19T11:32:01Z

include/watcher_sample_types.hpp

+inline constexpr uint32_t k_stype_none = UINT32_MAX;
+
+// Tracepoints: one event = one sample, no count companion, no live mode.
+// clang-format off


@nsavoire this part looks OK ? wdyt ?

nsavoire · 2026-03-19T17:29:40Z

src/pprof/ddprof_pprof.cc

+// Verify that sample_type_name() returns the strings the backend expects.
+static_assert(std::string_view(sample_type_name(
+                  DDOG_PROF_SAMPLE_TYPE_CPU_TIME)) == "cpu-time");
+static_assert(std::string_view(sample_type_name(
+                  DDOG_PROF_SAMPLE_TYPE_CPU_SAMPLES)) == "cpu-samples");
+static_assert(std::string_view(sample_type_name(
+                  DDOG_PROF_SAMPLE_TYPE_ALLOC_SPACE)) == "alloc-space");
+static_assert(std::string_view(sample_type_name(
+                  DDOG_PROF_SAMPLE_TYPE_ALLOC_SAMPLES)) == "alloc-samples");
+static_assert(std::string_view(sample_type_name(
+                  DDOG_PROF_SAMPLE_TYPE_INUSE_SPACE)) == "inuse-space");
+static_assert(std::string_view(sample_type_name(
+                  DDOG_PROF_SAMPLE_TYPE_INUSE_OBJECTS)) == "inuse-objects");
+static_assert(std::string_view(sample_type_name(
+                  DDOG_PROF_SAMPLE_TYPE_TRACEPOINT)) == "tracepoint");
+static_assert(std::string_view(
+                  sample_type_name(DDOG_PROF_SAMPLE_TYPE_SAMPLE)) == "sample");


I am not sure these are useful: sample_type_name function is very simple

If these change, we would BE would not see the relevant profile types.

nsavoire · 2026-03-19T22:24:38Z

include/perf_watcher.hpp

-  X(sEMU,       "Emu. Faults",        PERF_TYPE_SOFTWARE,   PERF_COUNT_SW_EMULATION_FAULTS,        99,           DDPROF_PWT_TRACEPOINT,  IS_FREQ)                 \
-  X(sDUM,       "Dummy",              PERF_TYPE_SOFTWARE,   PERF_COUNT_SW_DUMMY,                   1,            DDPROF_PWT_NOCOUNT,     {})                      \
-  X(sALLOC,     "Allocations",        kDDPROF_TYPE_CUSTOM,  kDDPROF_COUNT_ALLOCATIONS,             524288,       DDPROF_PWT_ALLOC_SPACE, SKIP_FRAMES)
+  X(hCPU,       "CPU Cycles",         PERF_TYPE_HARDWARE,   PERF_COUNT_HW_CPU_CYCLES,              99,           k_stype_tracepoint,    true,         IS_FREQ)     \


You could replace pprof_active with a dummy sample type (similarly to what was done before):

inline constexpr WatcherSampleTypes k_stype_dummy = { {k_stype_none, k_stype_none}, {k_stype_none, k_stype_none}}; inline constexpr bool is_pprof_active(const WatcherSampleTypes &st) { return st.sample_types[0] != k_stype_none; }

r1viollet commented Mar 5, 2026

View reviewed changes

r1viollet force-pushed the r1viollet/libdatadog_28.0.2 branch from 47c35b8 to 4326981 Compare March 5, 2026 17:48

r1viollet commented Mar 6, 2026

View reviewed changes

src/pprof/ddprof_pprof.cc Show resolved Hide resolved

r1viollet and others added 2 commits March 6, 2026 13:22

r1viollet commented Mar 6, 2026

View reviewed changes

src/perf_watcher.cc Show resolved Hide resolved

style: align sample_type_name() cases with static_assert order

475e553

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

nsavoire reviewed Mar 6, 2026

View reviewed changes

include/watcher_sample_types.hpp Outdated Show resolved Hide resolved

nsavoire reviewed Mar 6, 2026

View reviewed changes

include/watcher_sample_types.hpp Outdated Show resolved Hide resolved

chore: upgrade libdatadog to v29.0.0, fix sample type sentinel

d0147c5

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

r1viollet marked this pull request as ready for review March 19, 2026 10:51

r1viollet changed the title ~~libdatadog 28.0.2: integrate libdatadog sample types~~ libdatadog 29.0.0: integrate libdatadog sample types Mar 19, 2026

r1viollet commented Mar 19, 2026

View reviewed changes

include/watcher_sample_types.hpp Outdated Show resolved Hide resolved

r1viollet added 2 commits March 19, 2026 12:22

libdatadog 29: Add an include of libdatadog headers

9b1872c

This avoids forward declaring profile types

libdatadog 29 - remove documentation on profile types

18c4729

The documentation can be regenated from source if needed

r1viollet requested a review from nsavoire March 19, 2026 11:31

r1viollet commented Mar 19, 2026

View reviewed changes

r1viollet assigned nsavoire Mar 19, 2026

xroche mentioned this pull request Mar 19, 2026

Make libdatadog fetching optional #514

Open

nsavoire reviewed Mar 19, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

libdatadog 29.0.0: integrate libdatadog sample types#504

libdatadog 29.0.0: integrate libdatadog sample types#504
r1viollet wants to merge 8 commits intomainfrom
r1viollet/libdatadog_28.0.2

r1viollet commented Mar 5, 2026 •

edited

Loading

Uh oh!

pr-commenter bot commented Mar 5, 2026 •

edited

Loading

Uh oh!

pr-commenter bot commented Mar 5, 2026 •

edited

Loading

Uh oh!

r1viollet Mar 5, 2026

Uh oh!

r1viollet commented Mar 5, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

r1viollet Mar 19, 2026

Uh oh!

nsavoire Mar 19, 2026

Uh oh!

r1viollet Mar 19, 2026

Uh oh!

nsavoire Mar 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

r1viollet commented Mar 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Motivation

How to test the change?

Uh oh!

pr-commenter bot commented Mar 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmark results for collatz

Parameters

Summary

Uh oh!

pr-commenter bot commented Mar 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmark results for BadBoggleSolver_run

Parameters

Summary

Uh oh!

r1viollet Mar 5, 2026

Choose a reason for hiding this comment

Uh oh!

r1viollet commented Mar 5, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

r1viollet Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

nsavoire Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

r1viollet Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

nsavoire Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

r1viollet commented Mar 5, 2026 •

edited

Loading

pr-commenter bot commented Mar 5, 2026 •

edited

Loading

pr-commenter bot commented Mar 5, 2026 •

edited

Loading