[
https://issues.apache.org/jira/browse/TINKERPOP-1379?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jason Plurad updated TINKERPOP-1379:
------------------------------------
Description:
https://github.com/apache/tinkerpop/blob/master/gremlin-core/src/main/java/org/apache/tinkerpop/gremlin/process/traversal/step/filter/TailGlobalStep.java#L71-L74
This code doesn't account for the excess removed from {{tailBulk}}. This can
cause the code to set incorrect bulk values when there are multiple traversers
in the tail buffer.
I observed this behavior intermittently in TitanGraph
(https://github.com/thinkaurelius/titan/pull/1312), which doesn't allow user
defined ids, so the ordering of the traversers in the TraverserSet is a lot
more random than the unit tests using TinkerGraph.
This issue is reproducible with TinkerGraph (master/3.2.1). Instead of loading
from one of the data files, manually create the graph with the ids in inverted
order.
{noformat}
graph = TinkerGraph.open();
// graph.io(IoCore.gryo()).readGraph("tinkerpop-modern.kryo");
final Vertex v1 = graph.addVertex(T.id, 6L, T.label, "person", "name", "marko",
"age", 29);
final Vertex v2 = graph.addVertex(T.id, 5L, T.label, "person", "name", "vadas",
"age", 27);
final Vertex v3 = graph.addVertex(T.id, 4L, T.label, "software", "name", "lop",
"lang", "java");
final Vertex v4 = graph.addVertex(T.id, 3L, T.label, "person", "name", "josh",
"age", 32);
final Vertex v5 = graph.addVertex(T.id, 2L, T.label, "software", "name",
"ripple", "lang", "java");
final Vertex v6 = graph.addVertex(T.id, 1L, T.label, "person", "name", "peter",
"age", 35);
v1.addEdge("knows", v2, "weight", 0.5d);
v1.addEdge("knows", v4, "weight", 1.0d);
v1.addEdge("created", v2, "weight", 0.4d);
v4.addEdge("knows", v5, "weight", 1.0d);
v4.addEdge("knows", v3, "weight", 0.4d);
v6.addEdge("knows", v3, "weight", 0.2d);
if (graph.features().graph().supportsTransactions()) graph.tx().commit();
final GraphTraversalSource g = graph.traversal();
String result = g.V().repeat(both()).times(3).tail(7).count().next().toString();
boolean success = "7".equals(result);
{noformat}
The fix is this:
{noformat}
if (excess > 0) {
oldest.setBulk(oldestBulk-excess);
// Account for the loss of excess in the tail buffer
this.tailBulk -= excess;
}
{noformat}
was:
https://github.com/apache/tinkerpop/blob/master/gremlin-core/src/main/java/org/apache/tinkerpop/gremlin/process/traversal/step/filter/TailGlobalStep.java#L71-L74
This code doesn't account for the excess removed from {{tailBulk}}. This can
cause the code to set incorrect bulk values when there are multiple traversers
in the tail buffer.
I observed this behavior intermittently in TitanGraph, which doesn't allow user
defined ids, so the ordering of the traversers in the TraverserSet is a lot
more random than the unit tests using TinkerGraph.
This issue is reproducible with TinkerGraph (master/3.2.1). Instead of loading
from one of the data files, manually create the graph with the ids in inverted
order.
{noformat}
graph = TinkerGraph.open();
// graph.io(IoCore.gryo()).readGraph("tinkerpop-modern.kryo");
final Vertex v1 = graph.addVertex(T.id, 6L, T.label, "person", "name", "marko",
"age", 29);
final Vertex v2 = graph.addVertex(T.id, 5L, T.label, "person", "name", "vadas",
"age", 27);
final Vertex v3 = graph.addVertex(T.id, 4L, T.label, "software", "name", "lop",
"lang", "java");
final Vertex v4 = graph.addVertex(T.id, 3L, T.label, "person", "name", "josh",
"age", 32);
final Vertex v5 = graph.addVertex(T.id, 2L, T.label, "software", "name",
"ripple", "lang", "java");
final Vertex v6 = graph.addVertex(T.id, 1L, T.label, "person", "name", "peter",
"age", 35);
v1.addEdge("knows", v2, "weight", 0.5d);
v1.addEdge("knows", v4, "weight", 1.0d);
v1.addEdge("created", v2, "weight", 0.4d);
v4.addEdge("knows", v5, "weight", 1.0d);
v4.addEdge("knows", v3, "weight", 0.4d);
v6.addEdge("knows", v3, "weight", 0.2d);
if (graph.features().graph().supportsTransactions()) graph.tx().commit();
final GraphTraversalSource g = graph.traversal();
String result = g.V().repeat(both()).times(3).tail(7).count().next().toString();
boolean success = "7".equals(result);
{noformat}
The fix is this:
{noformat}
if (excess > 0) {
oldest.setBulk(oldestBulk-excess);
// Account for the loss of excess in the tail buffer
this.tailBulk -= excess;
}
{noformat}
> unaccounted excess in TailGlobalStep
> ------------------------------------
>
> Key: TINKERPOP-1379
> URL: https://issues.apache.org/jira/browse/TINKERPOP-1379
> Project: TinkerPop
> Issue Type: Bug
> Components: process
> Affects Versions: 3.2.1
> Reporter: Jason Plurad
>
> https://github.com/apache/tinkerpop/blob/master/gremlin-core/src/main/java/org/apache/tinkerpop/gremlin/process/traversal/step/filter/TailGlobalStep.java#L71-L74
> This code doesn't account for the excess removed from {{tailBulk}}. This can
> cause the code to set incorrect bulk values when there are multiple
> traversers in the tail buffer.
> I observed this behavior intermittently in TitanGraph
> (https://github.com/thinkaurelius/titan/pull/1312), which doesn't allow user
> defined ids, so the ordering of the traversers in the TraverserSet is a lot
> more random than the unit tests using TinkerGraph.
> This issue is reproducible with TinkerGraph (master/3.2.1). Instead of
> loading from one of the data files, manually create the graph with the ids in
> inverted order.
> {noformat}
> graph = TinkerGraph.open();
> // graph.io(IoCore.gryo()).readGraph("tinkerpop-modern.kryo");
> final Vertex v1 = graph.addVertex(T.id, 6L, T.label, "person", "name",
> "marko", "age", 29);
> final Vertex v2 = graph.addVertex(T.id, 5L, T.label, "person", "name",
> "vadas", "age", 27);
> final Vertex v3 = graph.addVertex(T.id, 4L, T.label, "software", "name",
> "lop", "lang", "java");
> final Vertex v4 = graph.addVertex(T.id, 3L, T.label, "person", "name",
> "josh", "age", 32);
> final Vertex v5 = graph.addVertex(T.id, 2L, T.label, "software", "name",
> "ripple", "lang", "java");
> final Vertex v6 = graph.addVertex(T.id, 1L, T.label, "person", "name",
> "peter", "age", 35);
> v1.addEdge("knows", v2, "weight", 0.5d);
> v1.addEdge("knows", v4, "weight", 1.0d);
> v1.addEdge("created", v2, "weight", 0.4d);
> v4.addEdge("knows", v5, "weight", 1.0d);
> v4.addEdge("knows", v3, "weight", 0.4d);
> v6.addEdge("knows", v3, "weight", 0.2d);
> if (graph.features().graph().supportsTransactions()) graph.tx().commit();
> final GraphTraversalSource g = graph.traversal();
> String result =
> g.V().repeat(both()).times(3).tail(7).count().next().toString();
> boolean success = "7".equals(result);
> {noformat}
> The fix is this:
> {noformat}
> if (excess > 0) {
> oldest.setBulk(oldestBulk-excess);
> // Account for the loss of excess in the tail buffer
> this.tailBulk -= excess;
> }
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
