platform/dbops/binaries/weaviate-src/usecases/schema/class.go

//                           _       _
// __      _____  __ ___   ___  __ _| |_ ___
// \ \ /\ / / _ \/ _` \ \ / / |/ _` | __/ _ \
//  \ V  V /  __/ (_| |\ V /| | (_| | ||  __/
//   \_/\_/ \___|\__,_| \_/ |_|\__,_|\__\___|
//
//  Copyright © 2016 - 2025 Weaviate B.V. All rights reserved.
//
//  CONTACT: hello@weaviate.io
//

package schema

import (
	"context"
	"encoding/json"
	"fmt"
	"os"
	"reflect"
	"strings"

	"github.com/pkg/errors"
	"github.com/prometheus/client_golang/prometheus"
	"github.com/weaviate/weaviate/entities/modelsext"
	schemaConfig "github.com/weaviate/weaviate/entities/schema/config"
	"github.com/weaviate/weaviate/entities/vectorindex/hnsw"

	"github.com/weaviate/weaviate/adapters/repos/db/inverted/stopwords"
	cschema "github.com/weaviate/weaviate/cluster/schema"
	"github.com/weaviate/weaviate/entities/backup"
	"github.com/weaviate/weaviate/entities/classcache"
	entcfg "github.com/weaviate/weaviate/entities/config"
	"github.com/weaviate/weaviate/entities/models"
	"github.com/weaviate/weaviate/entities/replication"
	"github.com/weaviate/weaviate/entities/schema"
	"github.com/weaviate/weaviate/entities/vectorindex"
	"github.com/weaviate/weaviate/entities/versioned"
	"github.com/weaviate/weaviate/usecases/auth/authorization"
	"github.com/weaviate/weaviate/usecases/config"
	configRuntime "github.com/weaviate/weaviate/usecases/config/runtime"
	"github.com/weaviate/weaviate/usecases/monitoring"
	"github.com/weaviate/weaviate/usecases/replica"
	"github.com/weaviate/weaviate/usecases/sharding"
	shardingcfg "github.com/weaviate/weaviate/usecases/sharding/config"
)

func (h *Handler) GetClass(ctx context.Context, principal *models.Principal, name string) (*models.Class, error) {
	if err := h.Authorizer.Authorize(ctx, principal, authorization.READ, authorization.CollectionsMetadata(name)...); err != nil {
		return nil, err
	}
	name = schema.UppercaseClassName(name)

	cl := h.schemaReader.ReadOnlyClass(name)
	return cl, nil
}

func (h *Handler) GetConsistentClass(ctx context.Context, principal *models.Principal,
	name string, consistency bool,
) (*models.Class, uint64, error) {
	if err := h.Authorizer.Authorize(ctx, principal, authorization.READ, authorization.CollectionsMetadata(name)...); err != nil {
		return nil, 0, err
	}

	name = schema.UppercaseClassName(name)

	if consistency {
		vclasses, err := h.schemaManager.QueryReadOnlyClasses(name)
		return vclasses[name].Class, vclasses[name].Version, err
	}
	class, err := h.schemaReader.ReadOnlyClassWithVersion(ctx, name, 0)
	return class, 0, err
}

// GetCachedClass will return the class from the cache if it exists. Note that the context cache
// will likely be at the "request" or "operation" level and not be shared between requests.
// Uses the Handler's getClassMethod to determine how to get the class data.
func (h *Handler) GetCachedClass(ctxWithClassCache context.Context,
	principal *models.Principal, names ...string,
) (map[string]versioned.Class, error) {
	if err := h.Authorizer.Authorize(ctxWithClassCache, principal, authorization.READ, authorization.CollectionsMetadata(names...)...); err != nil {
		return nil, err
	}

	return classcache.ClassesFromContext(ctxWithClassCache, func(names ...string) (map[string]versioned.Class, error) {
		return h.classGetter.getClasses(names)
	}, names...)
}

// GetCachedClassNoAuth will return the class from the cache if it exists. Note that the context cache
// will likely be at the "request" or "operation" level and not be shared between requests.
// Uses the Handler's getClassMethod to determine how to get the class data.
func (h *Handler) GetCachedClassNoAuth(ctxWithClassCache context.Context, names ...string) (map[string]versioned.Class, error) {
	return classcache.ClassesFromContext(ctxWithClassCache, func(names ...string) (map[string]versioned.Class, error) {
		return h.classGetter.getClasses(names)
	}, names...)
}

// AddClass to the schema
func (h *Handler) AddClass(ctx context.Context, principal *models.Principal,
	cls *models.Class,
) (*models.Class, uint64, error) {
	cls.Class = schema.UppercaseClassName(cls.Class)
	cls.Properties = schema.LowercaseAllPropertyNames(cls.Properties)

	err := h.Authorizer.Authorize(ctx, principal, authorization.CREATE, authorization.CollectionsMetadata(cls.Class)...)
	if err != nil {
		return nil, 0, err
	}

	classGetterWithAuth := func(name string) (*models.Class, error) {
		if err := h.Authorizer.Authorize(ctx, principal, authorization.READ, authorization.CollectionsMetadata(name)...); err != nil {
			return nil, err
		}
		return h.schemaReader.ReadOnlyClass(name), nil
	}

	if err := h.setNewClassDefaults(cls, h.config.Replication); err != nil {
		return nil, 0, err
	}

	if err := h.validateCanAddClass(ctx, cls, classGetterWithAuth, false); err != nil {
		return nil, 0, err
	}
	// migrate only after validation in completed
	h.migrateClassSettings(cls)
	if err := h.parser.ParseClass(cls); err != nil {
		return nil, 0, err
	}

	err = h.invertedConfigValidator(cls.InvertedIndexConfig)
	if err != nil {
		return nil, 0, err
	}

	existingCollectionsCount, err := h.schemaManager.QueryCollectionsCount()
	if err != nil {
		h.logger.WithField("error", err).Error("could not query the collections count")
	}

	limit := h.schemaConfig.MaximumAllowedCollectionsCount.Get()

	if limit != config.DefaultMaximumAllowedCollectionsCount && existingCollectionsCount >= limit {
		return nil, 0, fmt.Errorf(
			"cannot create collection: maximum number of collections (%d) reached - "+
				"please consider switching to multi-tenancy or increasing the collection count limit - "+
				"see https://weaviate.io/collections-count-limit to learn about available options and best practices "+
				"when working with multiple collections and tenants",
			limit)
	}

	shardState, err := sharding.InitState(cls.Class,
		cls.ShardingConfig.(shardingcfg.Config),
		h.clusterState.LocalName(), h.schemaManager.StorageCandidates(), cls.ReplicationConfig.Factor,
		schema.MultiTenancyEnabled(cls))
	if err != nil {
		return nil, 0, errors.Wrap(err, "init sharding state")
	}

	defaultQuantization := h.config.DefaultQuantization
	h.enableQuantization(cls, defaultQuantization)

	version, err := h.schemaManager.AddClass(ctx, cls, shardState)
	if err != nil {
		return nil, 0, err
	}
	return cls, version, err
}

func (h *Handler) enableQuantization(class *models.Class, defaultQuantization *configRuntime.DynamicValue[string]) {
	compression := defaultQuantization.Get()
	if !hasTargetVectors(class) || class.VectorIndexType != "" {
		class.VectorIndexConfig = setDefaultQuantization(class.VectorIndexType, class.VectorIndexConfig.(schemaConfig.VectorIndexConfig), compression)
	}

	for k, vectorConfig := range class.VectorConfig {
		vectorConfig.VectorIndexConfig = setDefaultQuantization(class.VectorIndexType, vectorConfig.VectorIndexConfig.(schemaConfig.VectorIndexConfig), compression)
		class.VectorConfig[k] = vectorConfig
	}
}

func setDefaultQuantization(vectorIndexType string, vectorIndexConfig schemaConfig.VectorIndexConfig, compression string) schemaConfig.VectorIndexConfig {
	if len(vectorIndexType) == 0 {
		vectorIndexType = vectorindex.DefaultVectorIndexType
	}
	if vectorIndexType == vectorindex.VectorIndexTypeHNSW && vectorIndexConfig.IndexType() == vectorindex.VectorIndexTypeHNSW {
		hnswConfig := vectorIndexConfig.(hnsw.UserConfig)
		pqEnabled := hnswConfig.PQ.Enabled
		sqEnabled := hnswConfig.SQ.Enabled
		rqEnabled := hnswConfig.RQ.Enabled
		bqEnabled := hnswConfig.BQ.Enabled
		skipDefaultQuantization := hnswConfig.SkipDefaultQuantization
		hnswConfig.TrackDefaultQuantization = false
		if pqEnabled || sqEnabled || rqEnabled || bqEnabled {
			return hnswConfig
		}
		if skipDefaultQuantization {
			return hnswConfig
		}
		switch compression {
		case "pq":
			hnswConfig.PQ.Enabled = true
		case "sq":
			hnswConfig.SQ.Enabled = true
		case "rq-1":
			hnswConfig.RQ.Enabled = true
			hnswConfig.RQ.Bits = 1
			hnswConfig.RQ.RescoreLimit = hnsw.DefaultBRQRescoreLimit
		case "rq-8":
			hnswConfig.RQ.Enabled = true
			hnswConfig.RQ.Bits = 8
			hnswConfig.RQ.RescoreLimit = hnsw.DefaultRQRescoreLimit
		case "bq":
			hnswConfig.BQ.Enabled = true
		default:
			return hnswConfig
		}
		hnswConfig.TrackDefaultQuantization = true
		return hnswConfig
	}
	return vectorIndexConfig
}

func (h *Handler) RestoreClass(ctx context.Context, d *backup.ClassDescriptor, m map[string]string, overwriteAlias bool) error {
	// get schema and sharding state
	class := &models.Class{}
	if err := json.Unmarshal(d.Schema, &class); err != nil {
		return fmt.Errorf("unmarshal class schema: %w", err)
	}
	var shardingState sharding.State
	if d.ShardingState != nil {
		err := json.Unmarshal(d.ShardingState, &shardingState)
		if err != nil {
			return fmt.Errorf("unmarshal sharding state: %w", err)
		}
	}

	aliases := make([]*models.Alias, 0)
	if d.AliasesIncluded {
		if err := json.Unmarshal(d.Aliases, &aliases); err != nil {
			return fmt.Errorf("unmarshal aliases: %w", err)
		}
	}

	metric, err := monitoring.GetMetrics().BackupRestoreClassDurations.GetMetricWithLabelValues(class.Class)
	if err == nil {
		timer := prometheus.NewTimer(metric)
		defer timer.ObserveDuration()
	}

	class.Class = schema.UppercaseClassName(class.Class)
	class.Properties = schema.LowercaseAllPropertyNames(class.Properties)

	if err := h.setClassDefaults(class, h.config.Replication); err != nil {
		return err
	}

	// no validation of reference for restore
	classGetterWrapper := func(name string) (*models.Class, error) {
		return h.schemaReader.ReadOnlyClass(name), nil
	}

	err = h.validateClassInvariants(ctx, class, classGetterWrapper, true)
	if err != nil {
		return err
	}
	// migrate only after validation in completed
	h.migrateClassSettings(class)

	if err := h.parser.ParseClass(class); err != nil {
		return err
	}

	err = h.invertedConfigValidator(class.InvertedIndexConfig)
	if err != nil {
		return err
	}

	shardingState.MigrateFromOldFormat()
	err = shardingState.MigrateShardingStateReplicationFactor()
	if err != nil {
		return fmt.Errorf("error while migrating replication factor: %w", err)
	}
	shardingState.ApplyNodeMapping(m)
	_, err = h.schemaManager.RestoreClass(ctx, class, &shardingState)
	if err != nil {
		return fmt.Errorf("error when trying to restore class: %w", err)
	}

	for _, alias := range aliases {
		var err error
		_, err = h.schemaManager.CreateAlias(ctx, alias.Alias, class)
		if errors.Is(err, cschema.ErrAliasExists) {
			// Overwrite if user asks to during restore
			if overwriteAlias {
				_, err = h.schemaManager.DeleteAlias(ctx, alias.Alias)
				if err != nil {
					return fmt.Errorf("failed to restore alias for class: delete alias failed: %w", err)
				}
				// retry again
				_, err = h.schemaManager.CreateAlias(ctx, alias.Alias, class)
				if err != nil {
					return fmt.Errorf("failed to restore alias for class: create alias failed: %w", err)
				}
				return nil
			}
			// Schema returned alias already exists error. So let user know
			// that there is a "flag overwrite" if she want's to overwrite alias.
			return fmt.Errorf("failed to restore alias for class: alias already exists. You can overwrite using `overwrite_alias` param when restoring")
		}

		if err != nil {
			return fmt.Errorf("failed to restore alias for class: %w", err)
		}
	}

	return nil
}

// DeleteClass from the schema
func (h *Handler) DeleteClass(ctx context.Context, principal *models.Principal, class string) error {
	err := h.Authorizer.Authorize(ctx, principal, authorization.DELETE, authorization.CollectionsMetadata(class)...)
	if err != nil {
		return err
	}

	class = schema.UppercaseClassName(class)

	if _, err = h.schemaManager.DeleteClass(ctx, class); err != nil {
		return err
	}

	return nil
}

func (h *Handler) UpdateClass(ctx context.Context, principal *models.Principal,
	className string, updated *models.Class,
) error {
	err := h.Authorizer.Authorize(ctx, principal, authorization.UPDATE, authorization.CollectionsMetadata(className)...)
	if err != nil || updated == nil {
		return err
	}

	return UpdateClassInternal(h, ctx, className, updated)
}

// bypass the auth check for internal class update requests
func UpdateClassInternal(h *Handler, ctx context.Context, className string, updated *models.Class,
) error {
	// make sure unset optionals on 'updated' don't lead to an error, as all
	// optionals would have been set with defaults on the initial already
	if err := h.setClassDefaults(updated, h.config.Replication); err != nil {
		return err
	}

	if err := h.parser.ParseClass(updated); err != nil {
		return err
	}

	// ideally, these calls would be encapsulated in ParseClass but ParseClass is
	// used in many different areas of the codebase that may cause BC issues with the
	// new validation logic. Issue ref: gh-5860
	// As our testing becomes more comprehensive, we can move these calls into ParseClass
	if err := h.parser.parseModuleConfig(updated); err != nil {
		return fmt.Errorf("parse module config: %w", err)
	}

	if err := h.parser.parseVectorConfig(updated); err != nil {
		return fmt.Errorf("parse vector config: %w", err)
	}

	if err := h.validateVectorSettings(updated); err != nil {
		return err
	}

	if initial := h.schemaReader.ReadOnlyClass(className); initial != nil {
		if err := validateImmutableFields(initial, updated); err != nil {
			return err
		}
	}
	// A nil sharding state means that the sharding state will not be updated.
	_, err := h.schemaManager.UpdateClass(ctx, updated, nil)
	return err
}

func (m *Handler) setNewClassDefaults(class *models.Class, globalCfg replication.GlobalConfig) error {
	if class.ShardingConfig != nil && schema.MultiTenancyEnabled(class) {
		return fmt.Errorf("cannot have both shardingConfig and multiTenancyConfig")
	} else if class.MultiTenancyConfig == nil {
		class.MultiTenancyConfig = &models.MultiTenancyConfig{}
	} else if class.MultiTenancyConfig.Enabled {
		class.ShardingConfig = shardingcfg.Config{DesiredCount: 0} // tenant shards will be created dynamically
	}

	if err := m.setClassDefaults(class, globalCfg); err != nil {
		return err
	}

	if class.ReplicationConfig == nil {
		class.ReplicationConfig = &models.ReplicationConfig{
			Factor:           int64(m.config.Replication.MinimumFactor),
			DeletionStrategy: models.ReplicationConfigDeletionStrategyNoAutomatedResolution,
		}
		return nil
	}

	if class.ReplicationConfig.DeletionStrategy == "" {
		class.ReplicationConfig.DeletionStrategy = models.ReplicationConfigDeletionStrategyNoAutomatedResolution
	}
	return nil
}

func (h *Handler) setClassDefaults(class *models.Class, globalCfg replication.GlobalConfig) error {
	// set legacy vector index defaults only when:
	// 	- no target vectors are configured
	//  - OR, there are target vectors configured AND there is a legacy vector configured
	if !hasTargetVectors(class) || modelsext.ClassHasLegacyVectorIndex(class) {
		if class.Vectorizer == "" {
			class.Vectorizer = h.config.DefaultVectorizerModule
		}

		if class.VectorIndexType == "" {
			class.VectorIndexType = vectorindex.DefaultVectorIndexType
		}

		if h.config.DefaultVectorDistanceMetric != "" {
			if class.VectorIndexConfig == nil {
				class.VectorIndexConfig = map[string]interface{}{"distance": h.config.DefaultVectorDistanceMetric}
			} else if vIdxCfgMap, ok := class.VectorIndexConfig.(map[string]interface{}); ok && vIdxCfgMap["distance"] == nil {
				class.VectorIndexConfig.(map[string]interface{})["distance"] = h.config.DefaultVectorDistanceMetric
			}
		}
	}

	setInvertedConfigDefaults(class)
	for _, prop := range class.Properties {
		setPropertyDefaults(prop)
	}

	if class.ReplicationConfig == nil {
		class.ReplicationConfig = &models.ReplicationConfig{Factor: int64(globalCfg.MinimumFactor)}
	}

	if class.ReplicationConfig.Factor > 0 && class.ReplicationConfig.Factor < int64(globalCfg.MinimumFactor) {
		return fmt.Errorf("invalid replication factor: setup requires a minimum replication factor of %d: got %d",
			globalCfg.MinimumFactor, class.ReplicationConfig.Factor)
	}

	if class.ReplicationConfig.Factor < 1 {
		class.ReplicationConfig.Factor = int64(globalCfg.MinimumFactor)
	}

	h.moduleConfig.SetClassDefaults(class)
	return nil
}

func setPropertyDefaults(props ...*models.Property) {
	setPropertyDefaultTokenization(props...)
	setPropertyDefaultIndexing(props...)
	for _, prop := range props {
		setNestedPropertiesDefaults(prop.NestedProperties)
	}
}

func setPropertyDefaultTokenization(props ...*models.Property) {
	for _, prop := range props {
		switch dataType, _ := schema.AsPrimitive(prop.DataType); dataType {
		case schema.DataTypeString, schema.DataTypeStringArray:
			// deprecated as of v1.19, default tokenization was word
			// which will be migrated to text+whitespace
			if prop.Tokenization == "" {
				prop.Tokenization = models.PropertyTokenizationWord
			}
		case schema.DataTypeText, schema.DataTypeTextArray:
			if prop.Tokenization == "" {
				if os.Getenv("DEFAULT_TOKENIZATION") != "" {
					prop.Tokenization = os.Getenv("DEFAULT_TOKENIZATION")
				} else {
					prop.Tokenization = models.PropertyTokenizationWord
				}
			}
		default:
			// tokenization not supported for other data types
		}
	}
}

func setPropertyDefaultIndexing(props ...*models.Property) {
	for _, prop := range props {
		// if IndexInverted is set but IndexFilterable and IndexSearchable are not
		// migrate IndexInverted later.
		if prop.IndexInverted != nil &&
			prop.IndexFilterable == nil &&
			prop.IndexSearchable == nil &&
			prop.IndexRangeFilters == nil {
			continue
		}

		vTrue := true
		vFalse := false
		if prop.IndexFilterable == nil {
			prop.IndexFilterable = &vTrue
		}
		if prop.IndexSearchable == nil {
			switch dataType, _ := schema.AsPrimitive(prop.DataType); dataType {
			case schema.DataTypeString, schema.DataTypeStringArray:
				// string/string[] are migrated to text/text[] later,
				// at this point they are still valid data types, therefore should be handled here
				prop.IndexSearchable = &vTrue
			case schema.DataTypeText, schema.DataTypeTextArray:
				prop.IndexSearchable = &vTrue
			default:
				prop.IndexSearchable = &vFalse
			}
		}
		if prop.IndexRangeFilters == nil {
			prop.IndexRangeFilters = &vFalse
		}
	}
}

func setNestedPropertiesDefaults(properties []*models.NestedProperty) {
	for _, property := range properties {
		primitiveDataType, isPrimitive := schema.AsPrimitive(property.DataType)
		nestedDataType, isNested := schema.AsNested(property.DataType)

		setNestedPropertyDefaultTokenization(property, primitiveDataType, nestedDataType, isPrimitive, isNested)
		setNestedPropertyDefaultIndexing(property, primitiveDataType, nestedDataType, isPrimitive, isNested)

		if isNested {
			setNestedPropertiesDefaults(property.NestedProperties)
		}
	}
}

func setNestedPropertyDefaultTokenization(property *models.NestedProperty,
	primitiveDataType, nestedDataType schema.DataType,
	isPrimitive, isNested bool,
) {
	if property.Tokenization == "" && isPrimitive {
		switch primitiveDataType {
		case schema.DataTypeText, schema.DataTypeTextArray:
			property.Tokenization = models.NestedPropertyTokenizationWord
		default:
			// do nothing
		}
	}
}

func setNestedPropertyDefaultIndexing(property *models.NestedProperty,
	primitiveDataType, nestedDataType schema.DataType,
	isPrimitive, isNested bool,
) {
	vTrue := true
	vFalse := false

	if property.IndexFilterable == nil {
		property.IndexFilterable = &vTrue

		if isPrimitive && primitiveDataType == schema.DataTypeBlob {
			property.IndexFilterable = &vFalse
		}
	}

	if property.IndexSearchable == nil {
		property.IndexSearchable = &vFalse

		if isPrimitive {
			switch primitiveDataType {
			case schema.DataTypeText, schema.DataTypeTextArray:
				property.IndexSearchable = &vTrue
			default:
				// do nothing
			}
		}
	}

	if property.IndexRangeFilters == nil {
		property.IndexRangeFilters = &vFalse
	}
}

func (h *Handler) migrateClassSettings(class *models.Class) {
	for _, prop := range class.Properties {
		migratePropertySettings(prop)
	}
}

func migratePropertySettings(props ...*models.Property) {
	migratePropertyDataTypeAndTokenization(props...)
	migratePropertyIndexInverted(props...)
}

// as of v1.19 DataTypeString and DataTypeStringArray are deprecated
// here both are changed to Text/TextArray
// and proper, backward compatible tokenization
func migratePropertyDataTypeAndTokenization(props ...*models.Property) {
	for _, prop := range props {
		switch dataType, _ := schema.AsPrimitive(prop.DataType); dataType {
		case schema.DataTypeString:
			prop.DataType = schema.DataTypeText.PropString()
		case schema.DataTypeStringArray:
			prop.DataType = schema.DataTypeTextArray.PropString()
		default:
			// other types need no migration and do not support tokenization
			continue
		}

		switch prop.Tokenization {
		case models.PropertyTokenizationWord:
			prop.Tokenization = models.PropertyTokenizationWhitespace
		case models.PropertyTokenizationField:
			// stays field
		}
	}
}

// as of v1.19 IndexInverted is deprecated and replaced with
// IndexFilterable (set inverted index)
// and IndexSearchable (map inverted index with term frequencies;
// therefore applicable only to text/text[] data types)
func migratePropertyIndexInverted(props ...*models.Property) {
	vFalse := false

	for _, prop := range props {
		// if none of new options is set, use inverted settings
		if prop.IndexInverted != nil &&
			prop.IndexFilterable == nil &&
			prop.IndexSearchable == nil &&
			prop.IndexRangeFilters == nil {
			prop.IndexFilterable = prop.IndexInverted
			switch dataType, _ := schema.AsPrimitive(prop.DataType); dataType {
			// string/string[] are already migrated into text/text[], can be skipped here
			case schema.DataTypeText, schema.DataTypeTextArray:
				prop.IndexSearchable = prop.IndexInverted
			default:
				prop.IndexSearchable = &vFalse
			}
			prop.IndexRangeFilters = &vFalse
		}
		// new options have precedence so inverted can be reset
		prop.IndexInverted = nil
	}
}

func (h *Handler) validateProperty(
	class *models.Class, existingPropertyNames map[string]bool,
	relaxCrossRefValidation bool, classGetterWithAuth func(string) (*models.Class, error), props ...*models.Property,
) error {
	for _, property := range props {
		if _, err := schema.ValidatePropertyName(property.Name); err != nil {
			return err
		}

		if err := schema.ValidateReservedPropertyName(property.Name); err != nil {
			return err
		}

		if existingPropertyNames[strings.ToLower(property.Name)] {
			return fmt.Errorf("class %q: conflict for property %q: already in use or provided multiple times", class.Class, property.Name)
		}

		// Validate data type of property.
		propertyDataType, err := schema.FindPropertyDataTypeWithRefsAndAuth(classGetterWithAuth, property.DataType,
			relaxCrossRefValidation, schema.ClassName(class.Class))
		if err != nil {
			return fmt.Errorf("property '%s': invalid dataType: %v: %w", property.Name, property.DataType, err)
		}

		if propertyDataType.IsNested() {
			if err := validateNestedProperties(property.NestedProperties, property.Name); err != nil {
				return err
			}
		} else {
			if len(property.NestedProperties) > 0 {
				return fmt.Errorf("property '%s': nestedProperties not allowed for data types other than object/object[]",
					property.Name)
			}
		}

		if err := h.validatePropertyTokenization(property.Tokenization, propertyDataType); err != nil {
			return err
		}

		if err := h.validatePropertyIndexing(property); err != nil {
			return err
		}

		if err := h.validatePropModuleConfig(class, property); err != nil {
			return err
		}
	}

	return nil
}

func setInvertedConfigDefaults(class *models.Class) {
	if class.InvertedIndexConfig == nil {
		class.InvertedIndexConfig = &models.InvertedIndexConfig{
			UsingBlockMaxWAND: config.DefaultUsingBlockMaxWAND,
		}
	}

	if class.InvertedIndexConfig.CleanupIntervalSeconds == 0 {
		class.InvertedIndexConfig.CleanupIntervalSeconds = config.DefaultCleanupIntervalSeconds
	}

	if class.InvertedIndexConfig.Bm25 == nil {
		class.InvertedIndexConfig.Bm25 = &models.BM25Config{
			K1: config.DefaultBM25k1,
			B:  config.DefaultBM25b,
		}
	}

	if class.InvertedIndexConfig.Stopwords == nil {
		class.InvertedIndexConfig.Stopwords = &models.StopwordConfig{
			Preset: stopwords.EnglishPreset,
		}
	}
}

func (h *Handler) validateCanAddClass(ctx context.Context, class *models.Class, classGetterWithAuth func(string) (*models.Class, error),
	relaxCrossRefValidation bool,
) error {
	if modelsext.ClassHasLegacyVectorIndex(class) && len(class.VectorConfig) > 0 {
		return fmt.Errorf("creating a class with both a class level vector index and named vectors is forbidden")
	}

	return h.validateClassInvariants(ctx, class, classGetterWithAuth, relaxCrossRefValidation)
}

func (h *Handler) validateClassInvariants(
	ctx context.Context, class *models.Class, classGetterWithAuth func(string) (*models.Class, error),
	relaxCrossRefValidation bool,
) error {
	if _, err := schema.ValidateClassName(class.Class); err != nil {
		return err
	}

	existingPropertyNames := map[string]bool{}
	for _, property := range class.Properties {
		if err := h.validateProperty(class, existingPropertyNames, relaxCrossRefValidation, classGetterWithAuth, property); err != nil {
			return err
		}
		existingPropertyNames[strings.ToLower(property.Name)] = true
	}

	if err := h.validateVectorSettings(class); err != nil {
		return err
	}

	if err := h.moduleConfig.ValidateClass(ctx, class); err != nil {
		return err
	}

	if err := validateMT(class); err != nil {
		return err
	}

	if err := replica.ValidateConfig(class, h.config.Replication); err != nil {
		return err
	}

	// all is fine!
	return nil
}

func (h *Handler) validatePropertyTokenization(tokenization string, propertyDataType schema.PropertyDataType) error {
	if propertyDataType.IsPrimitive() {
		primitiveDataType := propertyDataType.AsPrimitive()

		switch primitiveDataType {
		case schema.DataTypeString, schema.DataTypeStringArray:
			// deprecated as of v1.19, will be migrated to DataTypeText/DataTypeTextArray
			switch tokenization {
			case models.PropertyTokenizationField, models.PropertyTokenizationWord:
				return nil
			}
		case schema.DataTypeText, schema.DataTypeTextArray:
			switch tokenization {
			case models.PropertyTokenizationField, models.PropertyTokenizationWord,
				models.PropertyTokenizationWhitespace, models.PropertyTokenizationLowercase,
				models.PropertyTokenizationTrigram:
				return nil
			case models.PropertyTokenizationGse:
				if !entcfg.Enabled(os.Getenv("USE_GSE")) && !entcfg.Enabled(os.Getenv("ENABLE_TOKENIZER_GSE")) {
					return fmt.Errorf("the GSE tokenizer is not enabled; set 'ENABLE_TOKENIZER_GSE' to 'true' to enable")
				}
				return nil
			case models.PropertyTokenizationKagomeKr:
				if !entcfg.Enabled(os.Getenv("ENABLE_TOKENIZER_KAGOME_KR")) {
					return fmt.Errorf("the Korean tokenizer is not enabled; set 'ENABLE_TOKENIZER_KAGOME_KR' to 'true' to enable")
				}
				return nil
			case models.PropertyTokenizationKagomeJa:
				if !entcfg.Enabled(os.Getenv("ENABLE_TOKENIZER_KAGOME_JA")) {
					return fmt.Errorf("the Japanese tokenizer is not enabled; set 'ENABLE_TOKENIZER_KAGOME_JA' to 'true' to enable")
				}
				return nil
			}
		default:
			if tokenization == "" {
				return nil
			}
			return fmt.Errorf("tokenization is not allowed for data type '%s'", primitiveDataType)
		}
		return fmt.Errorf("tokenization '%s' is not allowed for data type '%s'", tokenization, primitiveDataType)
	}

	if tokenization == "" {
		return nil
	}

	if propertyDataType.IsNested() {
		return fmt.Errorf("tokenization is not allowed for object/object[] data types")
	}
	return fmt.Errorf("tokenization is not allowed for reference data type")
}

func (h *Handler) validatePropertyIndexing(prop *models.Property) error {
	if prop.IndexInverted != nil {
		if prop.IndexFilterable != nil || prop.IndexSearchable != nil || prop.IndexRangeFilters != nil {
			return fmt.Errorf("`indexInverted` is deprecated and can not be set together with `indexFilterable`, " + "`indexSearchable` or `indexRangeFilters`")
		}
	}

	dataType, _ := schema.AsPrimitive(prop.DataType)
	if prop.IndexSearchable != nil {
		switch dataType {
		case schema.DataTypeString, schema.DataTypeStringArray:
			// string/string[] are migrated to text/text[] later,
			// at this point they are still valid data types, therefore should be handled here.
			// true or false allowed
		case schema.DataTypeText, schema.DataTypeTextArray:
			// true or false allowed
		default:
			if *prop.IndexSearchable {
				return fmt.Errorf("`indexSearchable` is allowed only for text/text[] data types. " +
					"For other data types set false or leave empty")
			}
		}
	}
	if prop.IndexRangeFilters != nil {
		switch dataType {
		case schema.DataTypeNumber, schema.DataTypeInt, schema.DataTypeDate:
			// true or false allowed
		case schema.DataTypeNumberArray, schema.DataTypeIntArray, schema.DataTypeDateArray:
			// not supported (yet?)
			fallthrough
		default:
			if *prop.IndexRangeFilters {
				return fmt.Errorf("`indexRangeFilters` is allowed only for number/int/date data types. " +
					"For other data types set false or leave empty")
			}
		}
	}

	return nil
}

func (h *Handler) validateVectorSettings(class *models.Class) error {
	if modelsext.ClassHasLegacyVectorIndex(class) {
		if err := h.validateVectorIndexType(class.VectorIndexType); err != nil {
			return err
		}

		if err := h.validateVectorizer(class.Vectorizer); err != nil {
			return err
		}

		if asMap, ok := class.VectorIndexConfig.(map[string]interface{}); ok && len(asMap) > 0 {
			parsed, err := h.parser.parseGivenVectorIndexConfig(class.VectorIndexType, class.VectorIndexConfig, h.parser.modules.IsMultiVector(class.Vectorizer), h.config.DefaultQuantization)
			if err != nil {
				return fmt.Errorf("class.VectorIndexConfig can not parse: %w", err)
			}
			if parsed.IsMultiVector() {
				return errors.New("class.VectorIndexConfig multi vector type index type is only configurable using named vectors")
			}
		}
	}

	for name, cfg := range class.VectorConfig {
		// check only if vectorizer correctly configured (map with single key being vectorizer name)
		// other cases are handled in module config validation
		if vm, ok := cfg.Vectorizer.(map[string]interface{}); ok && len(vm) == 1 {
			for vectorizer := range vm {
				if err := h.validateVectorizer(vectorizer); err != nil {
					return fmt.Errorf("target vector %q: %w", name, err)
				}
			}
		}
		if err := h.validateVectorIndexType(cfg.VectorIndexType); err != nil {
			return fmt.Errorf("target vector %q: %w", name, err)
		}
	}
	return nil
}

func (h *Handler) validateVectorizer(vectorizer string) error {
	if vectorizer == config.VectorizerModuleNone {
		return nil
	}

	if err := h.vectorizerValidator.ValidateVectorizer(vectorizer); err != nil {
		return errors.Wrap(err, "vectorizer")
	}

	return nil
}

func (h *Handler) validateVectorIndexType(vectorIndexType string) error {
	switch vectorIndexType {
	case vectorindex.VectorIndexTypeHNSW, vectorindex.VectorIndexTypeFLAT:
		return nil
	case vectorindex.VectorIndexTypeDYNAMIC:
		if !h.asyncIndexingEnabled {
			return fmt.Errorf("the dynamic index can only be created under async indexing environment (ASYNC_INDEXING=true)")
		}
		return nil
	default:
		return errors.Errorf("unrecognized or unsupported vectorIndexType %q",
			vectorIndexType)
	}
}

func validateMT(class *models.Class) error {
	enabled := schema.MultiTenancyEnabled(class)
	if !enabled && schema.AutoTenantCreationEnabled(class) {
		return fmt.Errorf("can't enable autoTenantCreation on a non-multi-tenant class")
	}

	if !enabled && schema.AutoTenantActivationEnabled(class) {
		return fmt.Errorf("can't enable autoTenantActivation on a non-multi-tenant class")
	}

	return nil
}

// validateUpdatingMT validates toggling MT and returns whether mt is enabled
func validateUpdatingMT(current, update *models.Class) (enabled bool, err error) {
	enabled = schema.MultiTenancyEnabled(current)
	if schema.MultiTenancyEnabled(update) != enabled {
		if enabled {
			err = fmt.Errorf("disabling multi-tenancy for an existing class is not supported")
		} else {
			err = fmt.Errorf("enabling multi-tenancy for an existing class is not supported")
		}
	} else {
		err = validateMT(update)
	}

	return
}

func validateImmutableFields(initial, updated *models.Class) error {
	immutableFields := []immutableText{
		{
			name:     "class name",
			accessor: func(c *models.Class) string { return c.Class },
		},
	}

	if err := validateImmutableTextFields(initial, updated, immutableFields...); err != nil {
		return err
	}

	for k, v := range updated.VectorConfig {
		if _, ok := initial.VectorConfig[k]; !ok {
			continue
		}

		if !reflect.DeepEqual(initial.VectorConfig[k].Vectorizer, v.Vectorizer) {
			return fmt.Errorf("vectorizer config of vector %q is immutable", k)
		}
	}

	return nil
}

type immutableText struct {
	accessor func(c *models.Class) string
	name     string
}

func validateImmutableTextFields(previous, next *models.Class,
	immutables ...immutableText,
) error {
	for _, immutable := range immutables {
		oldField := immutable.accessor(previous)
		newField := immutable.accessor(next)
		if oldField != newField {
			return errors.Errorf("%s is immutable: attempted change from %q to %q",
				immutable.name, oldField, newField)
		}
	}
	return nil
}

func validateLegacyVectorIndexConfigImmutableFields(initial, updated *models.Class) error {
	return validateImmutableTextFields(initial, updated, []immutableText{
		{
			name:     "vectorizer",
			accessor: func(c *models.Class) string { return c.Vectorizer },
		},
		{
			name:     "vector index type",
			accessor: func(c *models.Class) string { return c.VectorIndexType },
		},
	}...)
}