Concurrent Transactions on JG Edges


Aman
 

Hi, 

Is there a way by which I can prevent concurrent updates to edges? For vertices, I'm simply maintaining a version field with Lock consistency, so a concurrent update results in Permanent Locking Exception being thrown, but the same doesn't seem to work for edges.

--
Regards,
Aman


Amandeep Srivastava <amandeep.srivastava1996@...>
 

Hi, 

Is there a way by which I can prevent concurrent updates to edges? For vertices, I'm simply maintaining a version field with Lock consistency, so a concurrent update results in Permanent Locking Exception being thrown, but the same doesn't seem to work for edges.

--
Regards,
Aman


hadoopmarc@...
 

Hi Aman,

The JanusGraph docs on consistency, cqlLocking do not limit its use to vertices. Can you please show how you translated the example from https://docs.janusgraph.org/advanced-topics/eventual-consistency/#data-consistency for vertex properties to the locking of edge properties?

Best wishes,    Marc


Aman
 

Hi Marc,

Thank you for your response. This is how I'm using it:

1. I created a property key called version, which is incremented at the service layer every time an update is made on an element (vertex or edge). 

mgmt = graph.openManagement()
version = mgmt.makePropertyKey('version').dataType(Integer.class).make()
mgmt.setConsistency(name, ConsistencyModifier.LOCK)
mgmt.buildIndex("vertexIndex", Vertex.class).addKey(version, Mapping.DEFAULT.asParameter()).buildMixedIndex("esIndex")
mgmt.buildIndex("edgeIndex", Edge.class).addKey(version, Mapping.DEFAULT.asParameter()).buildMixedIndex("esIndex")
mgmt.commit()

In case mixed index already exists, I simply add the property in both vertex and edge mixed indices:
index = mgmt.getGraphIndex("edgeIndex or vertexIndex")
mgmt.addIndexKey(index, version)
In case

2. For every new edge label I create, I'm attaching this property to the edge.

mgmt = graph.openManagement() relatedTo = mgmt.makeEdgeLabel('relatedTo').make() mgmt.addProperties(relatedTo, version) mgmt.commit()

How I tested: When I'm making multiple concurrent calls via Jmeter to update an existing vertex, only one request passes, and rest throw a permanent locking exception. However, when I run same test for edges, multiple updates are successful. Few that fail are because of version check logic at service layer which kicks in a few ms after first few updates are successful.

I'm using Janus with Cassandra and Elastic Search.

Regards,

Aman


Aman
 

Hi,

Can someone please advise if previously mentioned way is incorrect for using lock consistencies on edge property keys (version in my case)?

Regards,
Aman

On Mon, 10 Jan, 2022, 10:26 pm Aman via lists.lfaidata.foundation, <amandeep.srivastava1996=gmail.com@...> wrote:
Hi Marc,

Thank you for your response. This is how I'm using it:

1. I created a property key called version, which is incremented at the service layer every time an update is made on an element (vertex or edge). 

mgmt = graph.openManagement()
version = mgmt.makePropertyKey('version').dataType(Integer.class).make()
mgmt.setConsistency(name, ConsistencyModifier.LOCK)
mgmt.buildIndex("vertexIndex", Vertex.class).addKey(version, Mapping.DEFAULT.asParameter()).buildMixedIndex("esIndex")
mgmt.buildIndex("edgeIndex", Edge.class).addKey(version, Mapping.DEFAULT.asParameter()).buildMixedIndex("esIndex")
mgmt.commit()

In case mixed index already exists, I simply add the property in both vertex and edge mixed indices:
index = mgmt.getGraphIndex("edgeIndex or vertexIndex")
mgmt.addIndexKey(index, version)
In case

2. For every new edge label I create, I'm attaching this property to the edge.

mgmt = graph.openManagement() relatedTo = mgmt.makeEdgeLabel('relatedTo').make() mgmt.addProperties(relatedTo, version) mgmt.commit()

How I tested: When I'm making multiple concurrent calls via Jmeter to update an existing vertex, only one request passes, and rest throw a permanent locking exception. However, when I run same test for edges, multiple updates are successful. Few that fail are because of version check logic at service layer which kicks in a few ms after first few updates are successful.

I'm using Janus with Cassandra and Elastic Search.

Regards,

Aman