Best Practices for Time Series Collections

Omit Fields Containing Empty Objects and Arrays from Documents

If your data contains empty objects, arrays, or strings, omit the empty fields from your documents to optimize compression.

For example, consider the following documents:

{
   timestamp: ISODate("2020-01-23T00:00:00.441Z"),
   coordinates: [1.0, 2.0]
},
{
   timestamp: ISODate("2020-01-23T00:00:10.441Z"),
   coordinates: []
},
{
   timestamp: ISODate("2020-01-23T00:00:20.441Z"),
   coordinates: [3.0, 5.0]
}

coordinates fields with populated values and coordinates fields with an empty array result in a schema change for the compressor. The schema change causes the second and third documents in the sequence to remain uncompressed.

Optimize compression by omitting the fields with empty values, as shown in the following documents:

{
   timestamp: ISODate("2020-01-23T00:00:00.441Z"),
   coordinates: [1.0, 2.0]
},
{
   timestamp: ISODate("2020-01-23T00:00:10.441Z")
},
{
   timestamp: ISODate("2020-01-23T00:00:20.441Z"),
   coordinates: [3.0, 5.0]
}

Round Numeric Data to Few Decimal Places

Round numeric data to the precision that your application requires. Rounding numeric data to fewer decimal places improves the compression ratio.

Use Standard Embedded Fields Best Practices

If an embedded data model supports your application's needs, use nested fields. The behavior of nested fields does not differ between regular collections and time series collections.

For example, consider a time series collection that contains weather documents similar to the following example:

{
   timestamp: ISODate("2024-06-17T10:00:00Z"),
   stationId: "ALPHA123",
   atmosphere: {
      temperature: 21.5,
      humidity: 68,
      pressure: 1013.25
   },
   wind: {
      speed: 7.2,
      direction: "NW"
   },
   precipitation: 0.5,
   visibility: 10
}

MongoDB uses column compression on each nested field individually, which provides the same compression quality as flattening the fields to the top level. The resulting compression is identical to the compression that would result from flattening all of the nested fields to top level fields.

If your workload has high cardinality, flattening nested objects may improve performance. For example, the following document contains the same data as the previous document but in a flattened format:

{
   timestamp: ISODate("2024-06-17T10:00:00Z"),
   stationId: "ALPHA123",
   atmosphere_temperature: 21.5,
   atmosphere_humidity: 68,
   atmosphere_pressure: 1013.25
   wind_speed: 7.2,
   wind_direction: "NW"
   precipitation: 0.5,
   visibility: 10
}

Batch Document Writes

When inserting multiple documents:

• To avoid network roundtrips, use a single insertMany() statement as opposed to multiple insertOne() statements.

• If possible, insert data that contains identical metaField values in the same batches.

• Set the ordered parameter to false.

For example, if you have two sensors that correspond to two metaField values, sensor A and sensor B, a batch that contains multiple measurements from a single sensor incurs the cost of one insert, rather than one insert per measurement.

The following operation inserts six documents, but only incurs the cost of two inserts (one per metaField value), because the documents are ordered by sensor. The ordered parameter is set to false to improve performance:

db.temperatures.insertMany(
   [
      {
         metaField: {
            sensor: "sensorA"
         },
         timestamp: ISODate("2021-05-18T00:00:00.000Z"),
         temperature: 10
      },
      {
         metaField: {
            sensor: "sensorA"
         },
         timestamp: ISODate("2021-05-19T00:00:00.000Z"),
         temperature: 12
      },
      {
         metaField: {
            sensor: "sensorA"
         },
         timestamp: ISODate("2021-05-20T00:00:00.000Z"),
         temperature: 13
      },
      {
         metaField: {
            sensor: "sensorB"
         },
         timestamp: ISODate("2021-05-18T00:00:00.000Z"),
         temperature: 20
      },
      {
         metaField: {
            sensor: "sensorB"
         },
         timestamp: ISODate("2021-05-19T00:00:00.000Z"),
         temperature: 25
      },
      {
         metadField: {
            sensor: "sensorB"
         },
         timestamp: ISODate("2021-05-20T00:00:00.000Z"),
         temperature: 26
      }
   ],
   { "ordered": false }
)

Use Consistent Field Order in Documents

Using a consistent field order in your documents improves insert and compression performance.

For example, inserting the following documents, all of which have the same field order, results in optimal performance.

{
   _id: ObjectId("6250a0ef02a1877734a9df57"),
   timestamp: ISODate("2020-01-23T00:00:00.441Z"),
   name: "sensor1",
   range: 1
},
{
   _id: ObjectId("6560a0ef02a1877734a9df66"),
   timestamp: ISODate("2020-01-23T01:00:00.441Z"),
   name: "sensor1",
   range: 5
}

In contrast, the following documents do not achieve optimal performance, because their field orders differ:

{
   range: 1,
   _id: ObjectId("6250a0ef02a1877734a9df57"),
   name: "sensor1",
   timestamp: ISODate("2020-01-23T00:00:00.441Z")
},
{
   _id: ObjectId("6560a0ef02a1877734a9df66"),
   name: "sensor1",
   timestamp: ISODate("2020-01-23T01:00:00.441Z"),
   range: 5
}

Increase the Number of Clients

Increasing the number of clients that write data to your collections can improve performance.

Use the metaField as your Shard Key

Using the metaField to shard your collection provides sufficienct cardinality as a shard key for time series collections.

Set a Strategic metaField When Creating the Collection

Your choice of metaField has the biggest impact on optimizing queries in your application.

• Select fields that rarely or never change as part of your metaField.

• If possible, select identifiers or other stable values that are common in filter expressions as part of your metaField.

• Avoid selecting fields that are not used for filtering as part of your metaField. Instead, use those fields as measurements.

For more information, see metaField Considerations.

Set Appropriate Bucket Granularity

When you create a time series collection, MongoDB groups incoming time series data into buckets. By accurately setting granularity, you control how frequently data is bucketed based on the ingestion rate of your data.

Starting in MongoDB 6.3, you can use the custom bucketing parameters bucketMaxSpanSeconds and bucketRoundingSeconds to specify bucket boundaries and more precisely control how time series data is bucketed.

You can improve performance by setting the granularity or custom bucketing parameters to the best match for the time span between incoming measurements from the same data source. For example, if you are recording weather data from thousands of sensors but only record data from each sensor once per 5 minutes, you can either set granularity to "minutes" or set the custom bucketing parameters to 300 (seconds).

In this case, setting the granularity to hours groups up to a month's worth of data ingest events into a single bucket, resulting in longer traversal times and slower queries. Setting it to seconds leads to multiple buckets per polling interval, many of which might contain only a single document.

The following table shows the maximum time interval included in one bucket of data when using a given granularity value:

Create Secondary Indexes

To improve query performance, create one or more secondary indexes on your timeField and metaField to support common query patterns. In versions 6.3 and higher, MongoDB creates a secondary index on the timeField and metaField automatically.

Additional Index Best Practices

• Use the metaField index for filtering and equality.

• Use the timeField and other indexed fields for range queries.

• General indexing strategies also apply to time series collections. For more information, see Indexing Strategies.

Query the metaField on Sub-Fields

MongoDB reorders the metaField of time-series collections, which may cause servers to store data in a different field order than applications. If a metaField is an object, queries on the metaField may produce inconsistent results because metaField order may vary between servers and applications. To optimize queries on a time-series metaField, query the metaField on scalar sub-fields rather than the entire metaField.

The following example creates a time series collection:

db.weather.insertMany( [
   {
      metaField: { sensorId: 5578, type: "temperature" },
      timestamp: ISODate( "2021-05-18T00:00:00.000Z" ),
      temp: 12
   },
   {
      metaField: { sensorId: 5578, type: "temperature" },
      timestamp: ISODate( "2021-05-18T04:00:00.000Z" ),
      temp: 11
   }
] )

The following query on the sensorId and type scalar sub-fields returns the first document that matches the query criteria:

db.weather.findOne( {
   "metaField.sensorId": 5578,
   "metaField.type": "temperature"
} )

Example output:

{
  _id: ObjectId("6572371964eb5ad43054d572"),
  metaField: { sensorId: 5578, type: 'temperature' },
  timestamp: ISODate( "2021-05-18T00:00:00.000Z" ),
  temp: 12
}

Use $group Instead of Distinct()

Due to the unique data structure of time series collections, MongoDB can't efficiently index them for distinct values. Avoid using the distinct command or db.collection.distinct() helper method on time series collections. Instead, use a $group aggregation to group documents by distinct values, as shown in the following example:

db.foo.createIndex({"meta.project":1, "meta.type":1})
db.foo.aggregate([{$match: {"meta.project": 10}},
                  {$group: {_id: "$meta.type"}}])

This works as follows:

1. Creating a compound index on meta.project and meta.type and supports the aggregation.

2. The $match stage filters for documents where meta.project = 10.

3. The $group stage uses meta.type as the group key to output one document per unique value.