我正在从另一个问题中看一个相当复杂的MongoDB聚合查询，我创建了一大堆虚拟文档（200个）以对其进行操作，并在150-350ms之间的某个位置大致评估了性能。
然后我通过添加一些预过滤和后排序/限制等方式对该查询进行了稍微扩展。我希望这实际上可以加快查询速度，但是现在大约需要500-600毫秒。
我尝试创建几个索引（尽管我对Mongo大学的课程有些不了解）。一个在{ ownerId: 1, groupIds: 1 }上，另一个我在{ panels.visConfig.dataConfig.columns.element: 1 }上尝试，但两者似乎都没有什么不同，并且我看不到.explain()的输出有太大变化。
我想知道是否有人可以给我一些提高性能的指示，以及当我尝试的索引不起作用时，任何索引创建等究竟如何工作？
这是示例文档。在这里，有些字段更大一些，layout.sm，layout.md，layout.lg，meta和viewConfig。通常，我估计每个文档中会有20个panels用于测试。
{
"name" : "A Name",
"description" : "A Description",
"ownerId" : "00000000-0000-0000-0000-000000000000",
"groupIds": [
"11111111-1111-1111-1111-111111111111",
"22222222-2222-2222-2222-222222222222"
],
"version" : 1,
"coverId" : "5a8ae2180963493834d3049e",
"layout" : {
"sm" : [],
"md" : [],
"lg" : []
},
"hitCount" : 1,
"panels" : [
{
"_id" : ObjectId("000000000000000000000002"),
"visConfig" : {
"_id" : ObjectId("000000000000000000000001"),
"dataConfig" : {
"questionId" : 18,
"columns" : [
{
"type" : "COUNT",
"element" : "SE"
},
{
"levelFunction" : "MonthInYear",
"type" : "LEVEL",
"property" : "CRE",
"element" : "SE"
},
{
"type" : "LEVEL",
"property" : "OUT",
"element" : "SE"
}
]
},
"viewConfig" : {},
"meta" : {},
},
},
],
}
还有一个示例查询（旨在轻松在Robo 3T中运行）
var userId = "00000000-0000-0000-0000-000000000000";
var userGroups = [];
var mnemonics = ["AB", "XY"];
var sortAndPagingOptions = {
take: 10,
skip: 10,
sort: { name: 1 },
};
/** Find all the boards that the user can see */
var visibleBoards = {
$match: {
$or: [
{ ownerId: userId }, // Any where the user is the owner
{ groupIds: { $in: ["$this", userGroups ] } }, // Any which are shared with the users groups
],
},
};
/**
* This pipeline stage does 3 things
* 1) Calculate the number (totalElements) of columns in use within a dataConfig
* 2) Calculate the number (matchingElements) of columns that overlap with the given mnemonics within the dataConfig
* 3) Find the single panel that represents the cover visualization
*/
var calculateOverlapWithMnemonics = {
$project: {
_id: 1,
name: 1,
description: 1,
ownerId: 1,
coverId: 1,
coverVis: {
$filter: { // We're going to filter down to an array of 1 panel
input: "$panels", // Look at all the panels
as: "panel", // Refer to the current panel as $$panel
cond: { $eq: ["$$panel._id", "$coverId"] }, // Find the panel who's _id matches the coverId of the board
},
},
stats: {
$reduce: { // We're going to reduce all the panels
input: "$panels.visConfig.dataConfig.columns.element", // We use the columns.element field within a dataConfig which is going to yield a set of arrays => ["SE", "SE"] ["DX"] ["AB"]
initialValue: { totalElements: 0, matchingElements: 0 }, // We're going to calculate a total and a match value
in: {
totalElements: { $add: ["$$value.totalElements", { $size: "$$this" }] }, // Add together the sizes of the dataConfig element arrays
matchingElements: {
$add: [ // Add together the following elements (those which match mnemonics)
"$$value.matchingElements",
{
$sum: { // Sum the numeric results from the $map function
$map: { // Map each String element, into a 0 or 1 number
input: "$$this",
in: {
$cond: [{ $in: ["$$this", mnemonics] }, 1, 0], // Yield a 0 or 1 depending on whether the element exists within the array of mnemonics provided
},
},
},
},
],
},
},
},
},
},
};
/**
* This pipeline stage does 2 things
* 1) Converts the matchingElements to a percetange value
* 2) Finds the first panel (coverVis) and represents it as an object instead of an array
*/
var calculatePercentages = {
$project: {
_id: 1,
name: 1,
description: 1,
ownerId: 1,
coverVis: { $arrayElemAt: ["$coverVis", 0] },
percent: {
$multiply: [
{ $divide: ["$stats.matchingElements", "$stats.totalElements"] },
100,
],
},
}
};
/** Only match documents that are over the percent threshold */
var overThreshold = {
$match: { percent: { $gte: 25 } },
};
var sort = { $sort: sortAndPagingOptions.sort };
var limit = { $limit: sortAndPagingOptions.take };
var skip = { $skip: sortAndPagingOptions.skip };
// Create the aggregation pipeline
var pipeline = [
visibleBoards,
calculateOverlapWithMnemonics,
calculatePercentages,
overThreshold,
sort,
skip,
limit,
];
db.test.aggregate(pipeline);