Possible bug: java.lang.StackOverflowError on large query

[mhilker]

We are trying to execute a large query to seed our database on 0.6.1 and are seeing a java.lang.StackOverflowError on the Gremlin Console. I am quite sure this bug has to be in janusgraph, because I have seen the same error in our applications API. The same query works fine on 0.6.0.

java.lang.StackOverflowError
        at org.codehaus.groovy.classgen.asm.InvocationWriter.writeDirectMethodCall(InvocationWriter.java:154)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.makeDirectCall(InvocationWriter.java:335)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.makeCall(InvocationWriter.java:439)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.makeCall(InvocationWriter.java:150)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.makeInvokeMethodCall(InvocationWriter.java:124)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.writeInvokeMethod(InvocationWriter.java:509)
        at org.codehaus.groovy.classgen.AsmClassGenerator.visitMethodCallExpression(AsmClassGenerator.java:868)
        at org.codehaus.groovy.ast.expr.MethodCallExpression.visit(MethodCallExpression.java:68)
        at org.codehaus.groovy.classgen.asm.CallSiteWriter.prepareSiteAndReceiver(CallSiteWriter.java:254)
        at org.codehaus.groovy.classgen.asm.CallSiteWriter.prepareSiteAndReceiver(CallSiteWriter.java:243)
        at org.codehaus.groovy.classgen.asm.CallSiteWriter.makeCallSite(CallSiteWriter.java:291)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.makeCachedCall(InvocationWriter.java:352)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.makeCall(InvocationWriter.java:442)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.makeCall(InvocationWriter.java:150)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.makeInvokeMethodCall(InvocationWriter.java:124)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.writeInvokeMethod(InvocationWriter.java:509)
        at org.codehaus.groovy.classgen.AsmClassGenerator.visitMethodCallExpression(AsmClassGenerator.java:868)
        at org.codehaus.groovy.ast.expr.MethodCallExpression.visit(MethodCallExpression.java:68)
        at org.codehaus.groovy.classgen.asm.CallSiteWriter.prepareSiteAndReceiver(CallSiteWriter.java:254)
        at org.codehaus.groovy.classgen.asm.CallSiteWriter.prepareSiteAndReceiver(CallSiteWriter.java:243)
        at org.codehaus.groovy.classgen.asm.CallSiteWriter.makeCallSite(CallSiteWriter.java:291)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.makeCachedCall(InvocationWriter.java:352)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.makeCall(InvocationWriter.java:442)
        at org.codehaus.groovy.classgen.asm.InvocationWriter.makeCall(InvocationWriter.java:150)
[this goes on for about 1000 lines...]

Steps to reproduce:

• download janusgraph 0.6.1
• spin up the server
• start the gremlin console
• connect to janusgraph
  • :remote connect tinkerpop.server conf/remote.yaml
  • :> graph=TinkerGraph.open()
  • :> g=graph.traversal()
• execute the query
  • :> [insert the query from below here]

I have added the newlines to the query for better readability.

g
.addV('vertex_a').property('name', 'vertex_a_1').as('vertex_a_1')
.addV('vertex_a').property('name', 'vertex_a_2').as('vertex_a_2')
.addV('vertex_b').property('name', 'vertex_b_1').as('vertex_b_1')
.addV('vertex_b').property('name', 'vertex_b_2').as('vertex_b_2')
.addV('vertex_b').property('name', 'vertex_b_3').as('vertex_b_3')
.addV('vertex_b').property('name', 'vertex_b_4').as('vertex_b_4')
.addV('vertex_b').property('name', 'vertex_b_5').as('vertex_b_5')
.addE('edge_a').from('vertex_a_1').to('vertex_b_1')
.addE('edge_b').from('vertex_a_1').to('vertex_b_2')
.addE('edge_b').from('vertex_a_1').to('vertex_b_3')
.addE('edge_b').from('vertex_a_2').to('vertex_b_4')
.addE('edge_b').from('vertex_a_1').to('vertex_b_5')
.addE('edge_c').from('vertex_a_1').to('vertex_b_4')

.addV('vertex_c').property('name', 'vertex_c_1').as('vertex_c_1')
.addV('vertex_c').property('name', 'vertex_c_2').as('vertex_c_2')
.addE('edge_d').from('vertex_b_1').to('vertex_c_1')
.addE('edge_d').from('vertex_b_2').to('vertex_c_2')
.addE('edge_e').from('vertex_b_2').to('vertex_c_1')
.addE('edge_e').from('vertex_b_3').to('vertex_c_1')
.addE('edge_f').from('vertex_b_4').to('vertex_c_1')
.addE('edge_e').from('vertex_b_1').to('vertex_c_2')
.addE('edge_e').from('vertex_b_3').to('vertex_c_2')

.addV('vertex_d').property('name', 'vertex_d_1').as('vertex_d_1')
.addV('vertex_d').property('name', 'vertex_d_2').as('vertex_d_2')
.addV('vertex_d').property('name', 'vertex_d_3').as('vertex_d_3')
.addE('edge_f').from('vertex_c_1').to('vertex_d_1')
.addE('edge_f').from('vertex_c_1').to('vertex_d_3')
.addE('edge_f').from('vertex_c_2').to('vertex_d_2')

.addV('vertex_e').property('name', 'vertex_e_1').as('vertex_e_1')
.addV('vertex_e').property('name', 'vertex_e_2').as('vertex_e_2')
.addV('vertex_e').property('name', 'vertex_e_3').as('vertex_e_3')
.addV('vertex_e').property('name', 'vertex_e_1').as('vertex_e_4')
.addV('vertex_f').property('name', 'vertex_f_1').as('vertex_f_1')
.addE('edge_f').from('vertex_d_1').to('vertex_e_1')
.addE('edge_f').from('vertex_d_1').to('vertex_e_2')
.addE('edge_f').from('vertex_d_2').to('vertex_e_3')
.addE('edge_f').from('vertex_d_3').to('vertex_e_4')
.addE('edge_f').from('vertex_d_1').to('vertex_f_1')
.addV('vertex_g').as('vertex_g_1')

.addV('vertex_h').as('vertex_h_1')
.addV('vertex_i').property('name', 'vertex_i_1').as('vertex_i_1')
.addV('vertex_i').property('name', 'vertex_i_2').as('vertex_i_2')
.addV('vertex_i').property('name', 'vertex_i_3').as('vertex_i_3')
.addV('vertex_i').property('name', 'vertex_i_4').as('vertex_i_4')
.addV('vertex_i').property('name', 'vertex_i_5').as('vertex_i_5')
.addV('vertex_j').as('vertex_j_1')
.addV('vertex_k').as('vertex_k_1')
.addV('vertex_l').as('vertex_l_1')
.addE('edge_f').from('vertex_e_1').to('vertex_g_1')
.addE('edge_f').from('vertex_e_1').to('vertex_h_1')
.addE('edge_f').from('vertex_e_1').to('vertex_i_1')
.addE('edge_f').from('vertex_e_1').to('vertex_i_2')
.addE('edge_f').from('vertex_e_1').to('vertex_i_3')
.addE('edge_f').from('vertex_e_1').to('vertex_i_4')
.addE('edge_f').from('vertex_e_1').to('vertex_i_5')
.addE('edge_f').from('vertex_e_1').to('vertex_j_1')
.addE('edge_f').from('vertex_e_1').to('vertex_k_1')
.addE('edge_f').from('vertex_e_1').to('vertex_l_1')
.addE('edge_g').from('vertex_g_1').to('vertex_h_1')
.addE('edge_g').from('vertex_h_1').to('vertex_i_1')
.addE('edge_g').from('vertex_h_1').to('vertex_i_2')
.addE('edge_g').from('vertex_h_1').to('vertex_i_3')
.addE('edge_g').from('vertex_h_1').to('vertex_i_4')
.addE('edge_g').from('vertex_h_1').to('vertex_i_5')
.addE('edge_g').from('vertex_i_1').to('vertex_k_1')
.addE('edge_g').from('vertex_k_1').to('vertex_i_1')
.addE('edge_g').from('vertex_i_2').to('vertex_k_1')
.addE('edge_g').from('vertex_i_3').to('vertex_k_1')
.addE('edge_g').from('vertex_i_4').to('vertex_k_1')
.addE('edge_g').from('vertex_i_5').to('vertex_k_1')
.addE('edge_g').from('vertex_k_1').to('vertex_j_1')
.addE('edge_h').from('vertex_j_1').to('vertex_e_2')
.addE('edge_g').from('vertex_j_1').to('vertex_l_1')

.addV('vertex_g').as('vertex_g_1')
.addV('vertex_i').property('name', 'vertex_i_1').as('vertex_i_1')
.addV('vertex_i').property('name', 'vertex_i_2').as('vertex_i_2')
.addV('vertex_m').property('name', 'vertex_m_1').as('vertex_m_1')
.addV('vertex_l').as('vertex_l_1')
.addE('edge_f').from('vertex_e_2').to('vertex_g_1')
.addE('edge_f').from('vertex_e_2').to('vertex_i_1')
.addE('edge_f').from('vertex_e_2').to('vertex_i_2')
.addE('edge_f').from('vertex_e_2').to('vertex_m_1')
.addE('edge_f').from('vertex_e_2').to('vertex_l_1')
.addE('edge_g').from('vertex_g_1').to('vertex_i_1')
.addE('edge_g').from('vertex_g_1').to('vertex_i_2')
.addE('edge_g').from('vertex_i_1').to('vertex_m_1')
.addE('edge_g').from('vertex_i_2').to('vertex_m_1')
.addE('edge_g').from('vertex_m_1').to('vertex_l_1')

[FlorianHockmann]

:> graph=TinkerGraph.open()
:> g=graph.traversal()

This part shows that you're not really using JanusGraph, but TinkerGraph which is an inmemory graph implementation from TinkerPop. It would be great if you could try out whether you can reproduce this also directly with Gremlin Server from TinkerPop. You can download that from here.
I would probably first try whether you can reproduce it with the newest version of Gremlin Server. If that doesn't work, then it would be good to see whether you can reproduce it with Gremlin Server 3.5.1 which is the version JanusGraph 0.6.1 uses.
If you can reproduce it with Gremlin Server 3.5.2, then please create an issue with the TinkerPop project. If it works with Gremlin Server 3.5.2, but not with 3.5.1, then it's probably already fixed and we just need to update JanusGraph to that newer TinkerPop version.

[mhilker]

I would probably first try whether you can reproduce it with the newest version of Gremlin Server. If that doesn't work, then it would be good to see whether you can reproduce it with Gremlin Server 3.5.1 which is the version JanusGraph 0.6.1 uses.
If you can reproduce it with Gremlin Server 3.5.2, then please create an issue with the TinkerPop project. If it works with Gremlin Server 3.5.2, but not with 3.5.1, then it's probably already fixed and we just need to update JanusGraph to that newer TinkerPop version.

I've tested both versions 3.5.1 and 3.5.2 of the gremlin-console and gremlin-server using docker/docker-compose. The query worked fine both times.

Started the gremlin server like this:

services:
  gremlin-server:
    # also tried version 3.5.1
    image: tinkerpop/gremlin-server:3.5.2
    ports:
      - 8182:8182

And the clients like this (quick and dirty with host network mode):

docker run -it --network=host tinkerpop/gremlin-console:3.5.1
docker run -it --network=host tinkerpop/gremlin-console:3.5.2

This part shows that you're not really using JanusGraph, but TinkerGraph which is an inmemory graph implementation from TinkerPop. It would be great if you could try out whether you can reproduce this also directly with Gremlin Server from TinkerPop. You can download that from here.

You are right. This is kind of confusing for me, because it seems like those TinkerGraphs were also persisted with the default store driver. At least, my data was still there after restarting the server.

Edit: It seems like the TinkerGraph implementation returned a standardjanusgraph by default. So it wasn't TinkerGraph in the first place, but JanusGraph alle the time?

gremlin> :remote connect tinkerpop.server conf/remote.yaml
==>Configured localhost/127.0.0.1:8182
gremlin> :> graph = TinkerGraph.open()
==>tinkergraph[vertices:0 edges:0]
gremlin> :> g = graph.traversal()
==>graphtraversalsource[standardjanusgraph[berkeleyje:/var/lib/janusgraph/data], standard]

[FlorianHockmann]

Oh, I overlooked one aspect of your description: You're not using a session. This means that the graph and g variables that you're creating are not used which also explains why you see the print graphtraversalsource[standardjanusgraph[berkeleyje: ....

You need to connect to the server in session mode if you want to use variables like this:

:remote connect tinkerpop.server conf/remote.yaml session

So it looks like the bug really involves JanusGraph.