GitHub user okram opened a pull request:
https://github.com/apache/tinkerpop/pull/449
TINKERPOP-1488: Make LazyBarrierStrategy part of the default TraversalStrategies
https://issues.apache.org/jira/browse/TINKERPOP-1488
`LazyBarrierStrategy` is now a default `OptimizationStrategy`. Rewrote `LazyBarrierStrategy` and with @dkuppitz's help, we were able to make all tested traversals either equal or faster than without `LazyBarrierStrategy`. This is a really cool optimization.
```
gremlin> graph = TinkerGraph.open()
==>tinkergraph[vertices:0 edges:0]
gremlin> graph.io(gryo()).readGraph('data/grateful-dead.kryo')
==>null
gremlin>
gremlin> g1 = graph.traversal().withoutStrategies(LazyBarrierStrategy.class)
==>graphtraversalsource[tinkergraph[vertices:808 edges:8049], standard]
gremlin> g2 = graph.traversal()
==>graphtraversalsource[tinkergraph[vertices:808 edges:8049], standard]
gremlin>
gremlin> clock(10)
{g1.V().out().in().out().count().iterate()}
==>1073.5150449999999
gremlin> clock(10)
{g2.V().out().in().out().count().iterate()}
==>8.8109061
gremlin>
gremlin> clock(100)
{g1.V().out().out().count().iterate()}
==>18.75418462
gremlin> clock(100)
{g2.V().out().out().count().iterate()}
==>2.04183101
gremlin>
gremlin> clock(100)
{g1.V().out().values('performances').count().iterate()}
==>1.67867484
gremlin> clock(100)
{g2.V().out().values('performances').count().iterate()}
==>1.4292696999999999
gremlin>
gremlin> clock(100)
{g1.V().out().values('performances').is(gt(10)).iterate()}
==>1.76701077
gremlin> clock(100)
{g2.V().out().values('performances').is(gt(10)).iterate()}
==>1.47417542
gremlin>
gremlin> clock(100)
{g1.V().has('name','Garcia').out().out().iterate()}
==>0.23742839999999998
gremlin> clock(100)
{g2.V().has('name','Garcia').out().out().iterate()}
==>0.24021971999999997
gremlin>
gremlin> clock(100)
{g1.V().as('a').out('sungBy').in('sungBy').where(neq('a')).iterate()}
==>26.31896202
gremlin> clock(100)
{g2.V().as('a').out('sungBy').in('sungBy').where(neq('a')).iterate()}
==>21.811683509999998
gremlin>
gremlin> clock(100)
{g1.V().has('name','DARK STAR').out('followedBy').aggregate('a').out('followedBy').in('followedBy').where(not(within('a'))).groupCount().by('name').iterate()}
==>24.6694559
gremlin> clock(100)
{g2.V().has('name','DARK STAR').out('followedBy').aggregate('a').out('followedBy').in('followedBy').where(not(within('a'))).groupCount().by('name').iterate()}
==>1.9376707899999999
```
If this is VOTE'd in then, I will merge it after TINKERPOP-1455 is merged so we can do parameterization via `Configuration`.
VOTE +1.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/apache/tinkerpop TINKERPOP-1488
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/tinkerpop/pull/449.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #449
commit 5f0e8438c92555b79e7387c10a9ce915b29c8840
Author: Marko A. Rodriguez <okrammarko@gmail.com>
Date: 2016-10-04T22:50:48Z
LazyBarrierStrategy is now a default strategy and is smart about labels, retractions, flatMaps, barriers, etc. In short, all the queries that were given to it are either equal in speed or faster.
commit 658630a9785952cada916377bb06283e9e85c6c4
Author: Marko A. Rodriguez <okrammarko@gmail.com>
Date: 2016-10-04T23:11:51Z
ProfileTests and LazyBArrierStrategy don't play well with each other because ProfileTests expect certain bulks/counts/etc. Thus, if testing and there is ProfileStep, then the strategy is not applied.
Improvement
Major
GitHub user okram opened a pull request:
https://github.com/apache/tinkerpop/pull/449
TINKERPOP-1488: Make LazyBarrierStrategy part of the default TraversalStrategieshttps://issues.apache.org/jira/browse/TINKERPOP-1488
`LazyBarrierStrategy` is now a default `OptimizationStrategy`. Rewrote `LazyBarrierStrategy` and with @dkuppitz's help, we were able to make all tested traversals either equal or faster than without `LazyBarrierStrategy`. This is a really cool optimization.
```
{g1.V().out().in().out().count().iterate()}gremlin> graph = TinkerGraph.open()
==>tinkergraph[vertices:0 edges:0]
gremlin> graph.io(gryo()).readGraph('data/grateful-dead.kryo')
==>null
gremlin>
gremlin> g1 = graph.traversal().withoutStrategies(LazyBarrierStrategy.class)
==>graphtraversalsource[tinkergraph[vertices:808 edges:8049], standard]
gremlin> g2 = graph.traversal()
==>graphtraversalsource[tinkergraph[vertices:808 edges:8049], standard]
gremlin>
gremlin> clock(10)
==>1073.5150449999999
{g2.V().out().in().out().count().iterate()}gremlin> clock(10)
==>8.8109061
{g1.V().out().out().count().iterate()}gremlin>
gremlin> clock(100)
==>18.75418462
{g2.V().out().out().count().iterate()}gremlin> clock(100)
==>2.04183101
{g1.V().out().values('performances').count().iterate()}gremlin>
gremlin> clock(100)
==>1.67867484
{g2.V().out().values('performances').count().iterate()}gremlin> clock(100)
==>1.4292696999999999
{g1.V().out().values('performances').is(gt(10)).iterate()}gremlin>
gremlin> clock(100)
==>1.76701077
{g2.V().out().values('performances').is(gt(10)).iterate()}gremlin> clock(100)
==>1.47417542
{g1.V().has('name','Garcia').out().out().iterate()}gremlin>
gremlin> clock(100)
==>0.23742839999999998
{g2.V().has('name','Garcia').out().out().iterate()}gremlin> clock(100)
==>0.24021971999999997
{g1.V().as('a').out('sungBy').in('sungBy').where(neq('a')).iterate()}gremlin>
gremlin> clock(100)
==>26.31896202
{g2.V().as('a').out('sungBy').in('sungBy').where(neq('a')).iterate()}gremlin> clock(100)
==>21.811683509999998
{g1.V().has('name','DARK STAR').out('followedBy').aggregate('a').out('followedBy').in('followedBy').where(not(within('a'))).groupCount().by('name').iterate()}gremlin>
gremlin> clock(100)
==>24.6694559
{g2.V().has('name','DARK STAR').out('followedBy').aggregate('a').out('followedBy').in('followedBy').where(not(within('a'))).groupCount().by('name').iterate()}gremlin> clock(100)
==>1.9376707899999999
```
If this is VOTE'd in then, I will merge it after
TINKERPOP-1455is merged so we can do parameterization via `Configuration`.VOTE +1.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/apache/tinkerpop
TINKERPOP-1488Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/tinkerpop/pull/449.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #449
commit 5f0e8438c92555b79e7387c10a9ce915b29c8840
Author: Marko A. Rodriguez <okrammarko@gmail.com>
Date: 2016-10-04T22:50:48Z
LazyBarrierStrategy is now a default strategy and is smart about labels, retractions, flatMaps, barriers, etc. In short, all the queries that were given to it are either equal in speed or faster.
commit 658630a9785952cada916377bb06283e9e85c6c4
Author: Marko A. Rodriguez <okrammarko@gmail.com>
Date: 2016-10-04T23:11:51Z
ProfileTests and LazyBArrierStrategy don't play well with each other because ProfileTests expect certain bulks/counts/etc. Thus, if testing and there is ProfileStep, then the strategy is not applied.