datadog/update-otel-deps
Use when a user asks to update, bump, or troubleshoot OpenTelemetry Collector dependencies in datadog-agent, including OCB build failures, ddflareextension golden files, static quality gates, or OTel transitive dependency conflicts.
npx skills add https://github.com/DataDog/datadog-agent --skill update-otel-deps
Update OpenTelemetry collector dependencies in datadog-agent.
Arguments: $ARGUMENTS — optional <target-version> (e.g. v0.125.0). Note: inv collector.update always upgrades to the latest published OTel version; if a specific version is requested, a manual search-and-replace is needed instead (see Manual Step 1).
A scheduled GitHub Actions workflow (.github/workflows/collector-generate-and-update.yml) runs every other Wednesday and automatically opens a draft PR with the OTel update. Always check this before doing anything manually.
unset GITHUB_TOKEN
gh run list --workflow=collector-generate-and-update.yml --branch=main --limit=5 \
--json databaseId,status,conclusion,createdAt,event,displayTitle
If the latest run succeeded: find and review the auto-generated draft PR (Step 0c), then use /dd:ci:fix to fix any remaining CI failures and undraft.
If the latest run failed: go to Step 0b to diagnose and fix before falling back to a manual update.
If no run has ever succeeded or the workflow looks consistently broken: check #opentelemetry-agent and then fall back to Manual Step 1.
Check #otel-agent-ops for the failure notification posted by the workflow; it includes a direct link to the failed run. Then fetch the logs:
RUN_ID=<id from step 0a>
unset GITHUB_TOKEN
gh run view $RUN_ID --log-failed 2>&1 | grep -E "error|Error|failed|exit code|XDG|bazel|OCB" | head -40
Common workflow failures and fixes:
| Symptom in logs | Cause | Fix |
|---|---|---|
| XDG_CACHE_HOME () must denote a directory in CI or exit code 2 on bazel mod tidy | Set up Bazel cache step missing or not running before collector.update | Ensure the .github/actions/bazel-cache step appears before Run Collector Update Script in the workflow |
| Branch named update-otel-collector-dependencies- (empty version) | os.environ in the Python subprocess doesn't propagate back to the shell; the old echo "OCB_VERSION=$OCB_VERSION" reads a blank value | Ensure the workflow extracts the version from tasks/collector.py after the update: OCB_VERSION=$(grep -m1 'OCB_VERSION = ' tasks/collector.py \| grep -oP '"\K[^"]+') |
| permission denied / HTTP 403 from dd-octo-sts-action | Workflow running from a non-main branch | The octo-sts policy only grants tokens on main; always trigger via schedule or gh workflow run … --ref main |
After fixing the workflow, commit to main and re-trigger:
unset GITHUB_TOKEN
gh workflow run collector-generate-and-update.yml --ref main
gh run watch
If the workflow infrastructure succeeds but the resulting changes break CI (golden file mismatches, OCB build errors, etc.), that means the OTel version bump itself needs manual fixes. Proceed to Manual Step 2 using the auto-generated PR's branch as your working branch — don't start a new branch from scratch.
unset GITHUB_TOKEN
gh pr list --search "Update OTel Collector dependencies" --state open \
--json number,title,url,headRefName
Review the diff, run /dd:ci:fix to fix CI failures, then undraft and request review once CI is green.
> If a specific target version was requested that differs from what the workflow produced, fall back to Manual Step 1.
Use this only when: the automated workflow has a persistent infrastructure failure that can't be fixed quickly, or a specific target version (not latest) is requested.
dda inv collector.update # bumps go.mod + OCB YAML files to latest OTel version
dda inv collector.generate # regenerates OTel Agent code from the new manifests
dda inv components.lint-components --fix
dda inv modules.add-all-replace
dda inv tidy # reconcile transitive dependencies
bazel run //:go_mod_tidy_all
dda inv generate-licenses # update license inventory
If a specific version was requested, skip inv collector.update and do a repo-wide search-and-replace of the old version string (find it in tasks/collector.py — the OCB_VERSION / OTEL_CONTRIB_VERSION constants). Then run the remaining commands above.
After the commands finish, scan for any files the task missed:
git diff HEAD --stat
OLD=$(git diff HEAD -- tasks/collector.py | grep '^-OCB_VERSION' | grep -oP '"[\d.]+"' | tr -d '"')
rg "$OLD" -g "*.go" -g "*.yml" -g "*.yaml" -g "*.mod" -g "*.sum" -l
Manually update any remaining files that still reference the old version.
This is the most common failure. The DD flare extension tests compare OTel runtime config output against golden files. Any upstream config field change breaks them.
Run the test locally to get the actual diff:
dda inv test --targets=./comp/otelcol/ddflareextension/impl/... --build-include=otlp -- -run TestGetConfDump
The failure output shows exactly which lines differ. Apply the diffs to the golden files in:
comp/otelcol/ddflareextension/impl/testdata/ (unit test golden files)test/new-e2e/tests/otel/otel-agent/testdata/ (E2E test golden files — apply the same changes)> The E2E testdata mirrors the unit testdata for runtime config fields. Keep them in sync.
The CI job datadog_otel_components_ocb_build verifies OTel modules can be built with OCB (OTel Collector Builder). Run it locally:
test/otel/testdata/ocb_build_script.sh
# or find the CI definition:
rg "ocb_build" .gitlab/ -g "*.yml" -l
Check logs under /tmp/otel-ci/ for the exact error.
Common causes and fixes:
| Cause | Fix |
|---|---|
| Breaking changes in OTel core collector API | Update the affected Agent OTel modules to adapt (e.g. comp/otelcol/) |
| Incompatible changes between Agent and upstream datadogexporter/datadogconnector | 1. Open a PR in opentelemetry-collector-contrib to point the upstream exporter at a commit version of Agent packages. 2. After merge, add replace statements in ocb_build_script.sh pointing to the contrib HEAD commit. |
| Missing OCB artifacts during collector release (transient) | Wait for upstream release artifacts to appear, or temporarily disable the CI job after discussing in #opentelemetry-agent. |
OTel upgrades often increase binary sizes (e.g. via gRPC version bumps in transitive deps). OTel upgrade PRs have a standing exemption from the binary size gates. To fix:
static_quality_gate_*).max_on_disk_size / max_on_wire_size values in test/static/static_quality_gates.yml.OTel collector updates dependencies more aggressively than datadog-agent. A version bump may force a newer version of a transitive dep (commonly docker or k8s packages) that has breaking API changes.
Preferred fix: Update Agent code to use the newer API required by OTel's transitive dep version. Look at the build errors after inv tidy to find which symbols changed.
Fallback: Pin the old version with a replace directive in the relevant go.mod:
replace github.com/docker/docker => github.com/docker/docker v24.0.x+incompatible
Then run dda inv tidy again. Use this only when upgrading the Agent code is not feasible in the same PR.
dda inv linter.go --targets=./comp/otelcol/... # lint the changed OTel components
dda inv test --targets=./comp/otelcol/ddflareextension/impl/... --build-include=otlp # confirm tests pass
Open a draft PR to let CI catch any remaining failures. The PR title convention is:
Update OTel Collector dependencies to v<VERSION>
Once the PR is open, run /dd:ci:fix to automatically fetch, analyze, and fix any remaining CI failures. If issues are not covered by the troubleshooting steps above, post in #opentelemetry-agent.
dda inv collector.update + collector.generate + tidy + generate-licenses all completed without error (automated or manual)/dd:ci:fix for any remaining failures)Take datadog/update-otel-deps from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.