[jira] [Commented] (TINKERPOP-1254) Support dropping traverser path information when it is no longer needed.
[ https://issues.apache.org/jira/browse/TINKERPOP-1254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15370018#comment-15370018 ] ASF GitHub Bot commented on TINKERPOP-1254: --- Github user twilmes commented on the issue: https://github.com/apache/tinkerpop/pull/358 That's excellent! Integration tests almost ran all the way through for me, got an odd failure during the Archetype - Server, noted below. I'm going to rerun anyway with the latest commit and using docker this time. ``` [INFO] [INFO] Reactor Summary: [INFO] [INFO] Apache TinkerPop ... SUCCESS [ 9.540 s] [INFO] Apache TinkerPop :: Gremlin Shaded . SUCCESS [ 3.640 s] [INFO] Apache TinkerPop :: Gremlin Core ... SUCCESS [01:27 min] [INFO] Apache TinkerPop :: Gremlin Test ... SUCCESS [ 11.518 s] [INFO] Apache TinkerPop :: Gremlin Groovy . SUCCESS [ 49.133 s] [INFO] Apache TinkerPop :: Gremlin Groovy Test SUCCESS [ 8.280 s] [INFO] Apache TinkerPop :: TinkerGraph Gremlin SUCCESS [03:40 min] [INFO] Apache TinkerPop :: Gremlin Benchmark .. SUCCESS [ 4.660 s] [INFO] Apache TinkerPop :: Hadoop Gremlin . SUCCESS [06:14 min] [INFO] Apache TinkerPop :: Spark Gremlin .. SUCCESS [19:37 min] [INFO] Apache TinkerPop :: Giraph Gremlin . SUCCESS [ 02:54 h] [INFO] Apache TinkerPop :: Neo4j Gremlin .. SUCCESS [ 4.160 s] [INFO] Apache TinkerPop :: Gremlin Driver . SUCCESS [ 11.340 s] [INFO] Apache TinkerPop :: Gremlin Server . SUCCESS [12:58 min] [INFO] Apache TinkerPop :: Gremlin Console SUCCESS [01:36 min] [INFO] Apache TinkerPop :: Gremlin Archetype .. SUCCESS [ 0.151 s] [INFO] Apache TinkerPop :: Archetype - TinkerGraph FAILURE [ 0.374 s] [INFO] Apache TinkerPop :: Archetype - Server . SKIPPED [INFO] [INFO] BUILD FAILURE [INFO] [INFO] Total time: 03:41 h [INFO] Finished at: 2016-07-10T15:06:49-05:00 [INFO] Final Memory: 66M/727M [INFO] [ERROR] Failed to execute goal org.apache.maven.plugins:maven-surefire-plugin:2.17:test (default-test) on project gremlin-archetype-tinkergraph: Execution default-test of goal org.apache.maven.plugins:maven-surefire-plugin:2.17:test failed: There was an error in the forke d process [ERROR] java.lang.NoClassDefFoundError: projects/standard/project/gremlin-archetype-tinkergraph/target/test-classes/com/test/example/AppTest (wrong name: com/test/example/AppTest) [ERROR] at java.lang.ClassLoader.defineClass1(Native Method) [ERROR] at java.lang.ClassLoader.defineClass(ClassLoader.java:760) [ERROR] at java.security.SecureClassLoader.defineClass(SecureClassLoader.java:142) ``` > Support dropping traverser path information when it is no longer needed. > > > Key: TINKERPOP-1254 > URL: https://issues.apache.org/jira/browse/TINKERPOP-1254 > Project: TinkerPop > Issue Type: Improvement > Components: process >Affects Versions: 3.1.1-incubating >Reporter: Marko A. Rodriguez >Assignee: Ted Wilmes > > The most expensive traversals (especially in OLAP) are those that can not be > "bulked." There are various reasons why two traversers at the same object can > not be bulked, but the primary reason is {{PATH}} or {{LABELED_PATH}}. That > is, when the history of the traverser is required, the probability of two > traversers having the same history is low. > A key to making traversals more efficient is to do as a much as possible to > remove historic information from a traverser so it can get bulked. How does > one do this? > {code} > g.V.as('a').out().as('b').out().where(neq('a').and().neq('b')).both().name > {code} > The {{LABELED_PATH}} of "a" and "b" are required up to the {{where()}} and at > which point, at {{both()}}, they are no longer required. It would be smart to > support: > {code} > traverser.dropLabels(Set) > traverser.dropPath() > {code} > We would then, via a {{TraversalOptimizationStrategy}} insert a step between > {{where()}} and {{both()}} called {{PathPruneStep}} which would be a > {{SideEffectStep}}. The strategy would know which labels were no longer > needed (via forward lookahead) and then do: > {code} > public class
[jira] [Commented] (TINKERPOP-1254) Support dropping traverser path information when it is no longer needed.
[ https://issues.apache.org/jira/browse/TINKERPOP-1254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15369701#comment-15369701 ] ASF GitHub Bot commented on TINKERPOP-1254: --- Github user twilmes commented on the issue: https://github.com/apache/tinkerpop/pull/358 I made a small update to `ReferencePath` to create new label Sets when a patch is detached. This had been causing issues where the first set of labels for a path where being shared across `MutablePaths` after detachment. A label would be removed from one, and therefore all of the traversers that had that label `Set` in their path, would be affected. The `PathProcessors` are now respecting keepLabels null and labels are not dropped if `PrunePathStrategy` is not applied. **PrunePathStrategy on** ``` g.V().match( as("a").in("sungBy").as("b"), as("a").in("sungBy").as("c"), as("b").out("writtenBy").as("d"), as("c").out("writtenBy").as("e"), as("d").has("name", "George_Harrison"), as("e").has("name", "Bob_Marley")).select("a").count().profile() Traversal Metrics Step Count Traversers Time (ms)% Dur = GraphStep(vertex,[]) 808 808 44.21799.97 MatchStep(AND,[[MatchStartStep(a), ProfileStep,... 1 1 0.004 0.01 MatchStartStep(a) 808 808 43.517 VertexStep(IN,[sungBy],vertex) 501 499 20.323 MatchEndStep(b) (profiling ignored) 0.000 MatchStartStep(a)2 2 0.006 VertexStep(IN,[sungBy],vertex) 156 156 2.129 MatchEndStep(c) (profiling ignored) 0.000 MatchStartStep(b) 501 499 5.126 VertexStep(OUT,[writtenBy],vertex) 509 504 3.423 MatchEndStep(d) (profiling ignored) 0.000 MatchStartStep(c) 156 156 1.083 VertexStep(OUT,[writtenBy],vertex) 157 157 1.029 MatchEndStep(e) (profiling ignored) 0.000 MatchStartStep(d) 509 266 1.685 HasStep([name.eq(George_Harrison)]) 2 2 0.002 MatchEndStep (profiling ignored) 0.000 MatchStartStep(e) 157 57 0.391 HasStep([name.eq(Bob_Marley)]) 1 1 0.001 MatchEndStep (profiling ignored) 0.000 SelectOneStep(a) 1 1 0.003 0.01 CountGlobalStep1 1 0.003 0.01 >TOTAL - - 44.228- ``` **PrunePathStrategy off** ``` Traversal Metrics Step Count Traversers Time (ms)% Dur = GraphStep(vertex,[]) 808 808 7.56599.84 MatchStep(AND,[[MatchStartStep(a), ProfileStep,... 1 1 0.007 0.10 MatchStartStep(a) 808 808 5.726 VertexStep(IN,[sungBy],vertex) 501 499 9.532 MatchEndStep(b) (profiling ignored) 0.000 MatchStartStep(a)2 2 0.007 VertexStep(IN,[sungBy],vertex) 156 156 1.803
[jira] [Commented] (TINKERPOP-1254) Support dropping traverser path information when it is no longer needed.
[ https://issues.apache.org/jira/browse/TINKERPOP-1254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15369020#comment-15369020 ] ASF GitHub Bot commented on TINKERPOP-1254: --- Github user dkuppitz commented on the issue: https://github.com/apache/tinkerpop/pull/358 `docker/build.sh -t -i -n` fails with: ``` Failed tests: DedupTest$Traversals>DedupTest.g_V_both_name_order_byXa_bX_dedup_value:110 expected:<[josh]> but was:<[marko]> ``` > Support dropping traverser path information when it is no longer needed. > > > Key: TINKERPOP-1254 > URL: https://issues.apache.org/jira/browse/TINKERPOP-1254 > Project: TinkerPop > Issue Type: Improvement > Components: process >Affects Versions: 3.1.1-incubating >Reporter: Marko A. Rodriguez >Assignee: Ted Wilmes > > The most expensive traversals (especially in OLAP) are those that can not be > "bulked." There are various reasons why two traversers at the same object can > not be bulked, but the primary reason is {{PATH}} or {{LABELED_PATH}}. That > is, when the history of the traverser is required, the probability of two > traversers having the same history is low. > A key to making traversals more efficient is to do as a much as possible to > remove historic information from a traverser so it can get bulked. How does > one do this? > {code} > g.V.as('a').out().as('b').out().where(neq('a').and().neq('b')).both().name > {code} > The {{LABELED_PATH}} of "a" and "b" are required up to the {{where()}} and at > which point, at {{both()}}, they are no longer required. It would be smart to > support: > {code} > traverser.dropLabels(Set) > traverser.dropPath() > {code} > We would then, via a {{TraversalOptimizationStrategy}} insert a step between > {{where()}} and {{both()}} called {{PathPruneStep}} which would be a > {{SideEffectStep}}. The strategy would know which labels were no longer > needed (via forward lookahead) and then do: > {code} > public class PathPruneStep { > final Set dropLabels = ... > final boolean dropPath = ... > public void sideEffect(final Traverser traverser) { > final Traverser start = this.starts.next(); > if(this.dropPath) start.dropPath(); > else start.dropLabels(labels); > } > } > {code} > Again, the more we can prune historic path data no longer needed, the higher > the probability of bulking. Think about this in terms of {{match()}}. > {code} > g.V().match( > a.out.b, > b.out.c, > c.neq.a, > c.out.b, > ).select("a") > {code} > All we need is "a" at the end. Thus, once a pattern has been passed and no > future patterns require that label, drop it! > This idea is related to TINKERPOP-331, but I don't think we should deal with > manipulating the species. Thus, I think 331 is too "low level." -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (TINKERPOP-1254) Support dropping traverser path information when it is no longer needed.
[ https://issues.apache.org/jira/browse/TINKERPOP-1254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15367673#comment-15367673 ] ASF GitHub Bot commented on TINKERPOP-1254: --- GitHub user twilmes opened a pull request: https://github.com/apache/tinkerpop/pull/358 TINKERPOP-1254 Support dropping traverser path information when it is no longer needed This PR adds support for path retraction to increase the likelihood of bulking in OLTP and OLAP modes. Traversal analysis is performed during the application of the PrunePathStrategy to identify labels that may be dropped at various points in the traversal. MatchStep also performs runtime analysis to determine which labels it can drop in addition to the labels identified during traversal strategy application. Here is a set of profiles showing the benefit of path dropping. These were generated with `TinkerGraphComputer` first against 3.2 and then TinkerPop-1254. **Note** that the times should not be compared here. The first was run on my anemic macbook which, and the second was run on an 8 core AWS m3.2xlarge instance that I've been using for testing. I'll be following up with more numbers on the same hardware but you can see here the dramatic drop in traversers with path pruning enabled. **TP 3.2** ``` gremlin> g.V().match(__.as('a').out().as('b'), __.as('b').out().as('c'), __.as('c').out().as('d')).select('d').count().profile() ==>Traversal Metrics Step Count Traversers Time (ms)% Dur = TinkerGraphStep(vertex,[]) 808 808 21.205 0.02 MatchStep(AND,[[MatchStartStep(a), ProfileStep,... 14465066 14465066 110317.81385.88 MatchStartStep(a) 808 808 10975.797 VertexStep(OUT,vertex)8049 80496953.071 MatchEndStep(b) 8049 80496727.167 MatchStartStep(b) 8049 79575461.242 VertexStep(OUT,vertex) 327370 3273706782.024 MatchEndStep(c) 327370 3273706238.268 MatchStartStep(c) 327370 3269831771.773 VertexStep(OUT,vertex)14465066 14465066 11489.228 MatchEndStep(d) 14465066 14465066 14301.313 SelectOneStep(d)14465066 14465066 13752.96610.71 CountGlobalStep1 14363.667 3.40 >TOTAL - - 128455.652- ``` **TinkerPop-1254** ``` gremlin> g.V().match(__.as('a').out().as('b'), __.as('b').out().as('c'), __.as('c').out().as('d')).select('d').count().profile() ==>Traversal Metrics Step Count Traversers Time (ms)% Dur = GraphStep(vertex,[]) 808 808 32.45319.96 MatchStep(AND,[[MatchStartStep(a), ProfileStep,... 14465066 7510 89.46355.01 MatchStartStep(a) 808 808 22.388 VertexStep(OUT,vertex)8049 7957 85.493 MatchEndStep(b) (profiling ignored) 0.000 MatchStartStep(b) 8049 563 7.488 VertexStep(OUT,vertex) 327370 7561 19.548 MatchEndStep(c) (profiling ignored) 0.000 MatchStartStep(c) 327370 452 4.247 VertexStep(OUT,vertex)14465066 7510 14.812 MatchEndStep(d) (profiling ignored) 0.000 SelectOneStep(d)
[jira] [Commented] (TINKERPOP-1254) Support dropping traverser path information when it is no longer needed.
[ https://issues.apache.org/jira/browse/TINKERPOP-1254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15318584#comment-15318584 ] Ted Wilmes commented on TINKERPOP-1254: --- I'm getting close to ready for a review. I ended up switching over to keeping vs. dropping the labels. Drop works out well if a traversal is totally flat but becomes problematic when you're dealing with steps that take nested traversals. Currently I'm finishing up some fixes related to errors while running on graph computer, and then I'll push so we have something concrete to discuss. > Support dropping traverser path information when it is no longer needed. > > > Key: TINKERPOP-1254 > URL: https://issues.apache.org/jira/browse/TINKERPOP-1254 > Project: TinkerPop > Issue Type: Improvement > Components: process >Affects Versions: 3.1.1-incubating >Reporter: Marko A. Rodriguez >Assignee: Ted Wilmes > > The most expensive traversals (especially in OLAP) are those that can not be > "bulked." There are various reasons why two traversers at the same object can > not be bulked, but the primary reason is {{PATH}} or {{LABELED_PATH}}. That > is, when the history of the traverser is required, the probability of two > traversers having the same history is low. > A key to making traversals more efficient is to do as a much as possible to > remove historic information from a traverser so it can get bulked. How does > one do this? > {code} > g.V.as('a').out().as('b').out().where(neq('a').and().neq('b')).both().name > {code} > The {{LABELED_PATH}} of "a" and "b" are required up to the {{where()}} and at > which point, at {{both()}}, they are no longer required. It would be smart to > support: > {code} > traverser.dropLabels(Set) > traverser.dropPath() > {code} > We would then, via a {{TraversalOptimizationStrategy}} insert a step between > {{where()}} and {{both()}} called {{PathPruneStep}} which would be a > {{SideEffectStep}}. The strategy would know which labels were no longer > needed (via forward lookahead) and then do: > {code} > public class PathPruneStep { > final Set dropLabels = ... > final boolean dropPath = ... > public void sideEffect(final Traverser traverser) { > final Traverser start = this.starts.next(); > if(this.dropPath) start.dropPath(); > else start.dropLabels(labels); > } > } > {code} > Again, the more we can prune historic path data no longer needed, the higher > the probability of bulking. Think about this in terms of {{match()}}. > {code} > g.V().match( > a.out.b, > b.out.c, > c.neq.a, > c.out.b, > ).select("a") > {code} > All we need is "a" at the end. Thus, once a pattern has been passed and no > future patterns require that label, drop it! > This idea is related to TINKERPOP-331, but I don't think we should deal with > manipulating the species. Thus, I think 331 is too "low level." -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (TINKERPOP-1254) Support dropping traverser path information when it is no longer needed.
[ https://issues.apache.org/jira/browse/TINKERPOP-1254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15318434#comment-15318434 ] Marko A. Rodriguez commented on TINKERPOP-1254: --- How are things going on this work? [~twilmes] Doing okay? Need any advice/reviews? > Support dropping traverser path information when it is no longer needed. > > > Key: TINKERPOP-1254 > URL: https://issues.apache.org/jira/browse/TINKERPOP-1254 > Project: TinkerPop > Issue Type: Improvement > Components: process >Affects Versions: 3.1.1-incubating >Reporter: Marko A. Rodriguez >Assignee: Ted Wilmes > > The most expensive traversals (especially in OLAP) are those that can not be > "bulked." There are various reasons why two traversers at the same object can > not be bulked, but the primary reason is {{PATH}} or {{LABELED_PATH}}. That > is, when the history of the traverser is required, the probability of two > traversers having the same history is low. > A key to making traversals more efficient is to do as a much as possible to > remove historic information from a traverser so it can get bulked. How does > one do this? > {code} > g.V.as('a').out().as('b').out().where(neq('a').and().neq('b')).both().name > {code} > The {{LABELED_PATH}} of "a" and "b" are required up to the {{where()}} and at > which point, at {{both()}}, they are no longer required. It would be smart to > support: > {code} > traverser.dropLabels(Set) > traverser.dropPath() > {code} > We would then, via a {{TraversalOptimizationStrategy}} insert a step between > {{where()}} and {{both()}} called {{PathPruneStep}} which would be a > {{SideEffectStep}}. The strategy would know which labels were no longer > needed (via forward lookahead) and then do: > {code} > public class PathPruneStep { > final Set dropLabels = ... > final boolean dropPath = ... > public void sideEffect(final Traverser traverser) { > final Traverser start = this.starts.next(); > if(this.dropPath) start.dropPath(); > else start.dropLabels(labels); > } > } > {code} > Again, the more we can prune historic path data no longer needed, the higher > the probability of bulking. Think about this in terms of {{match()}}. > {code} > g.V().match( > a.out.b, > b.out.c, > c.neq.a, > c.out.b, > ).select("a") > {code} > All we need is "a" at the end. Thus, once a pattern has been passed and no > future patterns require that label, drop it! > This idea is related to TINKERPOP-331, but I don't think we should deal with > manipulating the species. Thus, I think 331 is too "low level." -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (TINKERPOP-1254) Support dropping traverser path information when it is no longer needed.
[ https://issues.apache.org/jira/browse/TINKERPOP-1254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15308427#comment-15308427 ] Marko A. Rodriguez commented on TINKERPOP-1254: --- Hm. I would put the {{dropLabels}} logic into {{MatchStep}} itself. For each traverser, you can {{Traverser.getTags()}} to know which traversals it has already gone down. If the travesals it has NOT gone down don't need the labels of the tagged traversals, then drop them. That is, don't insert {{dropLabels()}}-steps, do the logic in {{MatchStep.computerAlgorithm()}} and {{MatchStep.standardAlgorithm()}}. > Support dropping traverser path information when it is no longer needed. > > > Key: TINKERPOP-1254 > URL: https://issues.apache.org/jira/browse/TINKERPOP-1254 > Project: TinkerPop > Issue Type: Improvement > Components: process >Affects Versions: 3.1.1-incubating >Reporter: Marko A. Rodriguez >Assignee: Ted Wilmes > > The most expensive traversals (especially in OLAP) are those that can not be > "bulked." There are various reasons why two traversers at the same object can > not be bulked, but the primary reason is {{PATH}} or {{LABELED_PATH}}. That > is, when the history of the traverser is required, the probability of two > traversers having the same history is low. > A key to making traversals more efficient is to do as a much as possible to > remove historic information from a traverser so it can get bulked. How does > one do this? > {code} > g.V.as('a').out().as('b').out().where(neq('a').and().neq('b')).both().name > {code} > The {{LABELED_PATH}} of "a" and "b" are required up to the {{where()}} and at > which point, at {{both()}}, they are no longer required. It would be smart to > support: > {code} > traverser.dropLabels(Set) > traverser.dropPath() > {code} > We would then, via a {{TraversalOptimizationStrategy}} insert a step between > {{where()}} and {{both()}} called {{PathPruneStep}} which would be a > {{SideEffectStep}}. The strategy would know which labels were no longer > needed (via forward lookahead) and then do: > {code} > public class PathPruneStep { > final Set dropLabels = ... > final boolean dropPath = ... > public void sideEffect(final Traverser traverser) { > final Traverser start = this.starts.next(); > if(this.dropPath) start.dropPath(); > else start.dropLabels(labels); > } > } > {code} > Again, the more we can prune historic path data no longer needed, the higher > the probability of bulking. Think about this in terms of {{match()}}. > {code} > g.V().match( > a.out.b, > b.out.c, > c.neq.a, > c.out.b, > ).select("a") > {code} > All we need is "a" at the end. Thus, once a pattern has been passed and no > future patterns require that label, drop it! > This idea is related to TINKERPOP-331, but I don't think we should deal with > manipulating the species. Thus, I think 331 is too "low level." -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (TINKERPOP-1254) Support dropping traverser path information when it is no longer needed.
[ https://issues.apache.org/jira/browse/TINKERPOP-1254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15308248#comment-15308248 ] Ted Wilmes commented on TINKERPOP-1254: --- Not sure if this makes sense, but I'm in the thick of this right now and taking an approach where I build a dependency tree for the match traversals. Contrived example here to illustrate: {code} match( __.as('a').out().as('b'), (1) __.as('b').out().as('c'), (2) __.as('c').out().as('d'), (3) __.as('b').out().as('e')(4) ).select('d', 'e').dropLabels('d', 'e') {code} Traversal (2) and (4) depend on (1), Traversal (3) depends on (2). The results of (3) and (4) are referenced in the {{select}}. This info is used to insert the following dropLabel steps: {code} match( __.as('a').out().as('b'), __.as('b').dropLabels('b').out().as('c'), __.as('c').dropLabels('c').out().as('d') __.as('b').dropLabels('b').out().as('e') ).select('d', 'e').dropLabels('d', 'e') {code} My thinking is even though the order execution is determined at runtime, we can figure out beforehand which match traversals depend on each other and insert the drop steps accordingly. Does this make sense or have I made some incorrect assumptions on dependencies between match traversals? > Support dropping traverser path information when it is no longer needed. > > > Key: TINKERPOP-1254 > URL: https://issues.apache.org/jira/browse/TINKERPOP-1254 > Project: TinkerPop > Issue Type: Improvement > Components: process >Affects Versions: 3.1.1-incubating >Reporter: Marko A. Rodriguez >Assignee: Ted Wilmes > > The most expensive traversals (especially in OLAP) are those that can not be > "bulked." There are various reasons why two traversers at the same object can > not be bulked, but the primary reason is {{PATH}} or {{LABELED_PATH}}. That > is, when the history of the traverser is required, the probability of two > traversers having the same history is low. > A key to making traversals more efficient is to do as a much as possible to > remove historic information from a traverser so it can get bulked. How does > one do this? > {code} > g.V.as('a').out().as('b').out().where(neq('a').and().neq('b')).both().name > {code} > The {{LABELED_PATH}} of "a" and "b" are required up to the {{where()}} and at > which point, at {{both()}}, they are no longer required. It would be smart to > support: > {code} > traverser.dropLabels(Set) > traverser.dropPath() > {code} > We would then, via a {{TraversalOptimizationStrategy}} insert a step between > {{where()}} and {{both()}} called {{PathPruneStep}} which would be a > {{SideEffectStep}}. The strategy would know which labels were no longer > needed (via forward lookahead) and then do: > {code} > public class PathPruneStep { > final Set dropLabels = ... > final boolean dropPath = ... > public void sideEffect(final Traverser traverser) { > final Traverser start = this.starts.next(); > if(this.dropPath) start.dropPath(); > else start.dropLabels(labels); > } > } > {code} > Again, the more we can prune historic path data no longer needed, the higher > the probability of bulking. Think about this in terms of {{match()}}. > {code} > g.V().match( > a.out.b, > b.out.c, > c.neq.a, > c.out.b, > ).select("a") > {code} > All we need is "a" at the end. Thus, once a pattern has been passed and no > future patterns require that label, drop it! > This idea is related to TINKERPOP-331, but I don't think we should deal with > manipulating the species. Thus, I think 331 is too "low level." -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (TINKERPOP-1254) Support dropping traverser path information when it is no longer needed.
[ https://issues.apache.org/jira/browse/TINKERPOP-1254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15307850#comment-15307850 ] Marko A. Rodriguez commented on TINKERPOP-1254: --- Excellent! I'm excited to review the PR and to see this feature in action. One thing, given that {{MatchStep}} will have pattern order execution determined dynamically at runtime, are you being smart about knowing when to drop and not drop with {{match()}}? > Support dropping traverser path information when it is no longer needed. > > > Key: TINKERPOP-1254 > URL: https://issues.apache.org/jira/browse/TINKERPOP-1254 > Project: TinkerPop > Issue Type: Improvement > Components: process >Affects Versions: 3.1.1-incubating >Reporter: Marko A. Rodriguez >Assignee: Ted Wilmes > > The most expensive traversals (especially in OLAP) are those that can not be > "bulked." There are various reasons why two traversers at the same object can > not be bulked, but the primary reason is {{PATH}} or {{LABELED_PATH}}. That > is, when the history of the traverser is required, the probability of two > traversers having the same history is low. > A key to making traversals more efficient is to do as a much as possible to > remove historic information from a traverser so it can get bulked. How does > one do this? > {code} > g.V.as('a').out().as('b').out().where(neq('a').and().neq('b')).both().name > {code} > The {{LABELED_PATH}} of "a" and "b" are required up to the {{where()}} and at > which point, at {{both()}}, they are no longer required. It would be smart to > support: > {code} > traverser.dropLabels(Set) > traverser.dropPath() > {code} > We would then, via a {{TraversalOptimizationStrategy}} insert a step between > {{where()}} and {{both()}} called {{PathPruneStep}} which would be a > {{SideEffectStep}}. The strategy would know which labels were no longer > needed (via forward lookahead) and then do: > {code} > public class PathPruneStep { > final Set dropLabels = ... > final boolean dropPath = ... > public void sideEffect(final Traverser traverser) { > final Traverser start = this.starts.next(); > if(this.dropPath) start.dropPath(); > else start.dropLabels(labels); > } > } > {code} > Again, the more we can prune historic path data no longer needed, the higher > the probability of bulking. Think about this in terms of {{match()}}. > {code} > g.V().match( > a.out.b, > b.out.c, > c.neq.a, > c.out.b, > ).select("a") > {code} > All we need is "a" at the end. Thus, once a pattern has been passed and no > future patterns require that label, drop it! > This idea is related to TINKERPOP-331, but I don't think we should deal with > manipulating the species. Thus, I think 331 is too "low level." -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (TINKERPOP-1254) Support dropping traverser path information when it is no longer needed.
[ https://issues.apache.org/jira/browse/TINKERPOP-1254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15307845#comment-15307845 ] Ted Wilmes commented on TINKERPOP-1254: --- I'm pretty close to PR-ready. Hopefully have something pushed later this week and ready for an in-depth review. I need to make a few final updates to the strategy and I also need to do some further testing to make sure my retract on the {{ImmutablePath}} doesn't lead to performance degradation as the number of Immutable paths temporarily increases. > Support dropping traverser path information when it is no longer needed. > > > Key: TINKERPOP-1254 > URL: https://issues.apache.org/jira/browse/TINKERPOP-1254 > Project: TinkerPop > Issue Type: Improvement > Components: process >Affects Versions: 3.1.1-incubating >Reporter: Marko A. Rodriguez >Assignee: Ted Wilmes > > The most expensive traversals (especially in OLAP) are those that can not be > "bulked." There are various reasons why two traversers at the same object can > not be bulked, but the primary reason is {{PATH}} or {{LABELED_PATH}}. That > is, when the history of the traverser is required, the probability of two > traversers having the same history is low. > A key to making traversals more efficient is to do as a much as possible to > remove historic information from a traverser so it can get bulked. How does > one do this? > {code} > g.V.as('a').out().as('b').out().where(neq('a').and().neq('b')).both().name > {code} > The {{LABELED_PATH}} of "a" and "b" are required up to the {{where()}} and at > which point, at {{both()}}, they are no longer required. It would be smart to > support: > {code} > traverser.dropLabels(Set) > traverser.dropPath() > {code} > We would then, via a {{TraversalOptimizationStrategy}} insert a step between > {{where()}} and {{both()}} called {{PathPruneStep}} which would be a > {{SideEffectStep}}. The strategy would know which labels were no longer > needed (via forward lookahead) and then do: > {code} > public class PathPruneStep { > final Set dropLabels = ... > final boolean dropPath = ... > public void sideEffect(final Traverser traverser) { > final Traverser start = this.starts.next(); > if(this.dropPath) start.dropPath(); > else start.dropLabels(labels); > } > } > {code} > Again, the more we can prune historic path data no longer needed, the higher > the probability of bulking. Think about this in terms of {{match()}}. > {code} > g.V().match( > a.out.b, > b.out.c, > c.neq.a, > c.out.b, > ).select("a") > {code} > All we need is "a" at the end. Thus, once a pattern has been passed and no > future patterns require that label, drop it! > This idea is related to TINKERPOP-331, but I don't think we should deal with > manipulating the species. Thus, I think 331 is too "low level." -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (TINKERPOP-1254) Support dropping traverser path information when it is no longer needed.
[ https://issues.apache.org/jira/browse/TINKERPOP-1254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15307793#comment-15307793 ] Marko A. Rodriguez commented on TINKERPOP-1254: --- How is this going [~twilmes]? This is getting more and more important -- especially as more people are using {{match()}}. > Support dropping traverser path information when it is no longer needed. > > > Key: TINKERPOP-1254 > URL: https://issues.apache.org/jira/browse/TINKERPOP-1254 > Project: TinkerPop > Issue Type: Improvement > Components: process >Affects Versions: 3.1.1-incubating >Reporter: Marko A. Rodriguez >Assignee: Ted Wilmes > > The most expensive traversals (especially in OLAP) are those that can not be > "bulked." There are various reasons why two traversers at the same object can > not be bulked, but the primary reason is {{PATH}} or {{LABELED_PATH}}. That > is, when the history of the traverser is required, the probability of two > traversers having the same history is low. > A key to making traversals more efficient is to do as a much as possible to > remove historic information from a traverser so it can get bulked. How does > one do this? > {code} > g.V.as('a').out().as('b').out().where(neq('a').and().neq('b')).both().name > {code} > The {{LABELED_PATH}} of "a" and "b" are required up to the {{where()}} and at > which point, at {{both()}}, they are no longer required. It would be smart to > support: > {code} > traverser.dropLabels(Set) > traverser.dropPath() > {code} > We would then, via a {{TraversalOptimizationStrategy}} insert a step between > {{where()}} and {{both()}} called {{PathPruneStep}} which would be a > {{SideEffectStep}}. The strategy would know which labels were no longer > needed (via forward lookahead) and then do: > {code} > public class PathPruneStep { > final Set dropLabels = ... > final boolean dropPath = ... > public void sideEffect(final Traverser traverser) { > final Traverser start = this.starts.next(); > if(this.dropPath) start.dropPath(); > else start.dropLabels(labels); > } > } > {code} > Again, the more we can prune historic path data no longer needed, the higher > the probability of bulking. Think about this in terms of {{match()}}. > {code} > g.V().match( > a.out.b, > b.out.c, > c.neq.a, > c.out.b, > ).select("a") > {code} > All we need is "a" at the end. Thus, once a pattern has been passed and no > future patterns require that label, drop it! > This idea is related to TINKERPOP-331, but I don't think we should deal with > manipulating the species. Thus, I think 331 is too "low level." -- This message was sent by Atlassian JIRA (v6.3.4#6332)