acc: Add test reproducing engine=unknown deploy telemetry

acceptance/bundle/telemetry/deploy-failed-engine-unknown/databricks.yml

@@ -0,0 +1,15 @@
+bundle:
+  name: test-bundle
+  uuid: 11111111-2222-3333-4444-555555555555
+
+# Restrict the sync to a single file under src/ so the file upload (and thus the
+# forced failure below) is deterministic and does not sweep up acceptance-harness
+# files written during the run.
+sync:
+  paths:
+    - src
+
+resources:
+  jobs:
+    job_one:
+      name: job one

acceptance/bundle/telemetry/deploy-failed-engine-unknown/output.txt

@@ -0,0 +1,29 @@
+
+>>> [CLI] bundle deploy
+Error: Failed to acquire deployment lock: Internal error
+Error: Internal error (500 INTERNAL_ERROR)
+
+Endpoint: POST [DATABRICKS_URL]/api/2.0/workspace-files/import-file/Workspace%2FUsers%2F[USERNAME]%2F.bundle%2Ftest-bundle%2Fdefault%2Fstate%2Fdeploy.lock?overwrite=false
+HTTP Status: 500 Internal Server Error
+API error_code: INTERNAL_ERROR
+API message: Internal error
+
+
+Exit code: 1
+
+>>> cat out.requests.txt
+
+=== resources_metadata is reported without state_engine on a failed deploy
+>>> cat telemetry.json
+{
+  "resources": [
+    {
+      "resource_type": "jobs",
+      "count": 1
+    }
+  ]
+}
+
+=== state_engine is null even though the engine was resolved before the failure
+>>> cat telemetry.json
+null

acceptance/bundle/telemetry/deploy-failed-engine-unknown/script

@@ -0,0 +1,16 @@
+errcode trace $CLI bundle deploy
+
+trace cat out.requests.txt | jq 'select(has("path") and .path == "/telemetry-ext") | .body.protoLogs[] | fromjson' > telemetry.json
+
+# resources_metadata is present (per-type counts come from the config), but there is
+# no state_engine key: the deploy failed before deployCore ran, so b.Metrics.StateEngine
+# is still the zero value and its omitempty drops it. This absence is the "unknown
+# engine" the dashboard reports for failed/cancelled deploys.
+title "resources_metadata is reported without state_engine on a failed deploy"
+trace cat telemetry.json | jq '.entry.databricks_cli_log.bundle_deploy_event.resources_metadata'
+
+title "state_engine is null even though the engine was resolved before the failure"
+trace cat telemetry.json | jq '.entry.databricks_cli_log.bundle_deploy_event.resources_metadata.state_engine'
+
+rm out.requests.txt
+rm telemetry.json

acceptance/bundle/telemetry/deploy-failed-engine-unknown/src/file.txt

@@ -0,0 +1 @@
+hello

acceptance/bundle/telemetry/deploy-failed-engine-unknown/test.toml

@@ -0,0 +1,23 @@
+# Reproduces the "unknown engine" deploy telemetry.
+#
+# A deploy that fails before it starts applying resources (here it fails while
+# acquiring the deployment lock) still emits a bundle_deploy_event via the deferred
+# LogDeployTelemetry in cmd/bundle/utils/process.go. But b.Metrics.StateEngine is
+# assigned only inside deployCore (bundle/phases/deploy.go), which never runs on this
+# path, so it is still the zero value. resources_metadata is therefore reported with
+# per-type counts (which come from the config) but with no state_engine field, and
+# the dashboard surfaces that null as engine="unknown" - even though the engine was
+# fully resolved (ResolveEngineSetting) long before the deploy failed.
+#
+# The output is engine-agnostic: the lock is acquired in the shared deploy phase
+# before any engine-specific plan/apply, and a failed deploy records no per-resource
+# state sizes for either engine.
+#
+# The stubbed endpoint is the one the workspace filer writes through, so the first
+# write - the deploy.lock file in lock.Acquire - is what fails here.
+Cloud = false
+
+[[Server]]
+Pattern = "POST /api/2.0/workspace-files/import-file/{path...}"
+Response.StatusCode = 500
+Response.Body = '''{"error_code": "INTERNAL_ERROR", "message": "Internal error"}'''