[APPS-2792] Wire local Node execution into the real dev server

[tyffical]

Motivation

• Milestone 2 of the local Node execution kickoff plan — wires the fork/IPC path [APPS-2792] Prototype: local Node execution for backend functions #461 prototyped and [APPS-2792] Harden local Node execution: signal reporting, shutdown cleanup, edge-case coverage #471 hardened into the real npm run dev request path for the first time. Before this PR, local-execution.ts was fully built and tested but unreachable from any actual dev-server request.
• Design doc: Local Node execution for High Code Apps backend functions — Rollout section: "no toggle at all — npm run dev unconditionally uses local execution; the cloud round-trip becomes a separately-purposed command."
• APPS-2792

Architecture

Browser (frontend)
   │  executeBackendFunction(name, args)
   ▼
Vite dev-server middleware
   │
   ├──────────────────────────┬───────────────────────────────┐
   ▼                          ▼
POST /__dd/executeAction      POST /__dd/executeActionViaCloud
(no auth required)            (auth required, unchanged)
   │                          │
   ▼                          ▼
executeScriptLocally()        executeScriptViaDatadog()
(#461/#471 — forked           (existing preview-async +
 child process, real          long-poll round trip against
 OS-process isolation)        the real Deno prod runtime)
   │                          │
   └──────────┬───────────────┘
              ▼
   Same ExecuteActionResponse contract either way:
   {success:true, result:{data}} | {success:false, error}

/__dd/executeAction is the customer-facing default and is now unconditionally local — no config field, no flag, per the design doc's rollout decision. /__dd/executeActionViaCloud is new and additive: it preserves today's exact cloud round-trip (same executeScriptViaDatadog/pollQueryExecution code, untouched) under a distinctly-purposed endpoint. No npm run dev:verify CLI command exists yet to give customers an ergonomic way to reach it — that's now tracked as Milestone 3 in the kickoff doc, not deferred — and the existing test suite has deep coverage of this path that would otherwise be silently orphaned.

Changes

What changed	File
/__dd/executeAction now runs backend functions via executeScriptLocally (a real forked child process) instead of the cloud round-trip	packages/plugins/apps/src/vite/dev-server.ts
/__dd/executeAction requires no auth, since local execution's $.Actions calls are stubbed today (see Follow-ups)	packages/plugins/apps/src/vite/dev-server.ts
New /__dd/executeActionViaCloud endpoint preserves the original executeScriptViaDatadog cloud round-trip byte-for-byte, including its auth gate	packages/plugins/apps/src/vite/dev-server.ts
Startup log warning updated to reference /__dd/executeActionViaCloud (the endpoint that actually needs credentials) instead of the old /__dd/executeAction name	packages/plugins/apps/src/vite/dev-server.ts
Wire killAllLocalExecutionChildren() (from #471) into the dev server's own shutdown via server.httpServer's close event	packages/plugins/apps/src/vite/index.ts
Renamed the entire original cloud-round-trip test suite to target /__dd/executeActionViaCloud instead of /__dd/executeAction, preserving full existing coverage	packages/plugins/apps/src/vite/dev-server.test.ts
Fixed the routing-level test for /__dd/executeAction to assert against real local-execution bundle content instead of a stale nock mock	packages/plugins/apps/src/vite/dev-server.test.ts
Added a companion routing test for /__dd/executeActionViaCloud	packages/plugins/apps/src/vite/dev-server.test.ts
Added a new 5-test suite covering the local-execution handler directly (missing functionRef, unknown function, real forked-child result, crash-as-500, no-credentials-configured)	packages/plugins/apps/src/vite/dev-server.test.ts
Register local-exec-child.js in buildPlugin.toBuild so it's copied into the published dist/src output next to the compiled bundle	packages/plugins/apps/package.json
Local execution's $ context now includes a Source: { initiator, runAsUser } object with a synthetic local-dev identity, so @datadog/apps-backend's injected runtime-context setup no longer throws	packages/plugins/apps/src/vite/local-exec-child.js
New integration test reusing the apps_backend_project fixture (real @datadog/apps-backend) confirms a function calling getExecutionUser()/getInitiatingUser() runs cleanly under local execution	packages/plugins/apps/src/vite/local-execution.integration.test.ts

QA Instructions

yarn install

BUILD_PLUGINS_ENV=test FORCE_COLOR=true JEST_CONFIG_TRANSPILE_ONLY=true VITE_CJS_IGNORE_WARNING=true NODE_OPTIONS="--experimental-vm-modules" yarn workspace @dd/tests exec jest --config jest.config.ts packages/plugins/apps/src/vite/dev-server.test.ts
# Expected: 26 tests passing ✅ VERIFIED

BUILD_PLUGINS_ENV=test FORCE_COLOR=true JEST_CONFIG_TRANSPILE_ONLY=true VITE_CJS_IGNORE_WARNING=true NODE_OPTIONS="--experimental-vm-modules" yarn workspace @dd/tests exec jest --config jest.config.ts packages/plugins/apps
# Expected: 24 test suites, 298 tests, all passing ✅ VERIFIED

cd ~/dd/build-plugins/packages/plugins/apps && npx tsc -b tsconfig.client.json --emitDeclarationOnly && npx tsc --noEmit
# Expected: no output, clean exit ✅ VERIFIED

yarn eslint packages/plugins/apps/src/vite/dev-server.ts packages/plugins/apps/src/vite/dev-server.test.ts packages/plugins/apps/src/vite/index.ts
# Expected: no output, clean exit ✅ VERIFIED

Manual QA — real dev server, both endpoints

Setup (one app covers every scenario below — @datadog/apps-backend and @datadog/action-catalog are scaffold defaults). Three terminals are used in total: Terminal A builds this branch and is done after that; Terminal B runs the dev server and stays occupied by it; Terminal C is opened later, once the dev server is running, to send requests to it.

Terminal A — folder: this repo (~/dd/build-plugins):

cd ~/dd/build-plugins
yarn install
yarn workspace @datadog/vite-plugin build

Terminal A isn't needed again after this — the built plugin is linked by absolute path below, not by anything left running in this terminal.

Terminal B — folder: a fresh scaffolded app outside this repo (~/qa-app):

# 1. Scaffold a fresh test app under $HOME
npm create @datadog/apps@latest -- ~/qa-app --template vite-react -y
cd ~/qa-app

# 2. Link Terminal A's build in
npm link ~/dd/build-plugins/packages/published/vite-plugin
# If this resolves to unbuilt TS source instead of dist (npm link's known gap with this
# package's workspace-mode `exports` field), edit vite.config.ts's import instead:
#   import { datadogVitePlugin } from '@datadog/vite-plugin/dist/src';

Still Terminal B, still in ~/qa-app:

# 3. Add QA backend functions
cat > src/qaCheck.backend.ts << 'EOF'
export async function qaCheck(n: number) {
    return { doubled: n * 2 };
}
EOF

cat > src/qaIdentity.backend.ts << 'EOF'
import { getExecutionUser, getInitiatingUser } from '@datadog/apps-backend/user';
export async function qaIdentity() {
    const [executionUser, initiatingUser] = await Promise.all([getExecutionUser(), getInitiatingUser()]);
    return { executionUser, initiatingUser };
}
EOF

cat > src/slowCheck.backend.ts << 'EOF'
export async function slowCheck() {
    await new Promise((r) => setTimeout(r, 15000));
    return { done: true };
}
EOF

cat > src/streamLogs.backend.ts << 'EOF'
export async function streamLogs() {
    console.log('[qa] step 1: starting');
    await new Promise((r) => setTimeout(r, 1000));
    console.log('[qa] step 2: one second in');
    await new Promise((r) => setTimeout(r, 1000));
    console.log('[qa] step 3: done');
    return { done: true };
}
EOF

cat > src/doubleNumber.backend.ts << 'EOF'
export async function doubleNumber(n: number) {
    return { doubled: n * 2 };
}
EOF

# Wire the imports so Vite's backend-function transform discovers them (prepended, portable across sed dialects)
{ printf "import './qaCheck.backend';\nimport './qaIdentity.backend';\nimport './slowCheck.backend';\nimport './streamLogs.backend';\nimport './doubleNumber.backend';\n\n"; cat src/App.tsx; } > /tmp/App.tsx.new && mv /tmp/App.tsx.new src/App.tsx

Still Terminal B, still in ~/qa-app — this command blocks and keeps running, so leave it in the foreground here:

# 4. Start local execution — no dd-auth, fixed port
npm run dev -- --port 5180 --strictPort

Terminal C — open a new terminal, folder: ~/qa-app again (Terminal B is now occupied by the running dev server; everything below runs here instead):

cd ~/qa-app

# 5. Warm up the module graph once (Vite transforms lazily on first request)
curl -sS http://localhost:5180/ > /dev/null
curl -sS http://localhost:5180/src/main.tsx > /dev/null

# 6. Capture each function's real hashed name (registered name is {hash}.{exportName}, not the bare export name) — extracted automatically, no copy-paste needed
QACHECK=$(curl -sS http://localhost:5180/src/qaCheck.backend.ts | sed -n 's/.*executeBackendFunction("\([^"]*\)".*/\1/p')
QAIDENTITY=$(curl -sS http://localhost:5180/src/qaIdentity.backend.ts | sed -n 's/.*executeBackendFunction("\([^"]*\)".*/\1/p')
STREAMLOGS=$(curl -sS http://localhost:5180/src/streamLogs.backend.ts | sed -n 's/.*executeBackendFunction("\([^"]*\)".*/\1/p')
DOUBLENUMBER=$(curl -sS http://localhost:5180/src/doubleNumber.backend.ts | sed -n 's/.*executeBackendFunction("\([^"]*\)".*/\1/p')
SLOWCHECK=$(curl -sS http://localhost:5180/src/slowCheck.backend.ts | sed -n 's/.*executeBackendFunction("\([^"]*\)".*/\1/p')
echo "$QACHECK / $QAIDENTITY / $STREAMLOGS / $DOUBLENUMBER / $SLOWCHECK"
# Expected: five non-empty "<hash>.<exportName>" strings, space-separated ✅ VERIFIED

Local execution tests:

curl -sS -X POST http://localhost:5180/__dd/executeAction \
  -H 'Content-Type: application/json' \
  -d "{\"functionName\":\"$QACHECK\",\"args\":[21]}"
# Expected: {"success":true,"result":{"data":{"doubled":42}}} ✅ VERIFIED

curl -sS -X POST http://localhost:5180/__dd/executeAction \
  -H 'Content-Type: application/json' \
  -d "{\"functionName\":\"$QAIDENTITY\",\"args\":[]}"
# Expected: {"success":true,"result":{"data":{"executionUser":{"id":"local-dev-user","orgId":"local-dev-org","email":null,"name":"Local Development"},"initiatingUser":{...}}}} ✅ VERIFIED

curl -sS -X POST http://localhost:5180/__dd/executeAction \
  -H 'Content-Type: application/json' \
  -d "{\"functionName\":\"$STREAMLOGS\",\"args\":[]}"
# Expected: {"success":true,"result":{"data":{"done":true}}} in the response, and the three [qa] step lines print live in the dev-server terminal, one second apart ✅ VERIFIED

Timing check:

curl -sS -X POST http://localhost:5180/__dd/executeAction -H 'Content-Type: application/json' \
  -d "{\"functionName\":\"$DOUBLENUMBER\",\"args\":[21]}" -w '\n⏱  %{time_total}s\n'
# Expected: {"success":true,"result":{"data":{"doubled":42}}}, ⏱ ~0.1s ✅ VERIFIED (0.111s)

Shutdown cleanup — no orphaned child process (still Terminal C, until the Ctrl-C note below):

curl -sS -X POST http://localhost:5180/__dd/executeAction -H 'Content-Type: application/json' \
  -d "{\"functionName\":\"$SLOWCHECK\",\"args\":[]}" &
sleep 1 && pgrep -f local-exec-child.js
# Expected: one matching PID (child running mid-request) ✅ VERIFIED

Switch to Terminal B (where npm run dev is still running in the foreground) and press Ctrl-C to stop it. Then back in Terminal C:

pgrep -f local-exec-child.js
# Expected: no output (child was killed with the dev server) ✅ VERIFIED

Cloud round trip — the same $QACHECK/$DOUBLENUMBER/$STREAMLOGS variables from step 6 stay valid — the registered name only depends on the file's relative path, not the running server instance.

Terminal B — folder: still ~/qa-app — restart with credentials, fixed port:

dd-auth --domain dd.datad0g.com -- npm run dev -- --port 5182 --strictPort

Terminal C — folder: still ~/qa-app:

curl -sS -X POST http://localhost:5182/__dd/executeActionViaCloud \
  -H 'Content-Type: application/json' \
  -d "{\"functionName\":\"$QACHECK\",\"args\":[21]}"
# Expected: {"success":true,"result":{"data":{"doubled":42}}} ✅ VERIFIED

# same server instance — confirms both endpoints coexist
curl -sS -X POST http://localhost:5182/__dd/executeAction \
  -H 'Content-Type: application/json' \
  -d "{\"functionName\":\"$QACHECK\",\"args\":[7]}"
# Expected: {"success":true,"result":{"data":{"doubled":14}}} ✅ VERIFIED

curl -sS -X POST http://localhost:5182/__dd/executeActionViaCloud -H 'Content-Type: application/json' \
  -d "{\"functionName\":\"$DOUBLENUMBER\",\"args\":[21]}" -w '\n⏱  %{time_total}s\n'
# Expected: {"success":true,"result":{"data":{"doubled":42}}}, ⏱ multi-hundred-ms (real network + queue overhead) ✅ VERIFIED (0.725s)

curl -sS -X POST http://localhost:5182/__dd/executeActionViaCloud \
  -H 'Content-Type: application/json' \
  -d "{\"functionName\":\"$STREAMLOGS\",\"args\":[]}"
# Expected: the three [qa] step lines print all at once, only after the response lands — not live ✅ VERIFIED

Action-catalog build-time validation — no server needed, so Ctrl-C Terminal B's dev server first. Runs in either terminal, folder ~/qa-app:

echo "import { postMessage } from '@datadog/action-catalog/slack/chat';" > src/_check.backend.ts
npx tsc --noEmit
# Expected: TS2305 "has no exported member 'postMessage'" ✅ VERIFIED
rm src/_check.backend.ts

Blast Radius

• This is the first PR in the stack that changes shipping behavior: npm run dev's /__dd/executeAction now executes locally by default, unconditionally, per the design doc's rollout decision — no feature flag, matching the "no toggle" recommendation (Datadog's own internal canary/experiment gate, if wanted before this becomes the default for all customers, is out of scope for this PR; see design doc's Rollout section).
• The existing cloud round-trip is fully preserved and still reachable, just under a new endpoint name (/__dd/executeActionViaCloud) — no existing caller of the old /__dd/executeAction endpoint is silently broken without an equivalent path forward, though callers relying specifically on cloud execution at the old URL will need to switch to the new one.
• $.Actions calls made during local execution remain stubbed (per [APPS-2792] Prototype: local Node execution for backend functions #461/[APPS-2792] Harden local Node execution: signal reporting, shutdown cleanup, edge-case coverage #471) — real action side effects are only available via /__dd/executeActionViaCloud today. This is a known, called-out gap tracked against the single-action-execution endpoint dependency, not something this PR silently papers over.
• The @datadog/apps-backend Source-context gap found during this PR's manual QA (see above) is now fixed and re-verified against a real scaffolded app — local execution is safe as the default for apps using that package. The synthetic local-dev identity means getExecutionUser()/getInitiatingUser() won't reflect a real user locally (no live-session identity is available without the credentials local execution deliberately doesn't require) — acceptable for now since nothing in the design doc's scope depends on that identity being real, but worth knowing if a backend function's own logic ever branches on it.
• Risk: medium. Low risk to code correctness (thoroughly tested, real manual QA above, including the two gaps found and fixed) but a genuine behavior change to a customer-facing default — flagged as the "no toggle" decision the design doc argues for explicitly, not an accidental side effect.

Out of Scope / Follow-ups

Item	Status	Next step
Local execution's $.Source uses a synthetic local-dev identity, not a real user's — getExecutionUser()/getInitiatingUser() won't reflect real identity locally	Independent	No whoami-style endpoint exists in this package's auth layer today; fetching real identity via the already-authenticated parent process (when credentials are configured) is a real, separate enhancement, not a blocker — nothing in scope today depends on the identity being real
npm run dev:verify as a distinctly-named command for the cloud round-trip	Tracked (Milestone 3)	Now a fully in-scope, gating deliverable — losing easy cloud-path access the moment npm run dev defaults to local execution is a real workflow regression, not just a missing nicety. Two PRs needed: mode-aware routing in this package (packages/plugins/apps/), plus a create-apps template change in web-ui (owned by @DataDog/app-builder-high-code) to add the script and an example backend function. /__dd/executeActionViaCloud is reachable today via a manual curl call in the meantime.
Backward-compatible rollout (bump.yaml major bump + MIGRATIONS.md entry)	Not started	Now tracked under Product Launch (not a standalone milestone), gated on Milestones 3/4/5/6 landing (Milestone 7, telemetry, expected in place too), per the kickoff doc
Real (non-stubbed) $.Actions execution during local runs	Blocked	Same dependency as #461/#471 — see the RFC's Open Questions section for the REST-vs-MCP comparison
Internal canary/experiment gate for gradual rollout before this is unconditional for all customers	Not started	Design doc's Rollout section proposes this as a Datadog-internal gate, invisible to customers — separate from this PR's scope

Documentation

• Kickoff doc — Milestone 2
• Design doc