Peel dbUser out of DatabaseClientDecorator into a param-injected TagExtractor

[dougqh]

What

First real dissolution of the DB decorator hierarchy: removes the dbUser template method from DatabaseClientDecorator and replaces it with a param-injected connection-tags TagExtractor. Stacked on #11820.

Why dbUser

A null-ratio scan of the DB template methods flagged it as the sparsest: 28/30 leaves returned null — only JDBC and Vertx-SQL provide a real user. That sparsity is the SQL/NoSQL false-generalization showing through: db.user is a SQL-auth concept the NoSQL/cache stores (Redis, Mongo, Cassandra, …) can only answer with null, yet every DB span paid a virtual dbUser() call to set DB_USER = null. (dbInstance 0.65 / dbHostname 0.52 are next, but they're DERIVED — they feed service-name split — so they wait on the declarative-derivation layer.)

The pattern (param-injected extractor)

DatabaseClientDecorator gains onConnection(span, conn, TagExtractor<CONNECTION>); the 2-arg overload delegates with a shared no-op extractor. The extractor is passed as a caller-side static final constant, so at a monomorphic advice site it devirtualizes + inlines when onConnection inlines — the key advantage over a virtual connectionExtractor() provider (which would re-megamorphize). This is also the seam for disentangling pure extraction from side-effects in the harder decorators later.

Changes

• Base: param-form onConnection + no-op default; remove abstract dbUser and the setTag(DB_USER, …) line.
• JDBC: fold db.user into ConnectionInfoExtractor, apply via super.onConnection(..., INSTANCE).
• Vertx-SQL: new VertxSqlConnectionExtractor, applied via the param form at its super.onConnection call (the provider-with-no-override case the param form is for).
• 26 NoSQL/cache decorators: drop the dead dbUser → null override.
• DatabaseClientConnectionBenchmark (jmh): the inlining acceptance harness (below).

Net −71 lines. db.user is set unconditionally by the two providers to preserve the prior (present-check-free) behavior exactly.

Inlining verification (the acceptance test — done)

DatabaseClientConnectionBenchmark exercises the real param-form onConnection at a mono site (one static final extractor — the production shape) vs mega (8 rotating extractor types — worst case). -XX:+PrintInlining:

• mono: DatabaseClientDecorator::onConnection — inline (hot) and the concrete extractor's extract() — inline (hot) (the abstract TagExtractor::extract slot is bypassed). Full devirt + inline. 79.0M ops/s.
• mega: 67.6M ops/s, only −14% — the JIT polymorphically inlined all 8 tiny bodies. (Honest caveat: mega is optimistic — trivial 11-byte bodies fit HotSpot's polymorphic-inline ceiling; fuller real bodies would fall to a virtual call sooner. Production is mono.)

So the devirt claim holds at a monomorphic site.

Testing

• agent-bootstrap decorator tests updated + green (incl. new param-form coverage). These are mock interaction-asserts for the old mechanism — flagged as Spock→JUnit migration candidates, not migrated here to keep the PR focused.
• JDBCInstrumentationV0Test: 87/87; ConnectionInfoExtractor (now incl. db.user) auto-injected as a transitive helper — no helperClassNames() edit.
• Vertx-SQL / NoSQL integration tests are container-gated (CI); VertxSqlConnectionExtractor rides the same transitive-helper collection (VertxSqlClientDecorator is an explicit helper).

Follow-ups

• dbInstance/dbHostname peel — gated on the declarative-derivation layer (their values feed setServiceName, a computed side-effect, not just a tag). Until that lands, onConnection (and its 2-arg overload) stays as the wrapper around that derivation.
• Spock→JUnit migration of the DatabaseClientDecoratorTest cascade.

🤖 Generated with Claude Code

[datadog-datadog-us1-prod]

🎯 Code Coverage (details)
• Patch Coverage: 66.67%
• Overall Coverage: 54.94% (-0.01%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 9bb193f | Docs | Datadog PR Page | Give us feedback!}

[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	13.81 s	13.98 s	[-1.9%; -0.5%] (maybe better)
startup:insecure-bank:tracing:Agent	12.71 s	13.00 s	[-3.0%; -1.6%] (significantly better)
startup:petclinic:appsec:Agent	16.62 s	16.74 s	[-1.9%; +0.4%] (no difference)
startup:petclinic:iast:Agent	16.54 s	16.89 s	[-3.1%; -1.1%] (significantly better)
startup:petclinic:profiling:Agent	16.41 s	16.74 s	[-2.8%; -1.1%] (significantly better)
startup:petclinic:sca:Agent	16.54 s	16.77 s	[-2.4%; -0.4%] (maybe better)
startup:petclinic:tracing:Agent	15.32 s	15.84 s	[-9.3%; +2.7%] (unstable)

Commit: 9bb193fc · 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.