Range query with compound multikey index scans more objects than needed

XMLWordPrintableJSON

    • Type: Bug
    • Resolution: Won't Fix
    • Priority: Major - P3
    • None
    • Affects Version/s: 2.0.5, 2.0.6
    • Component/s: Index Maintenance, Querying
    • Environment:
      Windows, Linux
    • ALL
    • None
    • None
    • None
    • None
    • None
    • None
    • None

      I noticed very strange behavior IMO when range queries used with compound index where one of indexed fields is array.
      Example, I have collection of documents with HierarchyLevel as integer and Parents as array of strings, and other fieds. I make simple query:

      {"HierarchyLevel" : {"$gte" : 1, "$lte" : 2}}

      When I use simple key index or compound with simple type fields query optimizer runs as expected:

      {
        "cursor" : "BtreeCursor HierarchyLevel_1",
        "nscanned" : 153,
        "nscannedObjects" : 153,
        "n" : 153,
        "millis" : 1,
        "nYields" : 0,
        "nChunkSkips" : 0,
        "isMultiKey" : false,
        "indexOnly" : false,
        "indexBounds" : {
                "HierarchyLevel" : [
                        [
                                1,
                                2
                        ]
                ]
        }
}

      {
        "cursor" : "BtreeCursor HierarchyLevel_1_Path_1",
        "nscanned" : 153,
        "nscannedObjects" : 153,
        "n" : 153,
        "millis" : 1,
        "nYields" : 0,
        "nChunkSkips" : 0,
        "isMultiKey" : false,
        "indexOnly" : false,
        "indexBounds" : {
                "HierarchyLevel" : [
                        [
                                1,
                                2
                        ]
                ],
                "Path" : [
                        [
                                {
                                        "$minElement" : 1
                                },
                                {
                                        "$maxElement" : 1
                                }
                        ]
                ]
        }
}

      But when I use compound index with array field, $lte condition is simply ignored, and all index entries from lower bound is scanned

      {
        "cursor" : "BtreeCursor HierarchyLevel_1_Parents_1",
        "nscanned" : 283630,
        "nscannedObjects" : 283630,
        "n" : 153,
        "millis" : 836,
        "nYields" : 0,
        "nChunkSkips" : 0,
        "isMultiKey" : true,
        "indexOnly" : false,
        "indexBounds" : {
                "HierarchyLevel" : [
                        [
                                1,
                                1.7976931348623157e+308
                        ]
                ],
                "Parents" : [
                        [
                                {
                                        "$minElement" : 1
                                },
                                {
                                        "$maxElement" : 1
                                }
                        ]
                ]
        }
}

      In this case count of scanned objects is bigger that count of actual documents, and query took lot of time.

        1. example_documents.js
          3 kB
          Sergejs Degtjars

            Assignee:
            hari.khalsa@10gen.com
            Reporter:
            Sergejs Degtjars
            Votes:
            3 Vote for this issue
            Watchers:
            21 Start watching this issue

              Created:
              Updated:
              Resolved: