Add batch deletion DAO support for stale metadata cleanup

CLAUDE.md

@@ -0,0 +1,85 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+DataHub GMA (General Metadata Architecture) is the backend framework for LinkedIn's DataHub metadata search & discovery
+platform. It provides type-safe, schema-driven metadata management with event-driven data consistency across multiple
+storage backends (SQL via Ebean, Elasticsearch for search). Built on LinkedIn's Pegasus data framework and Rest.li for
+REST APIs.
+
+**Java version**: 1.8 (Java 8) — required, newer versions will cause build failures.
+
+## Build Commands
+
+```bash
+./gradlew build                    # Full build + all tests
+./gradlew :module-name:build       # Build a specific module (e.g., :dao-api, :dao-impl:ebean-dao)
+./gradlew :module-name:test        # Run tests for a specific module
+./gradlew spotlessCheck            # Check code formatting (CI enforced)
+./gradlew spotlessApply            # Auto-fix formatting
+./gradlew checkstyleMain           # Run checkstyle on main sources
+./gradlew idea                     # Generate IntelliJ project files
+```
+
+Tests use **TestNG** (not JUnit) as the default test framework across all subprojects. The elasticsearch integration
+tests require Docker.
+
+**Apple Silicon (M-series Mac)**: Requires `brew install mariadb` and uncommenting three lines in
+`EmbeddedMariaInstance.java` (see `docs/developers.md`).
+
+## Module Architecture
+
+```
+core-models/          → Pegasus PDL schemas: Urn, AuditStamp, Url, Time (no Java logic)
+core-models-utils/    → URN utility helpers
+dao-api/              → DAO abstractions (BaseLocalDAO, BaseSearchDAO, BaseBrowseDAO),
+                        event producers, query utilities, retention policies
+dao-impl/
+  ebean-dao/          → SQL storage via Ebean ORM (EbeanLocalDAO, relationship queries)
+  elasticsearch-dao/  → ES 5.x/6.x search implementation
+  elasticsearch-dao-7/→ ES 7.x search implementation
+restli-resources/     → Rest.li resource base classes (BaseEntityResource,
+                        BaseSearchableEntityResource) mapping DAOs to REST endpoints
+validators/           → Schema validators ensuring PDL models conform to GMA conventions
+                        (AspectValidator, EntityValidator, SnapshotValidator, etc.)
+gradle-plugins/       → Annotation parsing (@gma) and code generation for metadata events
+testing/              → Test infrastructure, ES integration test harness, test models
+```
+
+## Key Architectural Patterns
+
+- **Urn (Universal Resource Name)**: `urn:li:entityType:entityKey` — the universal identifier for all entities. Typed
+  URN subclasses provide entity-specific keys.
+- **Aspect Union Pattern**: Each entity type defines a Pegasus union of its supported aspects. Validators enforce that
+  union members are record types only.
+- **Aspect Versioning**: Version 0 = latest. Each aspect write creates a new immutable version. Retention policies
+  (indefinite, time-based, version-based) control history.
+- **Layered Storage**: BaseLocalDAO (SQL, source of truth) → BaseSearchDAO (Elasticsearch, derived index) →
+  BaseBrowseDAO (hierarchical navigation). BaseRemoteDAO proxies to other GMS instances.
+- **Event Sourcing**: Writes to LocalDAO trigger MCE/MAE event emission via BaseMetadataEventProducer. The
+  `gradle-plugins` auto-generate event PDL schemas from `@gma` annotations on aspect PDL files.
+- **Generic Type Binding**: DAOs are heavily parameterized with generics (`<ASPECT_UNION, URN>`) and validate type
+  constraints at construction time using reflection via `ModelUtils`.
+
+## Pegasus/Rest.li Data Models
+
+PDL (Pegasus Data Language) schemas live in `src/main/pegasus/` directories and compile to Java `RecordTemplate`
+classes. Key namespaces:
+
+- `com.linkedin.common.*` — Core types (Urn, AuditStamp)
+- `com.linkedin.metadata.aspect.*` — Aspect wrappers
+- `com.linkedin.metadata.query.*` — Search/filter structures
+- `com.linkedin.metadata.events.*` — Change tracking types
+- `com.linkedin.metadata.snapshot.*` — Entity snapshots (versioned aspect collections)
+
+When modifying PDL schemas, the Pegasus gradle plugin regenerates Java bindings automatically during build.
+
+## Commit Convention
+
+Follow [Conventional Commits](https://www.conventionalcommits.org/): `<type>(scope): description`
+
+Types: `feat`, `fix`, `refactor`, `docs`, `test`, `perf`, `style`, `build`, `ci`
+
+Max line length: 88 characters. Use imperative present tense, no capitalized first letter, no trailing dot.

dao-api/src/main/java/com/linkedin/metadata/dao/EntityDeletionInfo.java

@@ -0,0 +1,27 @@
+package com.linkedin.metadata.dao;
+
+import java.sql.Timestamp;
+import java.util.Map;
+import lombok.Builder;
+import lombok.Value;
+
+
+/**
+ * A value class that holds deletion-relevant fields for a single entity, used by batch deletion validation.
+ * Contains status flags for deletion eligibility checks and all aspect column values for Kafka archival.
+ */
+@Value
+@Builder
+public class EntityDeletionInfo {
+
+  Timestamp deletedTs;
+
+  boolean statusRemoved;
+
+  String statusLastModifiedOn;
+
+  /**
+   * All aspect column values (column name → raw JSON string) for Kafka archival by the service layer.
+   */
+  Map<String, String> aspectColumns;
+}

dao-impl/ebean-dao/src/main/java/com/linkedin/metadata/dao/EbeanLocalAccess.java

@@ -72,6 +72,8 @@ public class EbeanLocalAccess<URN extends Urn> implements IEbeanLocalAccess<URN>
 
   // TODO confirm if the default page size is 1000 in other code context.
   private static final int DEFAULT_PAGE_SIZE = 1000;
+  private static final int MAX_BATCH_DELETE_SIZE = 2000;
+  private static final String STATUS_ASPECT_FQCN = "com.linkedin.common.Status";
   private static final String ASPECT_JSON_PLACEHOLDER = "__PLACEHOLDER__";
   private static final String DEFAULT_ACTOR = "urn:li:principal:UNKNOWN";
   private static final String EBEAN_SERVER_CONFIG = "EbeanServerConfig";
@@ -344,6 +346,58 @@ public int softDeleteAsset(@Nonnull URN urn, boolean isTestMode) {
     return _server.createSqlUpdate(deleteSqlStatement).execute();
   }
 
+  @Override
+  public Map<URN, EntityDeletionInfo> readDeletionInfoBatch(@Nonnull List<URN> urns, boolean isTestMode) {
+    if (urns.isEmpty()) {
+      return Collections.emptyMap();
+    }
+    if (urns.size() > MAX_BATCH_DELETE_SIZE) {
+      throw new IllegalArgumentException(
+          String.format("Batch size %d exceeds maximum of %d", urns.size(), MAX_BATCH_DELETE_SIZE));
+    }
+
+    final String statusColumnName = getStatusColumnName();
+    final Urn firstUrn = urns.get(0);
+    final String tableName = isTestMode ? getTestTableName(firstUrn) : getTableName(firstUrn);
+    final List<String> columns = getDeletionInfoColumns(tableName);
+    final String sql = SQLStatementUtils.createReadDeletionInfoByUrnsSql(urns, columns, isTestMode);
+    return EBeanDAOUtils.convertSqlRowsToEntityDeletionInfoMap(
+        _server.createSqlQuery(sql).findList(), _urnClass, statusColumnName);
+  }
+
+  /**
+   * Returns the column list needed for batch deletion info: urn, deleted_ts, and all aspect columns (a_*).
+   * Excludes index columns (i_*) and other derived columns to reduce data transfer.
+   */
+  private List<String> getDeletionInfoColumns(@Nonnull String tableName) {
+    return validator.getColumns(tableName).stream()
+        .filter(c -> c.equals("urn") || c.equals("deleted_ts") || c.startsWith(ASPECT_PREFIX))
+        .collect(Collectors.toList());
+  }
+
+  /**
+   * Resolves the entity table column name for the Status aspect via {@link SQLSchemaUtils}.
+   */
+  private String getStatusColumnName() {
+    return getAspectColumnName(_entityType, STATUS_ASPECT_FQCN);
+  }
+
+  @Override
+  public int batchSoftDeleteAssets(@Nonnull List<URN> urns, @Nonnull String cutoffTimestamp, boolean isTestMode) {
+    if (urns.isEmpty()) {
+      return 0;
+    }
+    if (urns.size() > MAX_BATCH_DELETE_SIZE) {
+      throw new IllegalArgumentException(
+          String.format("Batch size %d exceeds maximum of %d", urns.size(), MAX_BATCH_DELETE_SIZE));
+    }
+
+    final String statusColumnName = getStatusColumnName();
+    final String sql = SQLStatementUtils.createBatchSoftDeleteAssetSql(urns, cutoffTimestamp, statusColumnName,
+        isTestMode);
+    return _server.createSqlUpdate(sql).execute();
+  }
+
   @Override
   public List<URN> listUrns(@Nullable IndexFilter indexFilter, @Nullable IndexSortCriterion indexSortCriterion,
       @Nullable URN lastUrn, int pageSize) {

dao-impl/ebean-dao/src/main/java/com/linkedin/metadata/dao/EbeanLocalDAO.java

@@ -750,6 +750,20 @@ protected int permanentDelete(@Nonnull URN urn, boolean isTestMode) {
     return _localAccess.softDeleteAsset(urn, isTestMode);
   }
 
+  /**
+   * Delegates to {@link IEbeanLocalAccess#readDeletionInfoBatch}.
+   */
+  public Map<URN, EntityDeletionInfo> readDeletionInfoBatch(@Nonnull List<URN> urns, boolean isTestMode) {
+    return _localAccess.readDeletionInfoBatch(urns, isTestMode);
+  }
+
+  /**
+   * Delegates to {@link IEbeanLocalAccess#batchSoftDeleteAssets}.
+   */
+  public int batchSoftDeleteAssets(@Nonnull List<URN> urns, @Nonnull String cutoffTimestamp, boolean isTestMode) {
+    return _localAccess.batchSoftDeleteAssets(urns, cutoffTimestamp, isTestMode);
+  }
+
   @Override
   public <ASPECT extends RecordTemplate> void updateEntityTables(@Nonnull URN urn, @Nonnull Class<ASPECT> aspectClass) {
     if (_schemaConfig == SchemaConfig.OLD_SCHEMA_ONLY) {

dao-impl/ebean-dao/src/main/java/com/linkedin/metadata/dao/IEbeanLocalAccess.java

@@ -97,6 +97,30 @@ <ASPECT extends RecordTemplate> List<EbeanMetadataAspect> batchGetUnion(@Nonnull
    */
   int softDeleteAsset(@Nonnull URN urn, boolean isTestMode);
 
+  /**
+   * Read deletion-relevant fields for a batch of URNs in a single SELECT.
+   * Returns deletion-relevant fields for validation and all aspect columns for Kafka archival.
+   * URNs not found in the database will not have entries in the returned map.
+   *
+   * @param urns list of URNs to check
+   * @param isTestMode whether to use test schema
+   * @return map of URN to {@link EntityDeletionInfo}
+   */
+  Map<URN, EntityDeletionInfo> readDeletionInfoBatch(@Nonnull List<URN> urns, boolean isTestMode);
+
+  /**
+   * Batch soft-delete entities by setting deleted_ts = NOW() for URNs that meet all deletion criteria.
+   * The UPDATE includes guard clauses (deleted_ts IS NULL, Status.removed = true, lastmodifiedon &lt; cutoff)
+   * as defense-in-depth against race conditions. The Status aspect column name is resolved internally via
+   * {@link com.linkedin.metadata.dao.utils.SQLSchemaUtils#getAspectColumnName}.
+   *
+   * @param urns list of URNs to soft-delete
+   * @param cutoffTimestamp only delete if Status.lastmodifiedon is before this timestamp
+   * @param isTestMode whether to use test schema
+   * @return number of rows actually soft-deleted
+   */
+  int batchSoftDeleteAssets(@Nonnull List<URN> urns, @Nonnull String cutoffTimestamp, boolean isTestMode);
+
   /**
    * Returns list of urns that satisfy the given filter conditions.
    *

dao-impl/ebean-dao/src/main/java/com/linkedin/metadata/dao/InstrumentedEbeanLocalAccess.java

@@ -99,6 +99,18 @@ public int softDeleteAsset(@Nonnull URN urn, boolean isTestMode) {
     return instrument("softDeleteAsset", () -> _delegate.softDeleteAsset(urn, isTestMode));
   }
 
+  @Override
+  public Map<URN, EntityDeletionInfo> readDeletionInfoBatch(@Nonnull List<URN> urns, boolean isTestMode) {
+    return instrument("readDeletionInfoBatch.urns_" + urns.size(),
+        () -> _delegate.readDeletionInfoBatch(urns, isTestMode));
+  }
+
+  @Override
+  public int batchSoftDeleteAssets(@Nonnull List<URN> urns, @Nonnull String cutoffTimestamp, boolean isTestMode) {
+    return instrument("batchSoftDeleteAssets.urns_" + urns.size(),
+        () -> _delegate.batchSoftDeleteAssets(urns, cutoffTimestamp, isTestMode));
+  }
+
   @Override
   public List<URN> listUrns(@Nullable IndexFilter indexFilter,
       @Nullable IndexSortCriterion indexSortCriterion, @Nullable URN lastUrn, int pageSize) {

dao-impl/ebean-dao/src/main/java/com/linkedin/metadata/dao/utils/EBeanDAOUtils.java

@@ -1,6 +1,7 @@
 package com.linkedin.metadata.dao.utils;
 
 import com.linkedin.common.urn.Urn;
+import com.linkedin.data.DataMap;
 import com.linkedin.data.schema.RecordDataSchema;
 import com.linkedin.data.template.DataTemplateUtil;
 import com.linkedin.data.template.RecordTemplate;
@@ -12,6 +13,7 @@
 import com.linkedin.metadata.aspect.AuditedAspect;
 import com.linkedin.metadata.aspect.SoftDeletedAspect;
 import com.linkedin.metadata.dao.EbeanMetadataAspect;
+import com.linkedin.metadata.dao.EntityDeletionInfo;
 import com.linkedin.metadata.dao.ListResult;
 import com.linkedin.metadata.query.AspectField;
 import com.linkedin.metadata.query.Condition;
@@ -257,6 +259,73 @@ public static <ASPECT extends RecordTemplate> List<EbeanMetadataAspect> readSqlR
     }).collect(Collectors.toList());
   }
 
+  /**
+   * Parse a list of {@link SqlRow} results from an entity table into a map of
+   * URN to {@link EntityDeletionInfo}. Each row must contain urn, deleted_ts, and the Status aspect column.
+   * Rows that cannot be parsed as a valid URN are skipped with a warning.
+   *
+   * @param sqlRows list of {@link SqlRow} from entity table query
+   * @param urnClass URN class for deserialization
+   * @param statusColumnName the entity table column name for the Status aspect (e.g. "a_status")
+   * @param <URN> URN type
+   * @return map of URN to {@link EntityDeletionInfo}
+   */
+  public static <URN extends Urn> Map<URN, EntityDeletionInfo> convertSqlRowsToEntityDeletionInfoMap(
+      @Nonnull List<SqlRow> sqlRows, @Nonnull Class<URN> urnClass, @Nonnull String statusColumnName) {
+    final Map<URN, EntityDeletionInfo> result = new HashMap<>();
+    for (SqlRow row : sqlRows) {
+      final String urnStr = row.getString("urn");
+      try {
+        result.put(getUrn(urnStr, urnClass), toEntityDeletionInfo(row, statusColumnName));
+      } catch (IllegalArgumentException e) {
+        log.warn("Failed to parse URN string: {}, skipping row", urnStr, e);
+      }
+    }
+    return result;
+  }
+
+  /**
+   * Parse a single {@link SqlRow} from an entity table query into an {@link EntityDeletionInfo}.
+   * Extracts deletion eligibility fields from the Status aspect column (statusRemoved, statusLastModifiedOn)
+   * and collects all aspect columns for Kafka archival.
+   *
+   * @param row {@link SqlRow} from entity table query
+   * @param statusColumnName the entity table column name for the Status aspect (e.g. "a_status")
+   * @return {@link EntityDeletionInfo}
+   */
+  @Nonnull
+  static EntityDeletionInfo toEntityDeletionInfo(@Nonnull SqlRow row, @Nonnull String statusColumnName) {
+    // Collect all aspect columns (a_* prefixed, non-null), same pattern as readSqlRows()
+    final Map<String, String> aspectColumnValues = new HashMap<>();
+    for (String key : row.keySet()) {
+      if (key.startsWith(SQLSchemaUtils.ASPECT_PREFIX) && row.get(key) != null) {
+        aspectColumnValues.put(key, row.getString(key));
+      }
+    }
+
+    boolean statusRemoved = false;
+    String statusLastModifiedOn = null;
+    final String statusJson = row.getString(statusColumnName);
+    if (statusJson != null) {
+      final DataMap statusData = RecordUtils.toDataMap(statusJson);
+      final Object lastModObj = statusData.get("lastmodifiedon");
+      if (lastModObj != null) {
+        statusLastModifiedOn = lastModObj.toString();
+      }
+      final Object aspectObj = statusData.get("aspect");
+      if (aspectObj instanceof DataMap) {
+        statusRemoved = Boolean.TRUE.equals(((DataMap) aspectObj).get("removed"));
+      }
+    }
+
+    return EntityDeletionInfo.builder()
+        .deletedTs(row.getTimestamp("deleted_ts"))
+        .statusRemoved(statusRemoved)
+        .statusLastModifiedOn(statusLastModifiedOn)
+        .aspectColumns(aspectColumnValues)
+        .build();
+  }
+
   /**
    * Read EbeanMetadataAspect from {@link SqlRow}.
    * @param sqlRow {@link SqlRow}