[dougqh]

What Does This Do

Makes getTag and unsafeGetTag resolve OpenTelemetry names whose tag depends on the span's direction, using the span's kind:

Name	Inbound (server, consumer)	Outbound (client, producer)
server.address	http.hostname	peer.hostname
server.port	server.port (its own tag)	peer.port
network.peer.address	network.client.ip	—
client.port	peer.port (the client's port)	—

The fallback runs only after a plain lookup misses:

• Literal name wins: a tag set under the literal name (for example server.address set through the Datadog API, stored as a custom tag) is still returned first.
• Only null results change: a read that finds something today returns the same value.
• No direction, no resolution: internal, broker, custom and unset kinds resolve nothing.
• server.port on outbound spans: server.port is also a Datadog name (the server's own port, set only on inbound spans). On a client or producer span, a read of server.port that misses now returns peer.port, matching its OpenTelemetry meaning there. So a sampling rule such as server.port:5432 matches database client spans. A Datadog-meaning reader can't get a wrong value, because the Datadog server.port tag isn't set on outbound spans.

The direction comes from the cached span-kind ordinal, so the fallback doesn't touch the tag map or take a lock to find it.

Motivation

Tag sampling with OpenTelemetry names. Trace and span sampling rules match tags with span.getTag(name). Today, a rule written with the OpenTelemetry name of a direction-independent rename already works (http.request.method, http.response.status_code, url.full, …), because TagMap maps OTel names to Datadog names on lookup. But a rule on server.address silently never matched, because KnownTagCodec.keyOf deliberately returns 0 for direction-scoped names. Resolving direction was only done at the OpenTelemetry boundary, on writes.

Checked on master before this change, with a probe against real spans:

• http.method vs http.request.method: match;
• http.status_code vs http.response.status_code:2*: match;
• server.address on server spans (http.hostname) and on client spans (peer.hostname): no match.

Timing: sampling decides at the first inject or at the root's finish (trace sampling), or at span finish (span sampling). By then, decorators, span prototypes or the OpenTelemetry span builder have set span.kind.

Additional Notes

• Cost: none on hits. On a miss with a known direction, one extra registry lookup (directionalKeyOf). Internal readers are moving to tag ids, so the name path being a little slower on misses is acceptable.
• Tests (DDSpanContextTest):
  • resolution on server, consumer, client and producer spans;
  • no resolution where the name names no tag (an outbound-only rename on a server span, and the reverse), or the span is internal or broker, or has no kind;
  • the literal name wins;
  • a sampling rule on server.address matches a client span through TagsMatcher.
• Local run: full dd-trace-core; the only failures are the two Docker-based tests (no Docker locally).
• Overlap: Split TagMap into id and name stores, owned directly by the span #12918 rewrites DDSpanContext's tag storage. This fallback is a few lines and moves easily with it.

Jira ticket: [PROJ-IDENT]

🤖 Generated with Claude Code

[dd-octo-sts]

🟢 Java Benchmark SLOs — All performance SLOs passed

Suite	Status
Startup	🟢 pass

SLO thresholds are defined here based on automatically generated metrics. A warning is raised when results are within 5% of the threshold.

PR vs. master results

Scenario	Candidate	master	Δ (95% CI of mean)
startup:insecure-bank:iast:Agent	14.65 s	14.63 s	[-0.5%; +0.8%] (no difference)
startup:insecure-bank:tracing:Agent	13.53 s	13.63 s	[-1.6%; +0.2%] (no difference)
startup:petclinic:appsec:Agent	17.17 s	16.98 s	[+0.2%; +2.0%] (maybe worse)
startup:petclinic:iast:Agent	17.03 s	16.87 s	[+0.1%; +1.8%] (maybe worse)
startup:petclinic:profiling:Agent	16.53 s	16.75 s	[-2.2%; -0.4%] (maybe better)
startup:petclinic:sca:Agent	17.02 s	16.83 s	[+0.4%; +1.8%] (maybe worse)
startup:petclinic:tracing:Agent	16.10 s	15.68 s	[-1.8%; +7.1%] (no difference)

Commit: 81429bf5 · CI Pipeline · Benchmarking Platform UI

---

Load and DaCapo benchmarks can be triggered manually in the GitLab pipeline. Results will appear in the Benchmarking Platform UI after completion.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-21T12:16:03.815617Z	81429bf	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[datadog-official]

More details

Existing tag values retain precedence, while direction-scoped misses resolve consistently with the registry mappings and cached span kind.

Was this helpful? React 👍 or 👎

_{Open Bits AI session}

_{🤖 Bits Code Review · Commit 81429bf}

[mcculls]

LGTM

[dd-octo-sts]

/merge

[gh-worker-devflow-routing-ef8351]

View all feedbacks in Devflow UI.

2026-10-21 13:02:46 UTC ℹ️ Start processing command /merge

---

2026-10-21 13:02:50 UTC ℹ️ MergeQueue: pull request added to the queue

The expected merge time in master is approximately 1h (p90).

---

2026-10-21 14:16:45 UTC ℹ️ MergeQueue: This merge request was merged