[HADOOP-15343] NetworkTopology#getLoc should exit loop earlier rather than traverse all children

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 2.7.6
• Component/s: performance
• Labels: performance
• Assignee: Unassigned
• Reporter: Xiaoqiao He
• Created: 26/Mar/18 17:11
• Updated: 27/Mar/18 06:15
• Resolved: 27/Mar/18 04:53

Description

NetworkTopology#getLoc return a proper node after traverse ALL children of current InnerNode even if it has found expected result, based on branch-2.7. This issue may lead some performance loss especially for a large & busy cluster and many nodes under a rack. I think it should exit loop earlier rather than traverse all children of InnerNode.

    private Node getLoc(String loc) {
      if (loc == null || loc.length() == 0) return this;
            
      String[] path = loc.split(PATH_SEPARATOR_STR, 2);
      Node childnode = null;
      for(int i=0; i<children.size(); i++) {
        if (children.get(i).getName().equals(path[0])) {
          childnode = children.get(i);
        }
      }
      if (childnode == null) return null; // non-existing node
      if (path.length == 1) return childnode;
      if (childnode instanceof InnerNode) {
        return ((InnerNode)childnode).getLoc(path[1]);
      } else {
        return null;
      }
    }

Issue Links

• Is contained by: HADOOP-12185 NetworkTopology is not efficient adding/getting/removing nodes
• relates to: HADOOP-15345 Backport HADOOP-12185 to branch-2.7: NetworkTopology is not efficient adding/getting/removing nodes