Improving Query Performance for Conditional Summation in groupby

Kartik_Paul · 2024-08-08T05:41:42.124Z

I am running the following query on a dataset with one billion records. The query becomes significantly slower when I add a sort by userId. The execution time increases drastically. Can anyone guide me on how to optimize this query for better performance?

[  
    {
        "$group": {
            "_id": "$userId",
            "totalMonthly": { $sum: { $cond: [{ "$eq": ["$subscriptionType", "month"] }, 1, 0] } },
            "subscriptionAmt": { $sum: "$subscriptionAmt" },
            "subscriptionId": {
                "$first": "$subscriptionId"
            }
        }
    },
    {
        "$sort": { "totalMonthly": -1 }
    },
    {
        "$limit": 5
    }
]

steevej · 2024-08-09T14:00:21.059Z

Kartik_Paul:

The query becomes significantly slower when I add a sort by userId

Why do you $sort then?

If you must sort make sure you have the appropriate index.

Kartik_Paul · 2024-08-09T14:35:18.307Z

We have set the index and used sorting, but it has not improved the grouping performance, and it’s still taking a long time. Do you have any suggestions for making this pipeline faster?

steevej · 2024-08-13T18:35:14.865Z

What works sometimes is to $group with _id only and do any computation in a $lookup pipeline.

Something like:

{ $group : {
    _id : "$userId"
} } ,
{ $lookup : {
    from : "collection_name" ,
    localField : "_id" ,
    foreignField : "userId" ,
    as : "result" ,
    pipeline : [
        { $group : {
            _id : null ,
            totalMonthly : ... ,
            ... ,
            subscriptionId : ...
        } }
    ] 
} }

The theory, is that the first empty $group uses a lot less memory. While the computations are the same in the inner $group it is performed on a smaller set of documents and can be sent to the next stage faster.

But sorting on a computed value like totalMonthly will always be kind of slow since no index can be used. Sometimes the computed pattern is your best friend.

steevej · 2024-08-20T12:38:59.071Z

Hi @Kartik_Paul, any followup on this last proposal?

Kartik_Paul · 2024-08-20T14:14:05.946Z

We reviewed your last solution, but it takes longer than the existing query.

steevej · 2024-08-20T14:18:06.358Z

Thanks for the feedback.

Topic	Replies	Views	Activity
Time-series modelling: storage limitations Working with Data data-modeling time-series	0	419	May 2024
$out and $merge with causal consistency Working with Data aggregation replication views	0	187	Aug 2024
Darshan Hiranandani : Has anyone tried disabling writeConcern acknowledgment to enhance performance? Working with Data	1	67	Aug 2024
Error – Option “explain” cannot be used on an aggregate call with writeConcern Working with Data aggregation queries atlas-cluster	3	370	Sep 2024
How should i handle daily updates of thousands of records while now knowing which ones actually changes? Working with Data	0	26	Feb 9

Improving Query Performance for Conditional Summation in groupby

New & Unread Topics

Want to read more? Browse other topics in Working with Data or view latest topics.