[Exporter] Improving reliability of Emit function

I found in the large-scale testing that sometimes we don't handle emitting of the same
resource reliably, and this may lead to generation of duplicate resources (very small
amount, but still) - found this in a very specific case when notebooks were listed without
directories.

This PR fixes this problem:

- by tracking if resource is already in importing queue
- detecting duplicates during code generation

It also may improve performance a bit (2-3%).

exporter/codegen.go

@@ -904,22 +904,27 @@ func (ic *importContext) handleResourceWrite(generatedFile string, ch dataWriteC
 		return
 	}
 
-	//
 	newResources := make(map[string]struct{}, 100)
 	log.Printf("[DEBUG] started processing new writes for %s", generatedFile)
 	for f := range ch {
 		if f != nil {
-			log.Printf("[DEBUG] started writing resource body for %s", f.BlockName)
-			_, err = tf.WriteString(f.ResourceBody)
-			if err == nil {
-				newResources[f.BlockName] = struct{}{}
-				if f.ImportCommand != "" {
-					ic.waitGroup.Add(1)
-					importChan <- f.ImportCommand
+			// check if we have the same blockname already written. To avoid duplicates
+			_, exists := newResources[f.BlockName]
+			if !exists {
+				log.Printf("[DEBUG] started writing resource body for %s", f.BlockName)
+				_, err = tf.WriteString(f.ResourceBody)
+				if err == nil {
+					newResources[f.BlockName] = struct{}{}
+					if f.ImportCommand != "" {
+						ic.waitGroup.Add(1)
+						importChan <- f.ImportCommand
+					}
+					log.Printf("[DEBUG] finished writing resource body for %s", f.BlockName)
+				} else {
+					log.Printf("[ERROR] Error when writing to %s: %v", generatedFile, err)
 				}
-				log.Printf("[DEBUG] finished writing resource body for %s", f.BlockName)
 			} else {
-				log.Printf("[ERROR] Error when writing to %s: %v", generatedFile, err)
+				log.Printf("[WARN] Found duplicate resource: '%s'", f.BlockName)
 			}
 		} else {
 			log.Print("[WARN] got nil as resourceWriteData!")

exporter/context.go

@@ -204,7 +204,7 @@ var goroutinesNumber = map[string]int{
 	"databricks_sql_dashboard":     3,
 	"databricks_sql_widget":        4,
 	"databricks_sql_visualization": 4,
-	"databricks_query":             4,
+	"databricks_query":             6,
 	"databricks_alert":             2,
 	"databricks_permissions":       11,
 }
@@ -615,17 +615,20 @@ func (ic *importContext) HasInState(r *resource, onlyAdded bool) bool {
 	return ic.State.Has(r)
 }
 
-func (ic *importContext) setImportingState(s string, state bool) {
-	ic.importingMutex.Lock()
-	defer ic.importingMutex.Unlock()
-	ic.importing[s] = state
-}
-
 func (ic *importContext) Add(r *resource) {
 	if ic.HasInState(r, true) { // resource must exist and already marked as added
 		return
 	}
-	ic.setImportingState(r.String(), true) // mark resource as added
+	rString := r.String()
+	ic.importingMutex.Lock()
+	_, ok := ic.importing[rString]
+	if ok {
+		ic.importingMutex.Unlock()
+		log.Printf("[DEBUG] %s already being added", rString)
+		return
+	}
+	ic.importing[rString] = true // mark resource as added
+	ic.importingMutex.Unlock()
 	state := r.Data.State()
 	if state == nil {
 		log.Printf("[ERROR] state is nil for %s", r)
@@ -648,7 +651,6 @@ func (ic *importContext) Add(r *resource) {
 		Instances: []instanceApproximation{inst},
 		Resource:  r,
 	})
-	// in single-threaded scenario scope is toposorted
 	ic.Scope.Append(r)
 }
 
@@ -727,14 +729,25 @@ func (ic *importContext) Emit(r *resource) {
 		log.Printf("[DEBUG] %s already imported", r)
 		return
 	}
+	rString := r.String()
 	if ic.testEmits != nil {
 		log.Printf("[INFO] %s is emitted in test mode", r)
 		ic.testEmitsMutex.Lock()
-		ic.testEmits[r.String()] = true
+		ic.testEmits[rString] = true
 		ic.testEmitsMutex.Unlock()
 		return
 	}
-	ic.setImportingState(r.String(), false) // we're starting to add a new resource
+	// we need to check that we're not importing the same resource twice - this may happen under high concurrency
+	// for specific resources, for example, directories when they aren't part of the listing
+	ic.importingMutex.Lock()
+	res, ok := ic.importing[rString]
+	if ok {
+		ic.importingMutex.Unlock()
+		log.Printf("[DEBUG] %s already being imported: %v", rString, res)
+		return
+	}
+	ic.importing[rString] = false // // we're starting to add a new resource
+	ic.importingMutex.Unlock()
 	_, ok = ic.Resources[r.Resource]
 	if !ok {
 		log.Printf("[ERROR] %s is not available in provider", r)
@@ -745,8 +758,10 @@ func (ic *importContext) Emit(r *resource) {
 		log.Printf("[DEBUG] %s (%s service) is not part of the account level export", r.Resource, ir.Service)
 		return
 	}
-	// TODO: add similar condition for checking workspace-level objects only. After new ACLs import is merged
-
+	if !ic.accountLevel && !ir.WorkspaceLevel {
+		log.Printf("[DEBUG] %s (%s service) is not part of the workspace level export", r.Resource, ir.Service)
+		return
+	}
 	// from here, it should be done by the goroutine...  send resource into the channel
 	ch, exists := ic.channels[r.Resource]
 	if exists {