[ https://issues.apache.org/jira/browse/GEARPUMP-349?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16166167#comment-16166167 ]
ASF GitHub Bot commented on GEARPUMP-349: ----------------------------------------- Github user manuzhang commented on a diff in the pull request: https://github.com/apache/incubator-gearpump/pull/223#discussion_r138876832 --- Diff: core/src/main/scala/org/apache/gearpump/util/Graph.scala --- @@ -243,13 +259,34 @@ class Graph[N, E](vertexList: List[N], edgeList: List[(N, E, N)]) extends Serial * The node returned by Iterator is stable sorted. */ def topologicalOrderIterator: Iterator[N] = { - val newGraph = copy - var output = List.empty[N] - - while (!newGraph.isEmpty) { - output ++= newGraph.removeZeroInDegree + tryTopologicalOrderIterator.get + } + + private def tryTopologicalOrderIterator: Try[Iterator[N]] = { + Try { + val indegreeMap = mutable.Map.empty[N, Int] --- End diff -- why not generate `indegreeMap` with `vertices.map` ? > Graph#topologicalOrderIterator is slow for large graph > ------------------------------------------------------ > > Key: GEARPUMP-349 > URL: https://issues.apache.org/jira/browse/GEARPUMP-349 > Project: Apache Gearpump > Issue Type: Improvement > Components: core > Affects Versions: 0.8.4 > Reporter: Manu Zhang > Assignee: Huafeng Wang > Fix For: 0.8.5 > > > The algorithm is as follows > 1. find zero in-degree nodes from a copied graph. > 2. remove nodes from the copied graph and add them to the output > 3. repeat 1 > The issue is that step 1 traverses all remaining nodes each time, which costs > the algorithm {{O(n^2)}} time > {{Graph#hasCycle}} has a similar issue -- This message was sent by Atlassian JIRA (v6.4.14#64029)