Sheetal Shah created ATLAS-5381:
-----------------------------------
Summary: Entity delete and purge fail when traversing empty owned
reference vertices
Key: ATLAS-5381
URL: https://issues.apache.org/jira/browse/ATLAS-5381
Project: Atlas
Issue Type: Improvement
Components: atlas-core
Reporter: Sheetal Shah
h2. Summary
Entity delete and purge can fail when Atlas traverses owned references (e.g.
{{columns}} on {{{}hive_table{}}}) and hits empty reference vertices: the
vertex and relationship edge still exist, but the vertex has no properties
({{{}__typeName{}}}, {{{}__guid{}}}, etc.).
On read, Atlas tolerates this (ATLAS-4605) and returns {{null}} in arrays like
{{{}relationshipAttributes.columns{}}}. On delete/purge, {{getOwnedVertices()}}
throws and the parent operation fails.
h2. Error
{"errorCode":"ATLAS-400-00-014","errorMessage":"Type ENTITY with name null does
not exist"}
org.apache.atlas.exception.AtlasBaseException: Type ENTITY with name null does
not exist
at
org.apache.atlas.repository.store.graph.v1.DeleteHandlerV1.getOwnedVertices(DeleteHandlerV1.java:218)
at
org.apache.atlas.repository.store.graph.v1.DeleteHandlerV1.deleteEntities(DeleteHandlerV1.java:123)
at
org.apache.atlas.repository.store.graph.v2.AtlasEntityStoreV2.purgeVertices(AtlasEntityStoreV2.java:1385)
at
org.apache.atlas.repository.store.graph.v2.AtlasEntityStoreV2.purgeByIds(AtlasEntityStoreV2.java:586)
Root cause
h3. Graph state
Some owned reference vertices are in an inconsistent state:
* Relationship edge from parent → child still exists
* Target vertex exists in graph storage
* Target vertex has empty or missing properties ({{{}__typeName{}}} is null)
This can occur due to:
* Partial hard purge / incomplete composite cleanup
* Concurrent purge and delete on related entities
* Incomplete metadata sync (for example, column dropped in source system but
Atlas graph left with dangling edges)
* Failed or interrupted delete/purge leaving vertex shells behind
h3. Read vs delete behavior mismatch
Read path ({{{}EntityGraphRetriever{}}}) — handled since ATLAS-4605:
* When a reference vertex has no property keys, Atlas logs a warning and
returns {{null}} in the relationship array
* GET succeeds but shows {{null}} entries in arrays like {{columns}}
Delete/purge traversal path ({{{}DeleteHandlerV1.getOwnedVertices{}}}) — not
handled:
* Traverses owned reference attributes via relationship edges
* Calls {{toAtlasEntityHeader()}} on each referenced vertex
* If {{typeName}} is null, throws {{TYPE_NAME_INVALID}}
({{{}ATLAS-400-00-014{}}})
* Parent entity delete/purge fails entirely
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
