[fjakobs]

Why

• Agent traces never reach the MLflow experiment from an app deployed on Databricks Apps. The same app run locally does export them.
• The Apps proxy sends a W3C traceparent header with every request. AppKit's tracer provider picks up that context, so the first span of each request (the HTTP server span) has a remote parent (parentSpanContext.isRemote === true).
• @mlflow/core (0.4.0, and master as of today) treats a span as the trace root only if it has no parent at all. It ignores isRemote. See exporters/mlflow.js:36/86/151 and exporters/uc_table.js:38/89/153 in the published build.
• As a result, mlflow never registers a trace for the request, and every span logs No trace ID found for span <name>. Skipping. Nothing is exported. GatedMlflowSpanProcessor had the same root check.
• The No trace ID found warnings on plain (non-agent) requests have the same cause.
• Upstream report: [BUG] TypeScript tracing drops every trace whose root span has a remote parent (incoming traceparent) mlflow/mlflow#27545.

Change

• In GatedMlflowSpanProcessor (packages/appkit/src/plugins/agents/mlflow.ts), a span whose parent is remote now counts as the root of this process's trace.
• The gate passes mlflow a view of that span with no parent. The view is a Proxy that hides parentSpanContext and binds methods to the real span, so mlflow's writes such as setAttribute land on the real span. The real span is not changed and keeps its remote parent, so OTLP export still joins the caller's distributed trace.
• The gate now records which object it forwarded (WeakSet became WeakMap), so onStart and onEnd hand mlflow the same view. The outbound-CLIENT drop and the root check both use the new definition of root.
• The fix covers both the classic experiment processor and the Unity Catalog table processor. It is marked as a shortcut:, to be removed once upstream treats remote parents as roots.

Testing

• New unit tests in plugins/agents/tests/mlflow.test.ts. All 3 fail on main and pass with this change:
  • A deployed agent turn whose request span has a remote parent is exported. mlflow gets one parentless view at start and at end, and the real span keeps its remote parent.
  • A deployed plain request with a remote parent is discarded (popTrace). It is not left unregistered.
  • With a real SDK span started under a remote context, the view writes through to the real span and reads its live state.
• pnpm build, pnpm docs:build, pnpm check:fix, pnpm -r typecheck and pnpm test all pass.
• End to end against a Databricks workspace, using a minimal app (server() + agents() with one markdown agent on a foundation model endpoint, and an MLflow experiment bound through the agents plugin's experiment resource). The SDK was packed from this branch.

Run	Before	After
Local, no traceparent	trace exported	trace exported
Local, with traceparent	no trace, No trace ID found for span POST	trace exported, no warnings
Deployed on Databricks Apps (proxy sends traceparent)	no trace, No trace ID found for POST and the agent span	trace exported (mlflow_trace_id in the response), no warnings
Plain analytics request with traceparent	No trace ID found for every span	no trace, no warnings

• The exported trace has the right shape. The root is POST /invocations with an empty parent_span_id, and the agent span sits under it. Locally, the MLflow trace id matched the caller's traceparent trace id.
• On the deployed app, non-agent requests still create no traces, and the trace count stayed stable after the requests, so the exporters' own uploads don't loop back into traces.

This pull request and its description were written by Isaac.

[github-actions]

📦 Bundle size report

Compared against bundle-size-baseline.json (main).

@databricks/appkit

npm tarball (packed): 1.2 MB (+928 B) — gzipped download (dist + bin; excludes release-only docs/NOTICE).

dist	raw	gzip
JS (runtime)	1.3 MB (+998 B)	460 KB (+386 B)
Type declarations	465 KB	169 KB (-1 B)
Source maps	2.5 MB (+1.6 KB)	859 KB (+589 B)
Other	11 KB	3.7 KB
Total	4.2 MB (+2.6 KB)	1.5 MB (+974 B)

Per-entry composition (own code — deps external (as shipped))

Entry	Initial (gz)	Lazy (gz)	Total (gz)	node_modules (min)	Own code (min)
.	106 KB	2.5 KB	109 KB	external	344 KB
./beta	103 KB (+62 B)	482 B	104 KB (+62 B)	external	310 KB (+212 B)
./testing	45 KB	32 KB	77 KB	external	223 KB
./tsdown	520 B	0 B	520 B	external	813 B
./type-generator	26 KB	0 B	26 KB	external	74 KB

Chunks:

Entry	Chunk	Load	Size (gz)
.	index.js	initial	102 KB
.	utils.js	initial	4.6 KB
.	remote-tunnel-manager.js	lazy	2.5 KB
./beta	beta.js	initial	83 KB
./beta	modular.js	initial	6.0 KB
./beta	stream-manager.js	initial	5.9 KB
./beta	service-context.js	initial	4.2 KB
./beta	databricks.js	initial	3.3 KB
./beta	client.js	initial	647 B
./beta	index.js	initial	20 B
./beta	supervisor-api.js	lazy	192 B
./beta	databricks.js	lazy	176 B
./beta	index.js	lazy	114 B
./testing	manifest.js	initial	32 KB
./testing	index.js	initial	10 KB
./testing	wide-event-emitter.js	initial	2.9 KB
./testing	index.js	lazy	28 KB
./testing	remote-tunnel-manager.js	lazy	2.5 KB
./testing	utils.js	lazy	1.8 KB
./tsdown	index.js	initial	520 B
./type-generator	index.js	initial	26 KB

@databricks/appkit-ui

npm tarball (packed): 397 KB (-4 B) — gzipped download (dist + bin; excludes release-only docs/NOTICE).

dist	raw	gzip
JS (runtime)	429 KB	147 KB
Type declarations	267 KB	98 KB (-1 B)
Source maps	853 KB	286 KB
CSS	16 KB	3.2 KB
Total	1.5 MB	534 KB (-1 B)

Per-entry composition (consumer bundle — deps bundled, peerDeps external)

Entry	Initial (gz)	Lazy (gz)	Total (gz)	node_modules (min)	Own code (min)
./js	5.3 KB	49 KB	55 KB	208 KB	14 KB
./js/beta	2.1 KB	0 B	2.1 KB	0 B	5.0 KB
./react	433 KB	49 KB	481 KB	1.3 MB	177 KB
./react/beta	4.9 KB	0 B	4.9 KB	0 B	12 KB

Chunks:

Entry	Chunk	Load	Size (gz)
./js	index.js	initial	5.2 KB
./js	chunk	initial	120 B
./js	apache-arrow	lazy	49 KB
./js/beta	beta.js	initial	2.1 KB
./react	index.js	initial	431 KB
./react	tslib	initial	2.1 KB
./react	apache-arrow	lazy	49 KB
./react/beta	beta.js	initial	4.9 KB

[github-actions]

🤖 AppKit PR bot

🔬 Run evals

Start an eval for this PR from the evals-monitor app: Go to Evals Monitor →

📦 Try this PR's app template

Scaffolds a new app from this PR's SDK build. Run it in any folder (requires the GitHub CLI — gh auth login — and the Databricks CLI):

gh run download 42243687946 -R databricks/appkit -n appkit-template-0.90.0-pr.a54a550-fjakobs-la-2052-mlflow-traces-deployed-717 -D appkit-pr-717 \
  && unzip -o "appkit-pr-717/appkit-template-0.90.0-pr.a54a550-fjakobs-la-2052-mlflow-traces-deployed-717.zip" -d "appkit-pr-717" \
  && databricks apps init --template "appkit-pr-717"

The template pins @databricks/appkit and @databricks/appkit-ui to tarballs built from this branch, so the scaffolded app runs against this PR's code.