Direct engine fixes

acceptance/bundle/resources/registered_models/aliases_converge/databricks.yml.tmpl

@@ -0,0 +1,16 @@
+bundle:
+ name: deploy-registered-models-aliases-$UNIQUE_NAME
+
+resources:
+ registered_models:
+ my_registered_model:
+ name: my-registered-model-aliases-$UNIQUE_NAME
+ comment: $COMMENT
+ catalog_name: main
+ schema_name: default
+ # Aliases live on model versions and GET does not echo them back
+ # (DoRead uses IncludeAliases=false), so a config that sets them used to
+ # report a perpetual in-place update and never converge.
+ aliases:
+ - alias_name: champion
+ id: alias-champion

acceptance/bundle/resources/registered_models/aliases_converge/output.txt

@@ -0,0 +1,57 @@
+
+>>> export COMMENT=original comment
+
+=== Initial deployment with aliases set
+>>> [CLI] bundle deploy
+Uploading bundle files to /Workspace/Users/[USERNAME]/.bundle/deploy-registered-models-aliases-[UNIQUE_NAME]/default/files...
+Deploying resources...
+Updating deployment state...
+Deployment complete!
+
+=== Plan is a no-op: GET never echoes aliases, so they must not drift
+>>> [CLI] bundle plan
+Plan: 0 to add, 0 to change, 0 to delete, 1 unchanged
+
+=== The config-set aliases are skipped as input_only (confirms the matched rule)
+>>> [CLI] bundle plan --output json
+{
+ "action": "skip",
+ "reason": "input_only",
+ "old": [
+ {
+ "alias_name": "champion",
+ "id": "alias-champion"
+ }
+ ],
+ "new": [
+ {
+ "alias_name": "champion",
+ "id": "alias-champion"
+ }
+ ]
+}
+
+=== Edit the comment: an in-place update, not a recreate
+>>> [CLI] bundle plan
+update registered_models.my_registered_model
+
+Plan: 0 to add, 1 to change, 0 to delete, 0 unchanged
+
+>>> [CLI] bundle deploy
+Uploading bundle files to /Workspace/Users/[USERNAME]/.bundle/deploy-registered-models-aliases-[UNIQUE_NAME]/default/files...
+Deploying resources...
+Updating deployment state...
+Deployment complete!
+
+=== Re-plan after the edit converges: no further changes
+>>> [CLI] bundle plan
+Plan: 0 to add, 0 to change, 0 to delete, 1 unchanged
+
+>>> [CLI] bundle destroy --auto-approve
+The following resources will be deleted:
+ delete resources.registered_models.my_registered_model
+
+All files and directories at the following location will be deleted: /Workspace/Users/[USERNAME]/.bundle/deploy-registered-models-aliases-[UNIQUE_NAME]/default
+
+Deleting files...
+Destroy complete!

acceptance/bundle/resources/registered_models/aliases_converge/script

@@ -0,0 +1,27 @@
+echo "*" > .gitignore
+
+trace export COMMENT="original comment"
+envsubst < databricks.yml.tmpl > databricks.yml
+
+cleanup() {
+ trace $CLI bundle destroy --auto-approve
+}
+trap cleanup EXIT
+
+title "Initial deployment with aliases set"
+trace $CLI bundle deploy
+
+title "Plan is a no-op: GET never echoes aliases, so they must not drift"
+trace $CLI bundle plan | contains.py "Plan: 0 to add, 0 to change, 0 to delete, 1 unchanged"
+
+title "The config-set aliases are skipped as input_only (confirms the matched rule)"
+trace $CLI bundle plan --output json | jq '.plan[].changes.aliases'
+
+title "Edit the comment: an in-place update, not a recreate"
+export COMMENT="updated comment"
+envsubst < databricks.yml.tmpl > databricks.yml
+trace $CLI bundle plan | contains.py "Plan: 0 to add, 1 to change, 0 to delete"
+trace $CLI bundle deploy
+
+title "Re-plan after the edit converges: no further changes"
+trace $CLI bundle plan | contains.py "Plan: 0 to add, 0 to change, 0 to delete, 1 unchanged"

acceptance/bundle/resources/registered_models/aliases_converge/test.toml

@@ -0,0 +1,5 @@
+RecordRequests = false
+
+# The aliases input_only classification lives in the direct engine, so scope
+# this regression there.
+EnvMatrix.DATABRICKS_BUNDLE_ENGINE = ["direct"]

acceptance/bundle/resources/sql_warehouses/lifecycle-started-edit/databricks.yml.tmpl

@@ -0,0 +1,14 @@
+bundle:
+ name: lifecycle-started-edit-$UNIQUE_NAME
+
+workspace:
+ root_path: ~/.bundle/$UNIQUE_NAME
+
+resources:
+ sql_warehouses:
+ mywarehouse:
+ name: $UNIQUE_NAME
+ cluster_size: "2X-Small"
+ auto_stop_mins: 10
+ lifecycle:
+ started: false

acceptance/bundle/resources/sql_warehouses/lifecycle-started-edit/output.txt

@@ -0,0 +1,67 @@
+
+=== Deploy with started=false: warehouse created and then stopped
+>>> errcode [CLI] bundle deploy
+Uploading bundle files to /Workspace/Users/[USERNAME]/.bundle/[UNIQUE_NAME]/files...
+Deploying resources...
+Updating deployment state...
+Deployment complete!
+
+>>> print_requests.py //sql/warehouses
+{
+ "method": "POST",
+ "path": "/api/2.0/sql/warehouses",
+ "body": {
+ "auto_stop_mins": 10,
+ "cluster_size": "2X-Small",
+ "enable_photon": true,
+ "max_num_clusters": 1,
+ "name": "[UNIQUE_NAME]",
+ "spot_instance_policy": "COST_OPTIMIZED"
+ }
+}
+{
+ "method": "POST",
+ "path": "/api/2.0/sql/warehouses/[WAREHOUSE_ID]/stop"
+}
+
+>>> errcode [CLI] warehouses get [WAREHOUSE_ID]
+"STOPPED"
+
+=== Edit a field with started=false: Edit restarts the warehouse, so a Stop must follow
+>>> update_file.py databricks.yml auto_stop_mins: 10 auto_stop_mins: 20
+
+>>> errcode [CLI] bundle deploy
+Uploading bundle files to /Workspace/Users/[USERNAME]/.bundle/[UNIQUE_NAME]/files...
+Deploying resources...
+Updating deployment state...
+Deployment complete!
+
+>>> print_requests.py //sql/warehouses
+{
+ "method": "POST",
+ "path": "/api/2.0/sql/warehouses/[WAREHOUSE_ID]/edit",
+ "body": {
+ "auto_stop_mins": 20,
+ "cluster_size": "2X-Small",
+ "enable_photon": true,
+ "max_num_clusters": 1,
+ "name": "[UNIQUE_NAME]",
+ "spot_instance_policy": "COST_OPTIMIZED"
+ }
+}
+{
+ "method": "POST",
+ "path": "/api/2.0/sql/warehouses/[WAREHOUSE_ID]/stop"
+}
+
+>>> errcode [CLI] warehouses get [WAREHOUSE_ID]
+"STOPPED"
+
+>>> [CLI] bundle destroy --auto-approve
+The following resources will be deleted:
+ delete resources.sql_warehouses.mywarehouse
+
+All files and directories at the following location will be deleted: /Workspace/Users/[USERNAME]/.bundle/[UNIQUE_NAME]
+
+Deleting files...
+Destroy complete!

acceptance/bundle/resources/sql_warehouses/lifecycle-started-edit/script

@@ -0,0 +1,25 @@
+envsubst < databricks.yml.tmpl > databricks.yml
+
+cleanup() {
+ trace $CLI bundle destroy --auto-approve
+ rm -f out.requests.txt
+}
+trap cleanup EXIT
+
+title "Deploy with started=false: warehouse created and then stopped"
+trace errcode $CLI bundle deploy
+WAREHOUSE_ID=$($CLI bundle summary -o json | jq -r '.resources.sql_warehouses.mywarehouse.id')
+add_repl.py "$WAREHOUSE_ID" "WAREHOUSE_ID"
+trace print_requests.py //sql/warehouses
+rm -f out.requests.txt
+{ trace errcode $CLI warehouses get "$WAREHOUSE_ID" | jq '.state'; } || true
+
+# Editing a warehouse restarts it: Edit's long-running op waits for RUNNING
+# (SDK WaitGetWarehouseRunning), so a started=false config must follow the Edit
+# with a Stop. Before the fix this redeploy hung waiting for STOPPED.
+title "Edit a field with started=false: Edit restarts the warehouse, so a Stop must follow"
+trace update_file.py databricks.yml "auto_stop_mins: 10" "auto_stop_mins: 20"
+trace errcode $CLI bundle deploy
+trace print_requests.py //sql/warehouses
+rm -f out.requests.txt
+{ trace errcode $CLI warehouses get "$WAREHOUSE_ID" | jq '.state'; } || true

acceptance/bundle/resources/sql_warehouses/lifecycle-started-edit/test.toml

@@ -0,0 +1,11 @@
+Local = true
+RecordRequests = true
+
+# Starting warehouses is slow, so run on cloud nightly (CloudSlow) instead of every PR
+# to confirm the real Edit-restarts behaviour without per-PR timeouts.
+CloudSlow = true
+
+Ignore = [".databricks", "databricks.yml"]
+
+[EnvMatrix]
+ DATABRICKS_BUNDLE_ENGINE = ["direct"]

bundle/direct/dresources/resources.yml

@@ -346,6 +346,11 @@ resources:
  # Backend-computed; the user never sets it. Not annotated output_only in the spec.
  - field: browse_only
  reason: output_only
+ # Aliases are managed on model versions through a separate API, and DoRead
+ # passes IncludeAliases=false, so GET never echoes them back. Without this a
+ # config that sets aliases reports a perpetual update (remote stays empty).
+ - field: aliases
+ reason: input_only
  provided_id_fields:
  # The name can technically be updated without recreate. We recreate for now though
  # to match TF implementation.

bundle/direct/dresources/sql_warehouse.go

@@ -134,7 +134,8 @@ func hasWarehouseChanges(entry *PlanEntry) bool {
 
 // DoUpdate updates the warehouse in place.
 func (r *ResourceSqlWarehouse) DoUpdate(ctx context.Context, id string, config *SqlWarehouseState, entry *PlanEntry) (*SqlWarehouseRemote, error) {
-	if hasWarehouseChanges(entry) {
+	edited := hasWarehouseChanges(entry)
+	if edited {
  request := sql.EditWarehouseRequest{
  AutoStopMins: config.AutoStopMins,
  Channel: config.Channel,
@@ -169,6 +170,12 @@ func (r *ResourceSqlWarehouse) DoUpdate(ctx context.Context, id string, config *
 
 	desiredStarted := *config.Lifecycle.Started
 	alreadyRunning := remoteWarehouseIsRunning(entry)
+	if edited {
+ // Edit restarts the warehouse: its long-running op waits for RUNNING
+ // (SDK WaitGetWarehouseRunning), so reconcile from RUNNING, not the
+ // stale pre-plan state.
+ alreadyRunning = true
+	}
 	if desiredStarted && !alreadyRunning {
  // lifecycle.started=true: fire Start; WaitAfterUpdate polls for RUNNING.
  _, err := r.client.Warehouses.Start(ctx, sql.StartRequest{Id: id})