@strapi/core 0.0.0-experimental.ca5f02ef2ec166e219546983049c0543bed92f8d → 0.0.0-experimental.cad2fba5593c9ceddfc264fcecd9313a7f866d1a

package/dist/migrations/database/5.0.0-discard-drafts.js

@@ -1,8 +1,46 @@
 'use strict';
 
+var cuid2 = require('@paralleldrive/cuid2');
 var strapiUtils = require('@strapi/utils');
-var index = require('../../services/document-service/index.js');
+var createDebug = require('debug');
+var transformContentTypesToModels = require('../../utils/transform-content-types-to-models.js');
 
+const DEFAULT_PRIMARY_KEY_COLUMN = 'id';
+/**
+ * Determines the primary-key column name for a schema, handling the various shapes
+ * metadata can take (string, object, attribute flag) and defaulting to `id`.
+ */ const resolvePrimaryKeyColumn = (meta)=>{
+    const { primaryKey } = meta;
+    if (typeof primaryKey === 'string' && primaryKey) {
+        return primaryKey;
+    }
+    if (primaryKey && typeof primaryKey === 'object') {
+        const pkObject = primaryKey;
+        const columnName = pkObject.columnName;
+        if (typeof columnName === 'string' && columnName) {
+            return columnName;
+        }
+        const name = pkObject.name;
+        if (typeof name === 'string' && name) {
+            return name;
+        }
+    }
+    const attributes = meta.attributes ?? {};
+    for (const [attributeName, attribute] of Object.entries(attributes)){
+        const normalizedAttribute = attribute;
+        if (!normalizedAttribute) {
+            continue;
+        }
+        const { column } = normalizedAttribute;
+        if (column?.primary === true) {
+            if (typeof normalizedAttribute.columnName === 'string' && normalizedAttribute.columnName) {
+                return normalizedAttribute.columnName;
+            }
+            return attributeName;
+        }
+    }
+    return DEFAULT_PRIMARY_KEY_COLUMN;
+};
 /**
  * Check if the model has draft and publish enabled.
  */ const hasDraftAndPublish = async (trx, meta)=>{
@@ -49,7 +87,9 @@ var index = require('../../services/document-service/index.js');
     ])).insert((subQb)=>{
         // SELECT columnName1, columnName2, columnName3, ...
         subQb.select(...scalarAttributes.map((att)=>{
-            // Override 'publishedAt' and 'updatedAt' attributes
+            // NOTE: these literals reference Strapi's built-in system columns. They never get shortened by
+            // the identifier migration (5.0.0-01-convert-identifiers-long-than-max-length) so we can safely
+            // compare/use them directly here.
             if (att === 'published_at') {
                 return trx.raw('NULL as ??', 'published_at');
             }
@@ -59,46 +99,1386 @@ var index = require('../../services/document-service/index.js');
     });
 }
 /**
- * Load a batch of versions to discard.
- *
- * Versions with only a draft version will be ignored.
- * Only versions with a published version (which always have a draft version) will be discarded.
- */ async function* getBatchToDiscard({ db, trx, uid, defaultBatchSize = 1000 }) {
-    const client = db.config.connection.client;
+ * Orchestrates the relation cloning pipeline for a single content type. We duplicate
+ * every category of relation (self, inbound, outbound, components) using direct SQL so
+ * the migration can scale without calling `discardDraft` entry by entry.
+ */ async function copyRelationsToDrafts({ db, trx, uid }) {
+    const meta = db.metadata.get(uid);
+    if (!meta) {
+        return;
+    }
+    // Create mapping from published entry ID to draft entry ID
+    const publishedToDraftMap = await buildPublishedToDraftMap({
+        trx,
+        uid,
+        meta
+    });
+    if (!publishedToDraftMap || publishedToDraftMap.size === 0) {
+        return;
+    }
+    // Copy relations for this content type
+    await copyRelationsForContentType({
+        trx,
+        uid,
+        publishedToDraftMap
+    });
+    // Copy relations from other content types that target this content type
+    await copyRelationsFromOtherContentTypes({
+        trx,
+        uid,
+        publishedToDraftMap
+    });
+    // Copy relations from this content type that target other content types (category 3)
+    await copyRelationsToOtherContentTypes({
+        trx,
+        uid,
+        publishedToDraftMap
+    });
+    // Copy component relations from published entries to draft entries
+    await copyComponentRelations({
+        trx,
+        uid,
+        publishedToDraftMap
+    });
+}
+/**
+ * Splits large input arrays into smaller batches so we can run SQL queries without
+ * hitting parameter limits (SQLite) or payload limits (MySQL/Postgres).
+ */ function chunkArray(array, chunkSize) {
+    const chunks = [];
+    for(let i = 0; i < array.length; i += chunkSize){
+        chunks.push(array.slice(i, i + chunkSize));
+    }
+    return chunks;
+}
+/**
+ * Chooses a safe batch size for bulk operations depending on the database engine,
+ * falling back to smaller units on engines (notably SQLite) that have low limits.
+ */ function getBatchSize(trx, defaultSize = 1000) {
+    const client = trx.client.config.client;
     const isSQLite = typeof client === 'string' && [
         'sqlite',
         'sqlite3',
         'better-sqlite3'
     ].includes(client);
-    // The SQLite documentation states that the maximum number of terms in a
+    // SQLite documentation states that the maximum number of terms in a
     // compound SELECT statement is 500 by default.
     // See: https://www.sqlite.org/limits.html
-    // To ensure a successful migration, we limit the batch size to 500 for SQLite.
-    const batchSize = isSQLite ? Math.min(defaultBatchSize, 500) : defaultBatchSize;
-    let offset = 0;
-    let hasMore = true;
-    while(hasMore){
-        // Look for the published entries to discard
-        const batch = await db.queryBuilder(uid).select([
+    // We use 250 to be safe and account for other query complexity.
+    return isSQLite ? Math.min(defaultSize, 250) : defaultSize;
+}
+/**
+ * Applies stable ordering to join-table queries so cloning work is deterministic and
+ * matches the ordering logic used by the entity service (important for tests and DZ order).
+ */ const applyJoinTableOrdering = (qb, joinTable, sourceColumnName)=>{
+    const seenColumns = new Set();
+    const enqueueColumn = (column, direction = 'asc')=>{
+        if (!column || seenColumns.has(column)) {
+            return;
+        }
+        seenColumns.add(column);
+        qb.orderBy(column, direction);
+    };
+    enqueueColumn(sourceColumnName, 'asc');
+    if (Array.isArray(joinTable?.orderBy)) {
+        for (const clause of joinTable.orderBy){
+            if (!clause || typeof clause !== 'object') {
+                continue;
+            }
+            const [column, direction] = Object.entries(clause)[0] ?? [];
+            if (!column) {
+                continue;
+            }
+            const normalizedDirection = typeof direction === 'string' && direction.toLowerCase() === 'desc' ? 'desc' : 'asc';
+            enqueueColumn(column, normalizedDirection);
+        }
+    }
+    enqueueColumn(joinTable?.orderColumnName, 'asc');
+    enqueueColumn(joinTable?.orderColumn, 'asc');
+    enqueueColumn('id', 'asc');
+};
+const componentParentSchemasCache = new Map();
+const joinTableExistsCache = new Map();
+const componentMetaCache = new Map();
+const SKIPPED_RELATION_SAMPLE_LIMIT = 5;
+const supportsReturning = (trx)=>{
+    const client = trx.client.config.client;
+    if (typeof client !== 'string') {
+        return false;
+    }
+    return [
+        'postgres',
+        'pg'
+    ].includes(client.toLowerCase());
+};
+const skippedRelationStats = new Map();
+const DUPLICATE_ERROR_CODES = new Set([
+    '23505',
+    'ER_DUP_ENTRY',
+    'SQLITE_CONSTRAINT_UNIQUE'
+]);
+const debug = createDebug('strapi::migration::discard-drafts');
+/**
+ * Converts arbitrary id values into numbers when possible so we can safely index into
+ * mapping tables without worrying about string/number mismatches.
+ */ const normalizeId = (value)=>{
+    if (value == null) {
+        return null;
+    }
+    const num = Number(value);
+    if (Number.isNaN(num)) {
+        return null;
+    }
+    return num;
+};
+/**
+ * Wrapper around map lookups that first normalizes the provided identifier.
+ */ const getMappedValue = (map, key)=>{
+    if (!map) {
+        return undefined;
+    }
+    const normalized = normalizeId(key);
+    if (normalized == null) {
+        return undefined;
+    }
+    return map.get(normalized);
+};
+/**
+ * Extracts the inserted row identifier across the various shapes returned by different
+ * clients/drivers (numbers, objects, arrays). Falls back to `null` if nothing usable.
+ */ const resolveInsertedId = (insertResult)=>{
+    if (insertResult == null) {
+        return null;
+    }
+    if (typeof insertResult === 'number') {
+        return insertResult;
+    }
+    if (Array.isArray(insertResult)) {
+        if (insertResult.length === 0) {
+            return null;
+        }
+        const first = insertResult[0];
+        if (first == null) {
+            return null;
+        }
+        if (typeof first === 'number') {
+            return first;
+        }
+        if (typeof first === 'object') {
+            if ('id' in first) {
+                return Number(first.id);
+            }
+            const idKey = Object.keys(first).find((key)=>key.toLowerCase() === 'id');
+            if (idKey) {
+                return Number(first[idKey]);
+            }
+        }
+    }
+    if (typeof insertResult === 'object' && 'id' in insertResult) {
+        return Number(insertResult.id);
+    }
+    return null;
+};
+/**
+ * Detects vendor-specific duplicate key errors so we can safely ignore them when our
+ * goal is to insert-or-ignore without branching on the client everywhere.
+ */ const isDuplicateEntryError = (error)=>{
+    if (!error) {
+        return false;
+    }
+    if (DUPLICATE_ERROR_CODES.has(error.code)) {
+        return true;
+    }
+    const message = typeof error.message === 'string' ? error.message : '';
+    return message.includes('duplicate key') || message.includes('UNIQUE constraint failed');
+};
+/**
+ * Inserts a row while tolerating duplicates across all supported clients. Used when
+ * bulk operations fall back to single inserts to resolve constraint conflicts.
+ */ const insertRowWithDuplicateHandling = async (trx, tableName, row, context = {})=>{
+    try {
+        const client = trx.client.config.client;
+        if (client === 'postgres' || client === 'pg' || client === 'sqlite3' || client === 'better-sqlite3') {
+            await trx(tableName).insert(row).onConflict().ignore();
+            return;
+        }
+        if (client === 'mysql' || client === 'mysql2') {
+            await trx.raw(`INSERT IGNORE INTO ?? SET ?`, [
+                tableName,
+                row
+            ]);
+            return;
+        }
+        await trx(tableName).insert(row);
+    } catch (error) {
+        if (!isDuplicateEntryError(error)) {
+            const details = JSON.stringify(context);
+            const wrapped = new Error(`Failed to insert row into ${tableName}: ${error.message} | context=${details}`);
+            wrapped.cause = error;
+            throw wrapped;
+        }
+    }
+};
+/**
+ * Normalizes identifiers into a comparable string so we can dedupe target ids
+ * regardless of whether they come in as numbers, strings, or objects.
+ */ const toComparisonKey = (value)=>{
+    if (value === null || value === undefined) {
+        return 'null';
+    }
+    if (typeof value === 'object') {
+        try {
+            return JSON.stringify(value);
+        } catch  {
+            return String(value);
+        }
+    }
+    return String(value);
+};
+/**
+ * Formats ids for log output while keeping the log lightweight and JSON-safe.
+ */ const toDisplayValue = (value)=>{
+    if (value === null || value === undefined) {
+        return 'null';
+    }
+    if (typeof value === 'string' || typeof value === 'number' || typeof value === 'bigint') {
+        return String(value);
+    }
+    try {
+        return JSON.stringify(value);
+    } catch  {
+        return String(value);
+    }
+};
+/**
+ * Tracks how many relation rows we skipped because their targets are missing, so we
+ * can emit a consolidated warning once the migration finishes copying relations.
+ */ function recordSkippedRelations(context, skippedIds) {
+    if (!skippedIds.length) {
+        return;
+    }
+    const key = `${context.sourceUid}::${context.attributeName}::${context.joinTable ?? 'NO_JOIN_TABLE'}`;
+    let stats = skippedRelationStats.get(key);
+    if (!stats) {
+        stats = {
+            ...context,
+            count: 0,
+            samples: new Set()
+        };
+    }
+    stats.count += skippedIds.length;
+    for (const id of skippedIds){
+        if (stats.samples.size >= SKIPPED_RELATION_SAMPLE_LIMIT) {
+            break;
+        }
+        stats.samples.add(toDisplayValue(id));
+    }
+    skippedRelationStats.set(key, stats);
+}
+/**
+ * Emits aggregated warnings for all skipped relations and resets the counters.
+ * This keeps the log readable even when millions of orphaned rows exist.
+ */ function flushSkippedRelationLogs() {
+    if (skippedRelationStats.size === 0) {
+        return;
+    }
+    for (const stats of skippedRelationStats.values()){
+        const sampleArray = Array.from(stats.samples);
+        const sampleText = sampleArray.length > 0 ? sampleArray.join(', ') : 'n/a';
+        const targetInfo = stats.targetUid ? `target=${stats.targetUid}` : 'target=unknown';
+        const joinTableInfo = stats.joinTable ? `joinTable=${stats.joinTable}` : 'joinTable=n/a';
+        const ellipsis = stats.count > sampleArray.length ? ', ...' : '';
+        strapi.log.warn(`[discard-drafts] Skipped ${stats.count} relation(s) for ${stats.sourceUid}.${stats.attributeName} (${targetInfo}, ${joinTableInfo}). Example target ids: ${sampleText}${ellipsis}`);
+    }
+    strapi.log.warn('[discard-drafts] Some join-table relations referenced missing targets and were skipped. Review these warnings and clean up orphaned relations before rerunning the migration if needed.');
+    skippedRelationStats.clear();
+}
+/**
+ * Returns every schema (content type or component) that can embed the provided component.
+ * Cached because we consult it repeatedly while remapping nested components.
+ */ function listComponentParentSchemas(componentUid) {
+    if (!componentParentSchemasCache.has(componentUid)) {
+        const schemas = [
+            ...Object.values(strapi.contentTypes),
+            ...Object.values(strapi.components)
+        ];
+        const parents = schemas.filter((schema)=>{
+            if (!schema?.attributes) {
+                return false;
+            }
+            return Object.values(schema.attributes).some((attr)=>{
+                if (attr.type === 'component') {
+                    return attr.component === componentUid;
+                }
+                if (attr.type === 'dynamiczone') {
+                    return attr.components?.includes(componentUid);
+                }
+                return false;
+            });
+        }).map((schema)=>({
+                uid: schema.uid,
+                collectionName: schema.collectionName
+            }));
+        componentParentSchemasCache.set(componentUid, parents);
+    }
+    return componentParentSchemasCache.get(componentUid);
+}
+/**
+ * Memoized helper for checking whether a table exists. Avoids repeating expensive
+ * information_schema queries when we touch the same join table many times.
+ */ async function ensureTableExists(trx, tableName) {
+    if (!joinTableExistsCache.has(tableName)) {
+        const exists = await trx.schema.hasTable(tableName);
+        joinTableExistsCache.set(tableName, exists);
+    }
+    return joinTableExistsCache.get(tableName);
+}
+/**
+ * Filters out relation rows whose target entity no longer exists, returning the safe
+ * rows along with a list of skipped ids so we can log them later.
+ */ async function filterRelationsWithExistingTargets({ trx, targetUid, relations, getTargetId }) {
+    if (!relations.length) {
+        return {
+            relations,
+            skippedIds: []
+        };
+    }
+    if (!targetUid) {
+        return {
+            relations,
+            skippedIds: []
+        };
+    }
+    const targetMeta = strapi.db.metadata.get(targetUid);
+    if (!targetMeta) {
+        return {
+            relations,
+            skippedIds: []
+        };
+    }
+    const tableName = targetMeta.tableName;
+    const primaryKeyColumn = resolvePrimaryKeyColumn(targetMeta);
+    if (!tableName) {
+        return {
+            relations,
+            skippedIds: []
+        };
+    }
+    const uniqueIdMap = new Map();
+    for (const relation of relations){
+        const targetId = getTargetId(relation);
+        if (targetId == null) {
+            continue;
+        }
+        const key = toComparisonKey(targetId);
+        if (!uniqueIdMap.has(key)) {
+            uniqueIdMap.set(key, targetId);
+        }
+    }
+    if (uniqueIdMap.size === 0) {
+        return {
+            relations,
+            skippedIds: []
+        };
+    }
+    const hasTable = await ensureTableExists(trx, tableName);
+    if (!hasTable) {
+        return {
+            relations: [],
+            skippedIds: Array.from(uniqueIdMap.values())
+        };
+    }
+    const existingKeys = new Set();
+    const uniqueIds = Array.from(uniqueIdMap.values());
+    const idChunks = chunkArray(uniqueIds, getBatchSize(trx, 1000));
+    for (const chunk of idChunks){
+        const rows = await trx(tableName).select(primaryKeyColumn).whereIn(primaryKeyColumn, chunk);
+        for (const row of rows){
+            const value = row[primaryKeyColumn];
+            existingKeys.add(toComparisonKey(value));
+        }
+    }
+    const filteredRelations = [];
+    const skippedIds = [];
+    for (const relation of relations){
+        const targetId = getTargetId(relation);
+        if (targetId == null) {
+            skippedIds.push(targetId);
+            continue;
+        }
+        if (existingKeys.has(toComparisonKey(targetId))) {
+            filteredRelations.push(relation);
+        } else {
+            skippedIds.push(targetId);
+        }
+    }
+    return {
+        relations: filteredRelations,
+        skippedIds
+    };
+}
+/**
+ * Locates the owning entity (content type or component) for a given component instance.
+ * This mirrors the document service logic we need in order to decide whether a relation
+ * should propagate to drafts, and is cached for performance.
+ */ async function findComponentParentInstance(trx, identifiers, componentUid, componentId, excludeUid, caches) {
+    const cacheKey = `${componentUid}:${componentId}:${excludeUid ?? 'ALL'}`;
+    if (caches.parentInstanceCache.has(cacheKey)) {
+        return caches.parentInstanceCache.get(cacheKey);
+    }
+    const parentComponentIdColumn = transformContentTypesToModels.getComponentJoinColumnInverseName(identifiers);
+    const parentComponentTypeColumn = transformContentTypesToModels.getComponentTypeColumn(identifiers);
+    const parentEntityIdColumn = transformContentTypesToModels.getComponentJoinColumnEntityName(identifiers);
+    const potentialParents = listComponentParentSchemas(componentUid).filter((schema)=>schema.uid !== excludeUid);
+    for (const parentSchema of potentialParents){
+        if (!parentSchema.collectionName) {
+            continue;
+        }
+        const parentJoinTableName = transformContentTypesToModels.getComponentJoinTableName(parentSchema.collectionName, identifiers);
+        try {
+            if (!await ensureTableExists(trx, parentJoinTableName)) {
+                continue;
+            }
+            const parentRow = await trx(parentJoinTableName).where({
+                [parentComponentIdColumn]: componentId,
+                [parentComponentTypeColumn]: componentUid
+            }).first(parentEntityIdColumn);
+            if (parentRow) {
+                const parentInstance = {
+                    uid: parentSchema.uid,
+                    parentId: parentRow[parentEntityIdColumn]
+                };
+                caches.parentInstanceCache.set(cacheKey, parentInstance);
+                return parentInstance;
+            }
+        } catch  {
+            continue;
+        }
+    }
+    caches.parentInstanceCache.set(cacheKey, null);
+    return null;
+}
+/**
+ * Fetches and caches database metadata for a component uid. Saves repeated lookups while
+ * cloning the same component type thousands of times.
+ */ const getComponentMeta = (componentUid)=>{
+    if (!componentMetaCache.has(componentUid)) {
+        const meta = strapi.db.metadata.get(componentUid);
+        componentMetaCache.set(componentUid, meta ?? null);
+    }
+    return componentMetaCache.get(componentUid);
+};
+/**
+ * Determines whether a component's parent entity participates in draft/publish,
+ * short-circuiting a lot of recursion when deciding relation propagation rules.
+ */ async function hasDraftPublishAncestorForParent(trx, identifiers, parent, caches) {
+    const cacheKey = `${parent.uid}:${parent.parentId}`;
+    if (caches.parentDpCache.has(cacheKey)) {
+        return caches.parentDpCache.get(cacheKey);
+    }
+    const parentContentType = strapi.contentTypes[parent.uid];
+    if (parentContentType) {
+        const result = !!parentContentType?.options?.draftAndPublish;
+        caches.parentDpCache.set(cacheKey, result);
+        return result;
+    }
+    const parentComponent = strapi.components[parent.uid];
+    if (!parentComponent) {
+        caches.parentDpCache.set(cacheKey, false);
+        return false;
+    }
+    const result = await hasDraftPublishAncestorForComponent(trx, identifiers, parent.uid, parent.parentId, undefined, caches);
+    caches.parentDpCache.set(cacheKey, result);
+    return result;
+}
+/**
+ * Recursively checks whether a component lies beneath a draft/publish-enabled parent
+ * component or content type. We mirror discardDraft's propagation guard.
+ */ async function hasDraftPublishAncestorForComponent(trx, identifiers, componentUid, componentId, excludeUid, caches) {
+    const cacheKey = `${componentUid}:${componentId}:${'ALL'}`;
+    if (caches.ancestorDpCache.has(cacheKey)) {
+        return caches.ancestorDpCache.get(cacheKey);
+    }
+    const parent = await findComponentParentInstance(trx, identifiers, componentUid, componentId, excludeUid, caches);
+    if (!parent) {
+        caches.ancestorDpCache.set(cacheKey, false);
+        return false;
+    }
+    const result = await hasDraftPublishAncestorForParent(trx, identifiers, parent, caches);
+    caches.ancestorDpCache.set(cacheKey, result);
+    return result;
+}
+/**
+ * Abstracts `NOW()` handling so that timestamps stay consistent across databases—
+ * using Knex's native function when available and falling back to JS dates otherwise.
+ */ const resolveNowValue = (trx)=>{
+    if (typeof trx.fn?.now === 'function') {
+        return trx.fn.now();
+    }
+    return new Date();
+};
+/**
+ * Builds or retrieves the published→draft id map for a target content type, caching
+ * the result so nested relation remapping can reuse the work.
+ */ async function getDraftMapForTarget(trx, targetUid, draftMapCache) {
+    if (draftMapCache.has(targetUid)) {
+        return draftMapCache.get(targetUid) ?? null;
+    }
+    const targetMeta = strapi.db.metadata.get(targetUid);
+    if (!targetMeta) {
+        draftMapCache.set(targetUid, null);
+        return null;
+    }
+    const map = await buildPublishedToDraftMap({
+        trx,
+        uid: targetUid,
+        meta: targetMeta,
+        options: {
+            requireDraftAndPublish: true
+        }
+    });
+    draftMapCache.set(targetUid, map ?? null);
+    return map ?? null;
+}
+/**
+ * Maps relation foreign keys so that draft entities reference draft targets when those
+ * targets exist; otherwise we preserve the original reference (matching discardDraft).
+ */ async function mapTargetId(trx, originalId, targetUid, parentUid, parentPublishedToDraftMap, draftMapCache) {
+    if (originalId == null || !targetUid) {
+        return originalId;
+    }
+    if (targetUid === parentUid) {
+        return parentPublishedToDraftMap.get(Number(originalId)) ?? originalId;
+    }
+    const targetMap = await getDraftMapForTarget(trx, targetUid, draftMapCache);
+    if (!targetMap) {
+        return originalId;
+    }
+    return targetMap.get(Number(originalId)) ?? originalId;
+}
+/**
+ * Clones a database row and strips the `id` column so it can be reinserted as a new row.
+ */ const ensureObjectWithoutId = (row)=>{
+    const cloned = {
+        ...row
+    };
+    if ('id' in cloned) {
+        delete cloned.id;
+    }
+    return cloned;
+};
+/**
+ * Duplicates join-table relations for a component instance while remapping any foreign
+ * keys to draft targets. Mirrors the runtime clone logic but operates entirely in SQL.
+ */ async function cloneComponentRelationJoinTables(trx, componentMeta, componentUid, originalComponentId, newComponentId, parentUid, parentPublishedToDraftMap, draftMapCache) {
+    for (const [attributeName, attribute] of Object.entries(componentMeta.attributes)){
+        if (attribute.type !== 'relation' || !attribute.joinTable) {
+            continue;
+        }
+        const joinTable = attribute.joinTable;
+        const sourceColumnName = joinTable.joinColumn.name;
+        const targetColumnName = joinTable.inverseJoinColumn.name;
+        if (!componentMeta.relationsLogPrinted) {
+            debug(`[cloneComponentRelationJoinTables] Inspecting join table ${joinTable.name} for component ${componentUid}`);
+            componentMeta.relationsLogPrinted = true;
+        }
+        const relations = await trx(joinTable.name).select('*').where(sourceColumnName, originalComponentId);
+        if (relations.length === 0) {
+            continue;
+        }
+        const preparedRelations = [];
+        for (const relation of relations){
+            const clonedRelation = ensureObjectWithoutId(relation);
+            clonedRelation[sourceColumnName] = newComponentId;
+            if (targetColumnName in clonedRelation) {
+                const originalTargetId = clonedRelation[targetColumnName];
+                clonedRelation[targetColumnName] = await mapTargetId(trx, clonedRelation[targetColumnName], attribute.target, parentUid, parentPublishedToDraftMap, draftMapCache);
+                debug(`[cloneComponentRelationJoinTables] ${componentUid} join ${joinTable.name}: mapped ${targetColumnName} from ${originalTargetId} to ${clonedRelation[targetColumnName]} (target=${attribute.target})`);
+            }
+            preparedRelations.push(clonedRelation);
+        }
+        let relationsToInsert = preparedRelations;
+        if (preparedRelations.some((relation)=>targetColumnName in relation)) {
+            const { relations: safeRelations, skippedIds } = await filterRelationsWithExistingTargets({
+                trx,
+                targetUid: attribute.target,
+                relations: preparedRelations,
+                getTargetId: (relation)=>relation[targetColumnName]
+            });
+            recordSkippedRelations({
+                sourceUid: componentUid,
+                attributeName,
+                targetUid: attribute.target,
+                joinTable: joinTable.name
+            }, skippedIds);
+            relationsToInsert = safeRelations;
+        }
+        if (relationsToInsert.length === 0) {
+            continue;
+        }
+        for (const relation of relationsToInsert){
+            debug(`[cloneComponentRelationJoinTables] inserting relation into ${joinTable.name} (component=${componentUid}, source=${relation[sourceColumnName]})`);
+            await insertRowWithDuplicateHandling(trx, joinTable.name, relation, {
+                componentUid,
+                originalComponentId,
+                newComponentId,
+                joinTable: joinTable.name,
+                sourceColumnName,
+                targetColumnName,
+                targetUid: attribute.target,
+                parentUid
+            });
+        }
+    }
+}
+/**
+ * Clones a component row (including nested relations) so the newly created draft entity
+ * owns its own copy, matching what the document service would have produced.
+ */ async function cloneComponentInstance({ trx, componentUid, componentId, parentUid, parentPublishedToDraftMap, draftMapCache }) {
+    const componentMeta = getComponentMeta(componentUid);
+    if (!componentMeta) {
+        return componentId;
+    }
+    const componentTableName = componentMeta.tableName;
+    const componentPrimaryKey = Number.isNaN(Number(componentId)) ? componentId : Number(componentId);
+    const componentRow = await trx(componentTableName).select('*').where('id', componentPrimaryKey).first();
+    if (!componentRow) {
+        return componentId;
+    }
+    const newComponentRow = ensureObjectWithoutId(componentRow);
+    // `document_id`, `created_at`, `updated_at` are Strapi system columns whose names remain stable across the
+    // identifier-shortening migration, so it’s safe to check them directly here.
+    if ('document_id' in newComponentRow) {
+        newComponentRow.document_id = cuid2.createId();
+    }
+    if ('updated_at' in newComponentRow) {
+        newComponentRow.updated_at = resolveNowValue(trx);
+    }
+    if ('created_at' in newComponentRow && newComponentRow.created_at == null) {
+        newComponentRow.created_at = resolveNowValue(trx);
+    }
+    for (const attribute of Object.values(componentMeta.attributes)){
+        if (attribute.type !== 'relation') {
+            continue;
+        }
+        const joinColumn = attribute.joinColumn;
+        if (!joinColumn) {
+            continue;
+        }
+        const columnName = joinColumn.name;
+        if (!columnName || !(columnName in newComponentRow)) {
+            continue;
+        }
+        newComponentRow[columnName] = await mapTargetId(trx, newComponentRow[columnName], attribute.target, parentUid, parentPublishedToDraftMap, draftMapCache);
+    }
+    let insertResult;
+    if (supportsReturning(trx)) {
+        try {
+            insertResult = await trx(componentTableName).insert(newComponentRow, [
+                'id'
+            ]);
+        } catch (error) {
+            insertResult = await trx(componentTableName).insert(newComponentRow);
+        }
+    } else {
+        insertResult = await trx(componentTableName).insert(newComponentRow);
+    }
+    let newComponentId = resolveInsertedId(insertResult);
+    if (!newComponentId) {
+        if ('document_id' in newComponentRow && newComponentRow.document_id) {
+            const insertedRow = await trx(componentTableName).select('id').where('document_id', newComponentRow.document_id).orderBy('id', 'desc').first();
+            newComponentId = insertedRow?.id ?? null;
+        }
+        if (!newComponentId) {
+            const insertedRow = await trx(componentTableName).select('id').orderBy('id', 'desc').first();
+            newComponentId = insertedRow?.id ?? null;
+        }
+    }
+    if (!newComponentId) {
+        throw new Error(`Failed to clone component ${componentUid} (id: ${componentId})`);
+    }
+    newComponentId = Number(newComponentId);
+    if (Number.isNaN(newComponentId)) {
+        throw new Error(`Invalid cloned component identifier for ${componentUid} (id: ${componentId})`);
+    }
+    await cloneComponentRelationJoinTables(trx, componentMeta, componentUid, Number(componentPrimaryKey), newComponentId, parentUid, parentPublishedToDraftMap, draftMapCache);
+    return newComponentId;
+}
+/**
+ * Generates a map between published row ids and their corresponding draft ids so we can
+ * rewire relations in bulk. Handles localization nuances and caches the newest draft.
+ */ async function buildPublishedToDraftMap({ trx, uid, meta, options = {} }) {
+    if (!meta) {
+        return null;
+    }
+    const model = strapi.getModel(uid);
+    const hasDraftAndPublishEnabled = strapiUtils.contentTypes.hasDraftAndPublish(model);
+    if (options.requireDraftAndPublish && !hasDraftAndPublishEnabled) {
+        return null;
+    }
+    const [publishedEntries, draftEntries] = await Promise.all([
+        // `document_id`, `locale`, and `published_at` are core columns that keep their exact names after the
+        // identifier-shortening migration, so selecting them by literal is safe.
+        trx(meta.tableName).select([
             'id',
-            'documentId',
+            'document_id',
             'locale'
-        ]).where({
-            publishedAt: {
-                $ne: null
+        ]).whereNotNull('published_at'),
+        trx(meta.tableName).select([
+            'id',
+            'document_id',
+            'locale'
+        ]).whereNull('published_at')
+    ]);
+    if (publishedEntries.length === 0 || draftEntries.length === 0) {
+        return null;
+    }
+    const i18nService = strapi.plugin('i18n')?.service('content-types');
+    const contentType = strapi.contentTypes[uid];
+    const isLocalized = i18nService?.isLocalizedContentType(contentType) ?? false;
+    const draftByDocumentId = new Map();
+    for (const draft of draftEntries){
+        if (!draft.document_id) {
+            continue;
+        }
+        const key = isLocalized ? `${draft.document_id}:${draft.locale || ''}` : draft.document_id;
+        const existing = draftByDocumentId.get(key);
+        if (!existing) {
+            draftByDocumentId.set(key, draft);
+            continue;
+        }
+        const existingId = Number(existing.id);
+        const draftId = Number(draft.id);
+        if (Number.isNaN(existingId) || Number.isNaN(draftId)) {
+            draftByDocumentId.set(key, draft);
+            continue;
+        }
+        if (draftId > existingId) {
+            draftByDocumentId.set(key, draft);
+        }
+    }
+    const publishedToDraftMap = new Map();
+    for (const published of publishedEntries){
+        if (!published.document_id) {
+            continue;
+        }
+        const key = isLocalized ? `${published.document_id}:${published.locale || ''}` : published.document_id;
+        const draft = draftByDocumentId.get(key);
+        if (draft) {
+            const publishedId = normalizeId(published.id);
+            const draftId = normalizeId(draft.id);
+            if (publishedId == null || draftId == null) {
+                continue;
+            }
+            publishedToDraftMap.set(publishedId, draftId);
+        }
+    }
+    return publishedToDraftMap.size > 0 ? publishedToDraftMap : null;
+}
+/**
+ * Copy relations within the same content type (self-referential relations)
+ */ async function copyRelationsForContentType({ trx, uid, publishedToDraftMap }) {
+    const meta = strapi.db.metadata.get(uid);
+    if (!meta) return;
+    const publishedIds = Array.from(publishedToDraftMap.keys());
+    for (const attribute of Object.values(meta.attributes)){
+        if (attribute.type !== 'relation' || attribute.target !== uid) {
+            continue;
+        }
+        const joinTable = attribute.joinTable;
+        if (!joinTable) {
+            continue;
+        }
+        // Skip component join tables - they are handled by copyComponentRelations
+        if (joinTable.name.includes('_cmps')) {
+            continue;
+        }
+        const { name: sourceColumnName } = joinTable.joinColumn;
+        const { name: targetColumnName } = joinTable.inverseJoinColumn;
+        // Process in batches to avoid MySQL query size limits and SQLite expression tree limits
+        const publishedIdsChunks = chunkArray(publishedIds, getBatchSize(trx, 1000));
+        for (const publishedIdsChunk of publishedIdsChunks){
+            // Get relations where the source is a published entry (in batches)
+            const relationsQuery = trx(joinTable.name).select('*').whereIn(sourceColumnName, publishedIdsChunk);
+            applyJoinTableOrdering(relationsQuery, joinTable, sourceColumnName);
+            const relations = await relationsQuery;
+            if (relations.length === 0) {
+                continue;
+            }
+            // Create new relations pointing to draft entries
+            // Remove the 'id' field to avoid duplicate key errors
+            const newRelations = relations.map((relation)=>{
+                const newSourceId = getMappedValue(publishedToDraftMap, relation[sourceColumnName]);
+                const newTargetId = getMappedValue(publishedToDraftMap, relation[targetColumnName]);
+                if (!newSourceId || !newTargetId) {
+                    // Skip if no mapping found
+                    return null;
+                }
+                // Create new relation object without the 'id' field
+                const { id, ...relationWithoutId } = relation;
+                return {
+                    ...relationWithoutId,
+                    [sourceColumnName]: newSourceId,
+                    [targetColumnName]: newTargetId
+                };
+            }).filter(Boolean);
+            if (newRelations.length > 0) {
+                await trx.batchInsert(joinTable.name, newRelations, getBatchSize(trx, 1000));
+            }
+        }
+    }
+}
+/**
+ * Copy relations from other content types that target this content type
+ */ async function copyRelationsFromOtherContentTypes({ trx, uid, publishedToDraftMap }) {
+    const targetMeta = strapi.db.metadata.get(uid);
+    if (!targetMeta) {
+        return;
+    }
+    const publishedTargetIds = Array.from(publishedToDraftMap.keys()).map((value)=>normalizeId(value)).filter((value)=>value != null);
+    if (publishedTargetIds.length === 0) {
+        return;
+    }
+    const draftTargetIds = Array.from(publishedToDraftMap.values()).map((value)=>normalizeId(value)).filter((value)=>value != null);
+    const models = [
+        ...Object.values(strapi.contentTypes),
+        ...Object.values(strapi.components)
+    ];
+    // Compose a stable key for join-table rows so we can detect duplicates regardless
+    // of ordering or whether we already created the “draft” relation earlier.
+    const buildRelationKey = (relation, sourceColumnName, targetId)=>{
+        const sourceId = normalizeId(relation[sourceColumnName]) ?? relation[sourceColumnName];
+        const fieldValue = 'field' in relation ? relation.field ?? '' : '';
+        const componentTypeValue = 'component_type' in relation ? relation.component_type ?? '' : '';
+        return `${sourceId ?? 'null'}::${targetId ?? 'null'}::${fieldValue}::${componentTypeValue}`;
+    };
+    for (const model of models){
+        const dbModel = strapi.db.metadata.get(model.uid);
+        if (!dbModel) {
+            continue;
+        }
+        const sourceHasDraftAndPublish = Boolean(model.options?.draftAndPublish);
+        for (const attribute of Object.values(dbModel.attributes)){
+            if (attribute.type !== 'relation' || attribute.target !== uid) {
+                continue;
+            }
+            const joinTable = attribute.joinTable;
+            if (!joinTable) {
+                continue;
+            }
+            // Component join tables are handled separately when cloning components.
+            if (joinTable.name.includes('_cmps')) {
+                continue;
+            }
+            // If the source content type also has draft/publish, its own cloning routine will recreate its relations.
+            if (sourceHasDraftAndPublish) {
+                continue;
+            }
+            const { name: sourceColumnName } = joinTable.joinColumn;
+            const { name: targetColumnName } = joinTable.inverseJoinColumn;
+            const existingKeys = new Set();
+            if (draftTargetIds.length > 0) {
+                const draftIdChunks = chunkArray(draftTargetIds, getBatchSize(trx, 1000));
+                for (const draftChunk of draftIdChunks){
+                    const existingRelationsQuery = trx(joinTable.name).select('*').whereIn(targetColumnName, draftChunk);
+                    applyJoinTableOrdering(existingRelationsQuery, joinTable, sourceColumnName);
+                    const existingRelations = await existingRelationsQuery;
+                    for (const relation of existingRelations){
+                        existingKeys.add(buildRelationKey(relation, sourceColumnName, normalizeId(relation[targetColumnName])));
+                    }
+                }
+            }
+            const publishedIdChunks = chunkArray(publishedTargetIds, getBatchSize(trx, 1000));
+            for (const chunk of publishedIdChunks){
+                const relationsQuery = trx(joinTable.name).select('*').whereIn(targetColumnName, chunk);
+                applyJoinTableOrdering(relationsQuery, joinTable, sourceColumnName);
+                const relations = await relationsQuery;
+                if (relations.length === 0) {
+                    continue;
+                }
+                const newRelations = [];
+                for (const relation of relations){
+                    const newTargetId = getMappedValue(publishedToDraftMap, relation[targetColumnName]);
+                    if (!newTargetId) {
+                        continue;
+                    }
+                    const key = buildRelationKey(relation, sourceColumnName, newTargetId);
+                    if (existingKeys.has(key)) {
+                        continue;
+                    }
+                    existingKeys.add(key);
+                    const { id, ...relationWithoutId } = relation;
+                    newRelations.push({
+                        ...relationWithoutId,
+                        [targetColumnName]: newTargetId
+                    });
+                }
+                if (newRelations.length === 0) {
+                    continue;
+                }
+                try {
+                    await trx.batchInsert(joinTable.name, newRelations, getBatchSize(trx, 1000));
+                } catch (error) {
+                    if (!isDuplicateEntryError(error)) {
+                        throw error;
+                    }
+                    for (const relation of newRelations){
+                        await insertRowWithDuplicateHandling(trx, joinTable.name, relation, {
+                            reason: 'duplicate-draft-target-relation',
+                            sourceUid: model.uid,
+                            targetUid: uid
+                        });
+                    }
+                }
+            }
+        }
+    }
+}
+/**
+ * Copy relations from this content type that target other content types (category 3)
+ * Example: Article -> Categories/Tags
+ */ async function copyRelationsToOtherContentTypes({ trx, uid, publishedToDraftMap }) {
+    const meta = strapi.db.metadata.get(uid);
+    if (!meta) return;
+    const publishedIds = Array.from(publishedToDraftMap.keys());
+    // Cache target publishedToDraftMap to avoid duplicate calls for same target
+    const targetMapCache = new Map();
+    for (const [attributeName, attribute] of Object.entries(meta.attributes)){
+        if (attribute.type !== 'relation' || attribute.target === uid) {
+            continue;
+        }
+        const joinTable = attribute.joinTable;
+        if (!joinTable) {
+            continue;
+        }
+        // Skip component join tables - they are handled by copyComponentRelations
+        if (joinTable.name.includes('_cmps')) {
+            continue;
+        }
+        const { name: sourceColumnName } = joinTable.joinColumn;
+        const { name: targetColumnName } = joinTable.inverseJoinColumn;
+        // Get target content type's publishedToDraftMap if it has draft/publish (cached)
+        const targetUid = attribute.target;
+        if (!targetMapCache.has(targetUid)) {
+            const targetMeta = strapi.db.metadata.get(targetUid);
+            const targetMap = await buildPublishedToDraftMap({
+                trx,
+                uid: targetUid,
+                meta: targetMeta,
+                options: {
+                    requireDraftAndPublish: true
+                }
+            });
+            targetMapCache.set(targetUid, targetMap);
+        }
+        const targetPublishedToDraftMap = targetMapCache.get(targetUid);
+        // Process in batches to avoid MySQL query size limits and SQLite expression tree limits
+        const publishedIdsChunks = chunkArray(publishedIds, getBatchSize(trx, 1000));
+        for (const publishedIdsChunk of publishedIdsChunks){
+            // Get relations where the source is a published entry of our content type (in batches)
+            const relationsQuery = trx(joinTable.name).select('*').whereIn(sourceColumnName, publishedIdsChunk);
+            applyJoinTableOrdering(relationsQuery, joinTable, sourceColumnName);
+            const relations = await relationsQuery;
+            if (relations.length === 0) {
+                continue;
+            }
+            // Create new relations pointing to draft entries
+            // Remove the 'id' field to avoid duplicate key errors
+            const newRelations = relations.map((relation)=>{
+                const newSourceId = getMappedValue(publishedToDraftMap, relation[sourceColumnName]);
+                if (!newSourceId) {
+                    return null;
+                }
+                // Map target ID to draft if target has draft/publish enabled
+                // This matches discard() behavior: drafts relate to drafts
+                let newTargetId = relation[targetColumnName];
+                if (targetPublishedToDraftMap) {
+                    const mappedTargetId = getMappedValue(targetPublishedToDraftMap, relation[targetColumnName]);
+                    if (mappedTargetId) {
+                        newTargetId = mappedTargetId;
+                    }
+                // If no draft mapping, keep published target (target might not have DP or was deleted)
+                }
+                // Create new relation object without the 'id' field
+                const { id, ...relationWithoutId } = relation;
+                return {
+                    ...relationWithoutId,
+                    [sourceColumnName]: newSourceId,
+                    [targetColumnName]: newTargetId
+                };
+            }).filter(Boolean);
+            const { relations: safeRelations, skippedIds } = await filterRelationsWithExistingTargets({
+                trx,
+                targetUid,
+                relations: newRelations,
+                getTargetId: (relation)=>relation[targetColumnName]
+            });
+            recordSkippedRelations({
+                sourceUid: uid,
+                attributeName,
+                targetUid,
+                joinTable: joinTable.name
+            }, skippedIds);
+            if (safeRelations.length > 0) {
+                try {
+                    await trx.batchInsert(joinTable.name, safeRelations, 1000);
+                } catch (error) {
+                    // If batch insert fails due to duplicates, try with conflict handling
+                    if (error.code === '23505' || error.message?.includes('duplicate key')) {
+                        const client = trx.client.config.client;
+                        if (client === 'postgres' || client === 'pg') {
+                            for (const relation of safeRelations){
+                                try {
+                                    await trx(joinTable.name).insert(relation).onConflict().ignore();
+                                } catch (err) {
+                                    if (err.code !== '23505' && !err.message?.includes('duplicate key')) {
+                                        throw err;
+                                    }
+                                }
+                            }
+                        } else {
+                            throw error;
+                        }
+                    } else {
+                        throw error;
+                    }
+                }
+            }
+        }
+    }
+}
+/**
+ * Update JoinColumn relations (oneToOne, manyToOne foreign keys) to point to draft versions
+ * This matches discard() behavior: when creating drafts, foreign keys should point to draft targets
+ */ async function updateJoinColumnRelations({ db, trx, uid }) {
+    const meta = db.metadata.get(uid);
+    if (!meta) {
+        return;
+    }
+    // Create mapping from published entry ID to draft entry ID
+    const publishedToDraftMap = await buildPublishedToDraftMap({
+        trx,
+        uid,
+        meta
+    });
+    if (!publishedToDraftMap || publishedToDraftMap.size === 0) {
+        return;
+    }
+    // Cache target publishedToDraftMap to avoid duplicate calls for same target
+    const targetMapCache = new Map();
+    // Find all JoinColumn relations (oneToOne, manyToOne without joinTable)
+    for (const attribute of Object.values(meta.attributes)){
+        if (attribute.type !== 'relation') {
+            continue;
+        }
+        // Skip relations with joinTable (handled by copyRelationsToOtherContentTypes)
+        if (attribute.joinTable) {
+            continue;
+        }
+        // Only handle oneToOne and manyToOne relations that use joinColumn
+        const joinColumn = attribute.joinColumn;
+        if (!joinColumn) {
+            continue;
+        }
+        const targetUid = attribute.target;
+        const foreignKeyColumn = joinColumn.name;
+        // Get target content type's publishedToDraftMap if it has draft/publish (cached)
+        if (!targetMapCache.has(targetUid)) {
+            const targetMeta = strapi.db.metadata.get(targetUid);
+            const targetMap = await buildPublishedToDraftMap({
+                trx,
+                uid: targetUid,
+                meta: targetMeta,
+                options: {
+                    requireDraftAndPublish: true
+                }
+            });
+            targetMapCache.set(targetUid, targetMap);
+        }
+        const targetPublishedToDraftMap = targetMapCache.get(targetUid);
+        if (!targetPublishedToDraftMap) {
+            continue;
+        }
+        const draftIds = Array.from(publishedToDraftMap.values());
+        if (draftIds.length === 0) {
+            continue;
+        }
+        const draftIdsChunks = chunkArray(draftIds, getBatchSize(trx, 1000));
+        for (const draftIdsChunk of draftIdsChunks){
+            // Get draft entries with their foreign key values
+            const draftEntriesWithFk = await trx(meta.tableName).select([
+                'id',
+                foreignKeyColumn
+            ]).whereIn('id', draftIdsChunk).whereNotNull(foreignKeyColumn);
+            const updates = draftEntriesWithFk.reduce((acc, draftEntry)=>{
+                const publishedTargetIdRaw = draftEntry[foreignKeyColumn];
+                const normalizedPublishedTargetId = normalizeId(publishedTargetIdRaw);
+                const draftTargetId = normalizedPublishedTargetId == null ? undefined : targetPublishedToDraftMap.get(normalizedPublishedTargetId);
+                if (draftTargetId != null && normalizeId(draftTargetId) !== normalizedPublishedTargetId) {
+                    acc.push({
+                        id: draftEntry.id,
+                        draftTargetId
+                    });
+                }
+                return acc;
+            }, []);
+            if (updates.length === 0) {
+                continue;
+            }
+            const caseFragments = updates.map(()=>'WHEN ? THEN ?').join(' ');
+            const idsPlaceholders = updates.map(()=>'?').join(', ');
+            await trx.raw(`UPDATE ?? SET ?? = CASE ?? ${caseFragments} ELSE ?? END WHERE ?? IN (${idsPlaceholders})`, [
+                meta.tableName,
+                foreignKeyColumn,
+                'id',
+                ...updates.flatMap(({ id, draftTargetId })=>[
+                        id,
+                        draftTargetId
+                    ]),
+                foreignKeyColumn,
+                'id',
+                ...updates.map(({ id })=>id)
+            ]);
+        }
+    }
+}
+/**
+ * Copy component relations from published entries to draft entries
+ */ async function copyComponentRelations({ trx, uid, publishedToDraftMap }) {
+    const meta = strapi.db.metadata.get(uid);
+    if (!meta) {
+        return;
+    }
+    // Get collectionName from content type schema (Meta only has tableName which may be shortened)
+    const contentType = strapi.contentTypes[uid];
+    const collectionName = contentType?.collectionName;
+    if (!collectionName) {
+        return;
+    }
+    const identifiers = strapi.db.metadata.identifiers;
+    const joinTableName = transformContentTypesToModels.getComponentJoinTableName(collectionName, identifiers);
+    const entityIdColumn = transformContentTypesToModels.getComponentJoinColumnEntityName(identifiers);
+    const componentIdColumn = transformContentTypesToModels.getComponentJoinColumnInverseName(identifiers);
+    const componentTypeColumn = transformContentTypesToModels.getComponentTypeColumn(identifiers);
+    const fieldColumn = identifiers.FIELD_COLUMN;
+    const orderColumn = identifiers.ORDER_COLUMN;
+    // Check if component join table exists
+    const hasTable = await trx.schema.hasTable(joinTableName);
+    if (!hasTable) {
+        return;
+    }
+    const publishedIds = Array.from(publishedToDraftMap.keys());
+    // Process in batches to avoid MySQL query size limits and SQLite expression tree limits
+    const publishedIdsChunks = chunkArray(publishedIds, getBatchSize(trx, 1000));
+    for (const publishedIdsChunk of publishedIdsChunks){
+        // Get component relations for published entries
+        const componentRelations = await trx(joinTableName).select('*').whereIn(entityIdColumn, publishedIdsChunk);
+        if (componentRelations.length === 0) {
+            continue;
+        }
+        const componentCloneCache = new Map();
+        const componentTargetDraftMapCache = new Map();
+        const componentHierarchyCaches = {
+            parentInstanceCache: new Map(),
+            ancestorDpCache: new Map(),
+            parentDpCache: new Map()
+        };
+        // Filter component relations: only propagate if component's parent in the component hierarchy doesn't have draft/publish
+        // This matches discardDraft() behavior via shouldPropagateComponentRelationToNewVersion
+        //
+        // The logic: find what contains this component instance (could be a content type or another component).
+        // If it's a component, recursively check its parents. If any parent in the chain has DP, filter out the relation.
+        const filteredComponentRelations = await Promise.all(componentRelations.map(async (relation)=>{
+            const componentId = relation[componentIdColumn];
+            const componentType = relation[componentTypeColumn];
+            const entityId = relation[entityIdColumn];
+            const componentSchema = strapi.components[componentType];
+            if (!componentSchema) {
+                debug(`[copyComponentRelations] ${uid}: Keeping relation - unknown component type ${componentType} (entity: ${entityId}, componentId: ${componentId})`);
+                return relation;
+            }
+            const componentParent = await findComponentParentInstance(trx, identifiers, componentSchema.uid, componentId, uid, componentHierarchyCaches);
+            if (!componentParent) {
+                debug(`[copyComponentRelations] ${uid}: Keeping relation - component ${componentType} (id: ${componentId}) is directly on entity ${entityId} (no nested parent found)`);
+                return relation;
+            }
+            debug(`[copyComponentRelations] ${uid}: Component ${componentType} (id: ${componentId}, entity: ${entityId}) has parent in hierarchy: ${componentParent.uid} (parentId: ${componentParent.parentId})`);
+            const hasDPParent = await hasDraftPublishAncestorForParent(trx, identifiers, componentParent, componentHierarchyCaches);
+            if (hasDPParent) {
+                debug(`[copyComponentRelations] Filtering: component ${componentType} (id: ${componentId}, entity: ${entityId}) has DP parent in hierarchy (${componentParent.uid})`);
+                return null;
+            }
+            debug(`[copyComponentRelations] ${uid}: Keeping relation - component ${componentType} (id: ${componentId}, entity: ${entityId}) has no DP parent in hierarchy`);
+            return relation;
+        }));
+        // Filter out null values (filtered relations)
+        const relationsToProcess = filteredComponentRelations.filter(Boolean);
+        const filteredCount = componentRelations.length - relationsToProcess.length;
+        if (filteredCount > 0) {
+            debug(`[copyComponentRelations] ${uid}: Filtered ${filteredCount} of ${componentRelations.length} component relations (removed ${filteredCount} with DP parents)`);
+        }
+        // Create new component relations for draft entries
+        // Remove the 'id' field to avoid duplicate key errors
+        const mappedRelations = (await Promise.all(relationsToProcess.map(async (relation)=>{
+            const newEntityId = getMappedValue(publishedToDraftMap, relation[entityIdColumn]);
+            if (!newEntityId) {
+                return null;
+            }
+            const componentId = relation[componentIdColumn];
+            const componentType = relation[componentTypeColumn];
+            const componentKey = `${componentId}:${newEntityId}`;
+            let cloneMap = componentCloneCache.get(componentType);
+            if (!cloneMap) {
+                cloneMap = new Map();
+                componentCloneCache.set(componentType, cloneMap);
+            }
+            let newComponentId = cloneMap.get(componentKey);
+            if (!newComponentId) {
+                newComponentId = await cloneComponentInstance({
+                    trx,
+                    componentUid: componentType,
+                    componentId: Number(componentId),
+                    parentUid: uid,
+                    parentPublishedToDraftMap: publishedToDraftMap,
+                    draftMapCache: componentTargetDraftMapCache
+                });
+                cloneMap.set(componentKey, newComponentId);
+            }
+            const { id, ...relationWithoutId } = relation;
+            return {
+                ...relationWithoutId,
+                [entityIdColumn]: newEntityId,
+                [componentIdColumn]: newComponentId
+            };
+        }))).filter(Boolean);
+        // Deduplicate relations based on the unique constraint columns
+        // This prevents duplicates within the same batch that could cause conflicts
+        const uniqueKeyMap = new Map();
+        for (const relation of mappedRelations){
+            const uniqueKey = `${relation[entityIdColumn]}_${relation[componentIdColumn]}_${relation[fieldColumn]}_${relation[componentTypeColumn]}`;
+            if (!uniqueKeyMap.has(uniqueKey)) {
+                uniqueKeyMap.set(uniqueKey, relation);
+            }
+        }
+        const deduplicatedRelations = Array.from(uniqueKeyMap.values());
+        if (deduplicatedRelations.length === 0) {
+            continue;
+        }
+        // Check which relations already exist in the database to avoid conflicts
+        // We need to check all unique constraint columns (entity_id, cmp_id, field, component_type)
+        // Batch the check to avoid SQLite expression tree depth limits
+        // Use smaller batches for OR queries (50 for SQLite, 100 for others)
+        const batchSize = getBatchSize(trx, 50);
+        const relationChunks = chunkArray(deduplicatedRelations, batchSize);
+        const existingKeys = new Set();
+        for (const relationChunk of relationChunks){
+            const existingRelations = await trx(joinTableName).select([
+                entityIdColumn,
+                componentIdColumn,
+                fieldColumn,
+                componentTypeColumn
+            ]).where((qb)=>{
+                // Build OR conditions for each relation in this chunk
+                for (const relation of relationChunk){
+                    qb.orWhere((subQb)=>{
+                        subQb.where(entityIdColumn, relation[entityIdColumn]).where(componentIdColumn, relation[componentIdColumn]).where(fieldColumn, relation[fieldColumn]).where(componentTypeColumn, relation[componentTypeColumn]);
+                    });
+                }
+            });
+            // Add existing relation keys to the set
+            for (const existing of existingRelations){
+                const key = `${existing[entityIdColumn]}_${existing[componentIdColumn]}_${existing[fieldColumn]}_${existing[componentTypeColumn]}`;
+                existingKeys.add(key);
+            }
+        }
+        // Filter out relations that already exist
+        const newComponentRelations = deduplicatedRelations.filter((relation)=>{
+            const key = `${relation[entityIdColumn]}_${relation[componentIdColumn]}_${relation[fieldColumn]}_${relation[componentTypeColumn]}`;
+            return !existingKeys.has(key);
+        });
+        if (newComponentRelations.length > 0) {
+            // Insert component relations while surfacing unexpected constraint issues.
+            // Use INSERT ... ON CONFLICT DO NOTHING (PostgreSQL) or catch duplicate key errors explicitly for other clients.
+            const client = trx.client.config.client;
+            if (client === 'postgres' || client === 'pg') {
+                // PostgreSQL: Insert one at a time with ON CONFLICT DO NOTHING
+                // Use raw SQL for more reliable conflict handling
+                let insertedCount = 0;
+                let skippedCount = 0;
+                for (const relation of newComponentRelations){
+                    try {
+                        // Use raw SQL - ?? is for identifiers, ? is for values in Knex
+                        const orderValue = relation[orderColumn] ?? null;
+                        await trx.raw(`INSERT INTO ?? (??, ??, ??, ??, ??) VALUES (?, ?, ?, ?, ?) 
+               ON CONFLICT (??, ??, ??, ??) DO NOTHING`, [
+                            joinTableName,
+                            entityIdColumn,
+                            componentIdColumn,
+                            fieldColumn,
+                            componentTypeColumn,
+                            orderColumn,
+                            relation[entityIdColumn],
+                            relation[componentIdColumn],
+                            relation[fieldColumn],
+                            relation[componentTypeColumn],
+                            orderValue,
+                            entityIdColumn,
+                            componentIdColumn,
+                            fieldColumn,
+                            componentTypeColumn
+                        ]);
+                        insertedCount += 1;
+                    } catch (error) {
+                        // Ignore duplicate key errors (PostgreSQL error code 23505)
+                        // This can happen if the record already exists in the database
+                        if (error.code !== '23505' && !error.message?.includes('duplicate key')) {
+                            throw error;
+                        } else {
+                            skippedCount += 1;
+                        }
+                    }
+                }
+                if (insertedCount > 0 || skippedCount > 0) {
+                    debug(`[copyComponentRelations] ${uid}: Inserted ${insertedCount} component relations, skipped ${skippedCount} duplicates`);
+                }
+            } else if (client === 'mysql' || client === 'mysql2') {
+                try {
+                    await trx(joinTableName).insert(newComponentRelations);
+                } catch (error) {
+                    if (error.code !== 'ER_DUP_ENTRY') {
+                        throw error;
+                    }
+                    for (const relation of newComponentRelations){
+                        try {
+                            await trx(joinTableName).insert(relation);
+                        } catch (err) {
+                            if (err.code !== 'ER_DUP_ENTRY') {
+                                throw err;
+                            }
+                        }
+                    }
+                }
+            } else {
+                // SQLite: Try to insert and ignore errors
+                for (const relation of newComponentRelations){
+                    try {
+                        await trx(joinTableName).insert(relation);
+                    } catch (error) {
+                        // Ignore duplicate key errors
+                        if (error.code !== 'SQLITE_CONSTRAINT_UNIQUE' && error.code !== '23505') {
+                            throw error;
+                        }
+                    }
+                }
             }
-        }).limit(batchSize).offset(offset).orderBy('id').transacting(trx).execute();
-        if (batch.length < batchSize) {
-            hasMore = false;
         }
-        offset += batchSize;
-        yield batch;
     }
 }
 /**
  * 2 pass migration to create the draft entries for all the published entries.
- * And then discard the drafts to copy the relations.
+ * And then copy relations directly using database queries.
  */ const migrateUp = async (trx, db)=>{
+    strapi.log.info('[discard-drafts] Migration started');
     const dpModels = [];
     for (const meta of db.metadata.values()){
         const hasDP = await hasDraftAndPublish(trx, meta);
@@ -106,53 +1486,85 @@ var index = require('../../services/document-service/index.js');
             dpModels.push(meta);
         }
     }
+    debug(`Found ${dpModels.length} draft/publish content types to process`);
     /**
    * Create plain draft entries for all the entries that were published.
-   */ for (const model of dpModels){
+   */ strapi.log.info('[discard-drafts] Stage 1/3 – cloning published entries into draft rows');
+    for (const model of dpModels){
+        debug(` • cloning scalars for ${model.uid}`);
         await copyPublishedEntriesToDraft({
             db,
             trx,
             uid: model.uid
         });
     }
+    strapi.log.info('[discard-drafts] Stage 1/3 complete');
     /**
-   * Discard the drafts will copy the relations from the published entries to the newly created drafts.
-   *
-   * Load a batch of entries (batched to prevent loading millions of rows at once ),
-   * and discard them using the document service.
-   *
-   * NOTE: This is using a custom document service without any validations,
-   *       to prevent the migration from failing if users already had invalid data in V4.
-   *       E.g. @see https://github.com/strapi/strapi/issues/21583
-   */ const documentService = index.createDocumentService(strapi, {
-        async validateEntityCreation (_, data) {
-            return data;
-        },
-        async validateEntityUpdate (_, data) {
-            // Data can be partially empty on partial updates
-            // This migration doesn't trigger any update (or partial update),
-            // so it's safe to return the data as is.
-            return data;
-        }
-    });
+   * Copy relations from published entries to draft entries using direct database queries.
+   * This is much more efficient than calling discardDraft for each entry.
+   */ strapi.log.info('[discard-drafts] Stage 2/3 – copying relations and components to drafts');
     for (const model of dpModels){
-        const discardDraft = async (entry)=>documentService(model.uid).discardDraft({
-                documentId: entry.documentId,
-                locale: entry.locale
-            });
-        for await (const batch of getBatchToDiscard({
+        debug(` • copying relations for ${model.uid}`);
+        await copyRelationsToDrafts({
             db,
             trx,
             uid: model.uid
-        })){
-            // NOTE: concurrency had to be disabled to prevent a race condition with self-references
-            // TODO: improve performance in a safe way
-            await strapiUtils.async.map(batch, discardDraft, {
-                concurrency: 1
-            });
-        }
+        });
     }
+    flushSkippedRelationLogs();
+    strapi.log.info('[discard-drafts] Stage 2/3 complete');
+    /**
+   * Update JoinColumn relations (foreign keys) to point to draft versions
+   * This matches discard() behavior: drafts relate to drafts
+   */ strapi.log.info('[discard-drafts] Stage 3/3 – updating foreign key references to draft targets');
+    for (const model of dpModels){
+        debug(` • updating join columns for ${model.uid}`);
+        await updateJoinColumnRelations({
+            db,
+            trx,
+            uid: model.uid
+        });
+    }
+    strapi.log.info('[discard-drafts] Stage 3/3 complete');
+    strapi.log.info('[discard-drafts] Migration completed successfully');
 };
+/**
+ * Load a batch of versions to discard.
+ *
+ * Versions with only a draft version will be ignored.
+ * Only versions with a published version (which always have a draft version) will be discarded.
+ */ async function* getBatchToDiscard({ db, trx, uid, defaultBatchSize = 1000 }) {
+    const client = db.config.connection.client;
+    const isSQLite = typeof client === 'string' && [
+        'sqlite',
+        'sqlite3',
+        'better-sqlite3'
+    ].includes(client);
+    // The SQLite documentation states that the maximum number of terms in a
+    // compound SELECT statement is 500 by default.
+    // See: https://www.sqlite.org/limits.html
+    // To ensure a successful migration, we limit the batch size to 500 for SQLite.
+    const batchSize = isSQLite ? Math.min(defaultBatchSize, 500) : defaultBatchSize;
+    let offset = 0;
+    let hasMore = true;
+    while(hasMore){
+        // Look for the published entries to discard
+        const batch = await db.queryBuilder(uid).select([
+            'id',
+            'documentId',
+            'locale'
+        ]).where({
+            publishedAt: {
+                $ne: null
+            }
+        }).limit(batchSize).offset(offset).orderBy('id').transacting(trx).execute();
+        if (batch.length < batchSize) {
+            hasMore = false;
+        }
+        offset += batchSize;
+        yield batch;
+    }
+}
 const discardDocumentDrafts = {
     name: 'core::5.0.0-discard-drafts',
     async up (trx, db) {