By the numbers

Data collected on 2026-04-30 from main at commit f1b64d50.

A quantitative snapshot of the Datadog Agent codebase. Numbers come from git log, find, and direct file inspection. Per-person stats are intentionally excluded; for ownership info see Maintainers.

Size

The Agent is one of Datadog's largest open-source repositories.

xychart-beta horizontal
    title "Source files by language"
    x-axis ["Go", "Python", "Rust", "C / eBPF"]
    y-axis "File count" 0 --> 12000
    bar [10495, 523, 88, 66]

Language	Source files
Go (*.go)	10,495
Python (*.py)	523
Rust (*.rs)	88
C / eBPF (*.c)	66
Markdown (*.md)	315

The Go file count includes tests, generated mocks, and platform-specific shims. Tests typically end in _test.go. Generated mocks live alongside their definitions or under mocks/ subdirectories.

The repository also contains:

• ~50,000+ files total when including BUILD.bazel files, configuration, schemas, and documentation.
• A LICENSE-3rdparty.csv of about 800 KB enumerating every third-party dependency and its license — a side effect of the Go module sprawl.
• A CHANGELOG.rst of about 730 KB covering ten years of releases.

Top-level layout

Path	Role	Approximate size
pkg/	Shared Go libraries (legacy + current)	~64 subdirectories at top level
comp/	Component framework bundles	~48 bundles
cmd/	Binary entry points	24 binaries
tasks/	Python Invoke tasks	~80 task modules
test/	Unit, integration, E2E tests	dozens of subtrees
tools/	Development helper binaries	~12 subtrees
omnibus/ + packages/	Packaging logic	hundreds of files each
docs/	Developer + public documentation	~315 markdown files

Activity

The Agent is one of the most active codebases at Datadog.

Window	Commits
Last 90 days	2,547
Lifetime	34,158
First commit	2016-02-09
Latest commit (HEAD)	2026-04-29

That works out to a mean of roughly 28 commits per day over the last quarter. The repo merges close to ten years of history through 2026.

Bot-attributed commits

Roughly 7.7% of the last 90 days of commits were authored by automation bots — renovate[bot], dependabot[bot], github-actions[bot], etc. Concretely: 196 of 2,547 commits (excluding co-authored credits).

A small additional fraction of commits include Co-authored-by: Claude … or Co-authored-by: Claude Opus / Sonnet … trailers, indicating AI-assisted but human-authored work. Five such trailers appear in the last 90 days.

These numbers are a lower bound on automation usage. Inline AI tools like Cursor or Copilot leave no trace in the git history; many commits attributed to humans were likely co-written with an assistant.

Activity hotspots

Tooling for full per-directory churn isn't included in this snapshot, but you can reproduce it with:

git log --since='90 days ago' --pretty=format: --name-only \
  | sort | uniq -c | sort -rn | head -20

Areas that consistently dominate the recent history (last 90 days):

• pkg/network/ and pkg/security/ — System Probe and CWS see continuous eBPF + protocol work.
• comp/core/ and comp/forwarder/ — ongoing migration from pkg/ to the component framework.
• pkg/trace/ — the Trace Agent sees a steady cadence of sampler and concentrator changes.
• tasks/ and .gitlab/ — CI changes are very high-frequency in this repo.
• pkg/logs/ — adaptive sampling and source rework.

Languages and ecosystems

The Agent is a polyglot project. Each language plays a different role:

Language	Where	Why
Go	Almost everywhere	Primary implementation language, ~95% of source files.
Python	tasks/, cmd/agent/dist/checks/, pkg/jmxfetch/ test fixtures	Invoke build/test/release scripts; runtime check loader (CPython 3); JMX bridge driver.
C / C++	pkg/ebpf/c/, rtloader/, kernel headers	eBPF programs (kernel side) and the Python runtime loader (CPython embed).
Rust	rust/, pkg/dyninst/ adjacent	Small auxiliary crates for newer subsystems.
Bazel Starlark	bazel/rules/, *.bazel files	Build rules and toolchain definitions.
Ruby	omnibus/, chocolatey/	Legacy packaging system.
Java	(external) jmxfetch.jar	JMX collection subprocess shipped alongside the Agent.

Test-to-code ratio

A rough Go-only ratio:

find . -name '*_test.go' -not -path './.git/*' | wc -l   # tests
find . -name '*.go' -not -name '*_test.go' -not -path './.git/*' | wc -l  # source

Out of 10,495 total Go files, several thousand have the _test.go suffix. The exact ratio varies wildly by package: pkg/network/, pkg/security/, and pkg/aggregator/ have heavy unit and integration test coverage; many comp/<x>/<y>/fx/ Fx-glue packages have minimal tests by design.

In addition to in-repo Go tests, an entire end-to-end test fleet lives under test/new-e2e/ and provisions real cloud infrastructure (AWS, GCP, Azure). E2E tests use fakeintake (test/fakeintake/) to assert payloads end up where they should.

Configuration surface

• Build tags: ~60+ defined in tasks/build_tags.py.
• Runtime configuration knobs: hundreds of DD_* environment variables and datadog.yaml keys, computed from pkg/config/setup/. The schema is large enough that there is a dedicated agent createschema subcommand to dump it.
• Feature flags: many features are gated by config keys rather than build tags, allowing operators to disable work at runtime.

Dependency footprint

go.mod is the single largest in the repository (93 KB), with go.sum at ~692 KB. The presence of LICENSE-3rdparty.csv (800 KB) reflects the long tail of indirect dependencies that come with collecting from every cloud provider, container runtime, OTel exporter, and language tracer the project supports.

For details, see Dependencies.

Things that surprise newcomers

• The Agent is not one binary. The same repo produces 24 binaries under cmd/. They share Go modules and the component framework but are deployed independently.
• A single line of business logic can touch Go, Bazel, eBPF C, Python (Invoke), and YAML (CI) — every change tends to ripple outward.
• The check runtime loads CPython 3 via rtloader/; that's why the Agent has a CMake build and a C++ ABI in addition to all the Go.
• The tasks/ directory has more code than many entire open-source projects.