MongoDB Atlas Profiler slow queries each day on empty collections

MongoDB Atlas
, ,
1

If I visit MongoDB Atlas Profiler, I can observe on each and every day a similar pattern of slow queries. It looks like this:

image
image1703×665 67.3 KB

If ordered by Mean or Sum of operation execution time, the top 30 or more are collections with suffix _gtins_catalog that are empty, created when the user is creating an account.

Now, when I am investigating the queries for any of these collections they are usually several of them
and all of them have pipeline like this:

"originatingCommand": {
    "aggregate": "241e8df1-c055-424b-9e37-37012aa61f4e_gtins_catalog",
    "cursor": {
      "batchSize": 0
    },
    "pipeline": [
      {
        "$changeStream": {
          "startAfter": {
            "_data": "826470CC57000000012B0229296E04"
          },
          "fullDocument": "updateLookup",
          "showMigrationEvents": true
        }
      }
    ],
    "$db": "prod_sm",
...
}

And the Operation Execution Time can even reach up to 2 min.

And now is my question. I can’t figure out what cause this pipelines to be executed, especially on empty collections, and why they are this slow. Is this some internal operations that Mongo have to perform?
It would be very helpful If someone could cast a bit of light on this.

2

Hi @Grzegorz_Skupiewski

Thanks for being a part of MongoDB’s developer community! I’d like to take a closer look and see what may be causing this rogue command. At first glance, this does not look like an internal operation that MongoDB is running.

If you can click on one of the commands and investigate the query details, you should be able to see the “appName” that is calling this command. Would you mind checking on this and sharing what you find?

Thanks,
Frank

1 Like
3

Hi @Frank_Sun
Thank you for the reply, this is what you’ve asked for:

"appName": "mongot steady state"

Edit: Now as it came to my mind I probably should have mentioned, that even it these collections are empty, they have an Atlas Search Index created on them.

4

Hey @Grzegorz_Skupiewski - PM from Atlas Search here - can you share what cluster tier you are on?

5

Hi @Elle_Shwer, it is M30 (General).

6

I see, any chance you have a support package where you can submit a ticket about this?

7

Unfortunately, that is not an option for us. Any suspicions what might be a cause or hints where to look further?

8

Without much more info, the only interpretation I heard was: The query itself looks like the changestream mongot/AtlasSearch follows for steady state replication. It is possible for changestream queries to be slow on a collection without any writes due to writes on neighboring collections. But there is no precedent for this / it’s not something we’ve seen before. It’s hard to know beyond that.

9

Hi,
it seems we have the same problem here.

image
image926×710 29.5 KB

{
  "type": "command",
  "appName": "mongot steady state",
  "command": {
    "getMore": 4874570742714440000,
.....
}

These commands are all on empty collections with search index on them.

It seems more a bug than a problem in or data.

1 Like
10

Hello!
I started facing the same problem after I created an atlas index for my collection. unfortunately it affects the entire cluster performance.

Screenshot 2024-03-14 at 23.42.20
Screenshot 2024-03-14 at 23.42.20778×812 48.8 KB

1 Like
11

We’re having similar issues as well where almost all of our slow queries are related the app named “mongot steady state”.
It happens for all the collections where we’ve created Atlas search indexes

12

We are also on M30 General Tier. Facing same issue did anyone found any solution for this??