feat(webapp): add RUNTIME_API_ORIGIN to decouple runner traffic from external origin

.server-changes/runtime-api-origin.md

@@ -0,0 +1,6 @@
+---
+area: webapp
+type: feature
+---
+
+Add `RUNTIME_API_ORIGIN` env var to route managed runner traffic through an in-cluster URL, bypassing tracing gateways that rewrite the W3C `traceparent` header and break parent→child run links.

apps/webapp/app/env.server.ts

@@ -122,6 +122,21 @@ const EnvironmentSchema = z
     LOGIN_RATE_LIMITS_ENABLED: BoolEnv.default(true),
     APP_ORIGIN: z.string().default("http://localhost:3030"),
     API_ORIGIN: z.string().optional(),
+    // Origin that the webapp publishes to MANAGED (deployed) runner pods as
+    // both `TRIGGER_API_URL` and (as the first fallback) `TRIGGER_STREAM_URL`.
+    // When self-hosting behind a tracing-enabled gateway (Envoy/Istio/etc.)
+    // that rewrites the W3C `traceparent` on egress, point this at an
+    // in-cluster service URL so runner-to-webapp traffic stays inside the
+    // cluster and the parent->child run link in the trace tree is preserved.
+    // Intentionally NOT used for dev (CLI) task runs, which usually run on a
+    // developer's machine outside the cluster and would lose connectivity if
+    // forced onto an in-cluster URL. Empty string is normalized to unset so
+    // blank `${RUNTIME_API_ORIGIN:-}` passthroughs from caller environments
+    // don't short-circuit the `??` fallback chain.
+    RUNTIME_API_ORIGIN: z
+      .string()
+      .optional()
+      .transform((v) => v || undefined),
     STREAM_ORIGIN: z.string().optional(),
     ELECTRIC_ORIGIN: z.string().default("http://localhost:3060"),
     // A comma separated list of electric origins to shard into different electric instances by environmentId

apps/webapp/app/v3/environmentVariables/environmentVariablesRepository.server.ts

@@ -980,6 +980,13 @@ function renameVariables(variables: EnvironmentVariable[], renameMap: Record<str
   });
 }
 
+/**
+ * Resolves trigger-side built-in environment variables that are merged into a
+ * task run's env but can be overridden by user-defined variables on the
+ * environment. Values come from server-level `env` and are surfaced to runners
+ * so they can pick up rollouts (e.g. the realtime stream version) without a
+ * redeploy.
+ */
 async function resolveOverridableTriggerVariables(
   runtimeEnvironment: RuntimeEnvironmentForEnvRepo
 ) {
@@ -993,6 +1000,14 @@ async function resolveOverridableTriggerVariables(
   return result;
 }
 
+/**
+ * Resolves built-in environment variables that are injected into dev (CLI) task
+ * runs. Dev CLI typically runs on a developer's machine outside any cluster,
+ * so the runner-bypass `RUNTIME_API_ORIGIN` (which usually points at an
+ * in-cluster service URL) is intentionally NOT applied here -- using it would
+ * make the URL unreachable for the dev CLI. Dev keeps the original
+ * `API_ORIGIN`/`STREAM_ORIGIN`/`APP_ORIGIN` chain.
+ */
 async function resolveBuiltInDevVariables(runtimeEnvironment: RuntimeEnvironmentForEnvRepo) {
   let result: Array<EnvironmentVariable> = [
     {
@@ -1119,6 +1134,12 @@ async function resolveBuiltInDevVariables(runtimeEnvironment: RuntimeEnvironment
   return [...result, ...commonVariables];
 }
 
+/**
+ * Resolves the OpenTelemetry collector endpoint advertised to dev (CLI) task
+ * runs. Defaults to the webapp's own `/otel` route under `APP_ORIGIN` so a
+ * vanilla self-host works without extra wiring; `DEV_OTEL_EXPORTER_OTLP_ENDPOINT`
+ * can override it to point spans/logs at an external collector.
+ */
 async function resolveOverridableOtelDevVariables(
   runtimeEnvironment: RuntimeEnvironmentForEnvRepo
 ) {
@@ -1132,6 +1153,15 @@ async function resolveOverridableOtelDevVariables(
   return result;
 }
 
+/**
+ * Resolves built-in environment variables that are injected into managed
+ * (deployed) task runs. `TRIGGER_API_URL` and `TRIGGER_STREAM_URL` prefer
+ * `RUNTIME_API_ORIGIN` over `API_ORIGIN`/`STREAM_ORIGIN` so self-hosted
+ * deployments can keep runner-to-webapp traffic on a cluster-internal hop
+ * (bypassing tracing-enabled gateways that rewrite the W3C `traceparent`
+ * header on egress) without affecting the public origins exposed to external
+ * clients.
+ */
 async function resolveBuiltInProdVariables(
   runtimeEnvironment: RuntimeEnvironmentForEnvRepo,
   parentEnvironment?: RuntimeEnvironmentForEnvRepo
@@ -1143,11 +1173,11 @@ async function resolveBuiltInProdVariables(
     },
     {
       key: "TRIGGER_API_URL",
-      value: env.API_ORIGIN ?? env.APP_ORIGIN,
+      value: env.RUNTIME_API_ORIGIN ?? env.API_ORIGIN ?? env.APP_ORIGIN,
     },
     {
       key: "TRIGGER_STREAM_URL",
-      value: env.STREAM_ORIGIN ?? env.API_ORIGIN ?? env.APP_ORIGIN,
+      value: env.RUNTIME_API_ORIGIN ?? env.STREAM_ORIGIN ?? env.API_ORIGIN ?? env.APP_ORIGIN,
     },
     {
       key: "TRIGGER_RUNTIME_WAIT_THRESHOLD_IN_MS",

hosting/docker/.env.example

@@ -46,6 +46,16 @@ API_ORIGIN=http://localhost:8030
 DEV_OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:8030/otel
 # You may need to set this when testing locally or when using the combined setup
 # API_ORIGIN=http://webapp:3000
+# Optional: origin advertised to MANAGED (deployed) runner pods as both
+# TRIGGER_API_URL and TRIGGER_STREAM_URL (intentional: keeps all managed
+# runner traffic on the same bypass hop). Dev (CLI) task runs are NOT
+# affected -- they keep using API_ORIGIN/APP_ORIGIN so a developer running
+# `trigger.dev dev` from outside the cluster doesn't lose connectivity.
+# Set this to an in-cluster service URL when running behind a tracing-enabled
+# gateway that rewrites the W3C `traceparent` header on egress (e.g. Envoy/
+# Istio with tracing on). If you need streams on a dedicated endpoint (CDN,
+# etc.), keep RUNTIME_API_ORIGIN unset and use STREAM_ORIGIN instead.
+# RUNTIME_API_ORIGIN=http://webapp:3000
 
 # Webapp - memory management
 # - This sets the maximum memory allocation for Node.js heap in MiB (e.g. "4096" for 4GB)

hosting/docker/webapp/docker-compose.yml

@@ -43,6 +43,7 @@ services:
       APP_ORIGIN: ${APP_ORIGIN:-http://localhost:8030}
       LOGIN_ORIGIN: ${LOGIN_ORIGIN:-http://localhost:8030}
       API_ORIGIN: ${API_ORIGIN:-http://localhost:8030}
+      RUNTIME_API_ORIGIN: ${RUNTIME_API_ORIGIN:-}
       ELECTRIC_ORIGIN: http://electric:3000
       DATABASE_URL: ${DATABASE_URL:-postgresql://postgres:postgres@postgres:5432/main?schema=public&sslmode=disable}
       DIRECT_URL: ${DIRECT_URL:-postgresql://postgres:postgres@postgres:5432/main?schema=public&sslmode=disable}

hosting/k8s/helm/templates/webapp.yaml

@@ -186,6 +186,10 @@ spec:
               value: {{ .Values.webapp.loginOrigin | quote }}
             - name: API_ORIGIN
               value: {{ .Values.webapp.apiOrigin | quote }}
+            {{- with .Values.webapp.runtimeApiOrigin }}
+            - name: RUNTIME_API_ORIGIN
+              value: {{ . | quote }}
+            {{- end }}
             - name: ELECTRIC_ORIGIN
               value: {{ include "trigger-v4.electric.url" . | quote }}
             {{- if include "trigger-v4.postgres.useSecretUrl" . }}

hosting/k8s/helm/values.yaml

@@ -68,6 +68,12 @@ webapp:
   appOrigin: "http://localhost:3040"
   loginOrigin: "http://localhost:3040"
   apiOrigin: "http://localhost:3040"
+  # Origin advertised to runner pods as TRIGGER_API_URL.
+  # When unset (default), runners use apiOrigin/appOrigin. Set this to an
+  # in-cluster service URL to keep runner->webapp traffic inside the cluster,
+  # bypassing gateways/proxies (e.g. Envoy with tracing enabled) that rewrite
+  # the W3C `traceparent` header on egress and break the parent->child run link.
+  runtimeApiOrigin: ""
 
   replicaCount: 1