시계열 컬렉션 만들기

MongoDB 5.0 introduced native time series collections — a specialised collection type optimised for storing and querying measurements that change over time. Common use cases include IoT sensor readings, application metrics, financial ticks, and server monitoring data. Unlike regular collections, time series collections use a columnar storage format internally, dramatically reducing storage space and improving query performance on time-range filters.

Every time series collection requires three key options when created. The timeField is the document field that holds the timestamp (must be a BSON Date). The metaField identifies the series — for example, a sensor ID or device name. The granularity hint ('seconds', 'minutes', or 'hours') tells MongoDB how frequently measurements arrive, allowing it to optimise bucket sizing internally.

Use db.createCollection() with a timeseries option object to create a time series collection. You cannot convert an existing regular collection to time series — you must create it fresh. The collection will appear in show collections with a special timeseries type indicator.

The granularity option controls how MongoDB groups measurements into internal bucket documents. With 'seconds', buckets span one hour (3,600 measurements per bucket). With 'minutes', buckets span 24 hours. With 'hours', buckets span 30 days. Choosing the wrong granularity means more bucket documents and worse compression — always match granularity to your actual data arrival rate.

Documents inserted into a time series collection must include the timeField as a proper BSON Date. The metaField value identifies which series this measurement belongs to (e.g., a device ID). All other fields are called measurement fields and can hold any BSON value. MongoDB will reject documents where the timeField is missing or not a Date.

Insert into a time series collection exactly as you would a regular collection — using insertOne() or insertMany(). MongoDB handles the internal bucketing automatically. It is best practice to batch inserts with insertMany() when loading historical data, as this amortises the overhead of bucket creation across many measurements.

Time series collections support automatic data expiration via the expireAfterSeconds option. Once set, MongoDB's background TTL thread deletes entire buckets when all measurements in the bucket are older than the threshold. This is more efficient than a regular TTL index because entire internal bucket documents are dropped at once rather than individual measurement documents.

Queries on time series collections look identical to regular find() queries. MongoDB automatically uses the internal bucket structure to skip irrelevant buckets when filtering by time range. Filtering on the metaField is also highly efficient. Avoid querying only on measurement fields without a time or meta filter, as this forces a full collection scan across all buckets.

You can increase the granularity of an existing time series collection (e.g., from 'seconds' to 'minutes') using the collMod command. However, you cannot decrease it — MongoDB will return an error if you try to move from 'minutes' back to 'seconds'. Updating the expireAfterSeconds setting is also possible via collMod without recreating the collection.

Time series collections have a few important restrictions compared to regular collections. You cannot shard a time series collection on the timeField alone — a hashed metaField component is required. Updates and deletes are limited: before MongoDB 5.1, only deletes by metaField or time range were supported. Additionally, time series collections do not support unique indexes, sparse indexes, or capped collections.

After creating a time series collection, inspect it using db.getCollectionInfos() to confirm the timeseries options are correctly stored. You can also run db.sensorReadings.stats() to see storage statistics, including the number of internal bucket documents MongoDB is maintaining behind the scenes.

Test your understanding of MongoDB & NoSQL Databases concepts from this lesson.

In this lesson you learned: time series collections use a columnar bucket format for high-compression append-heavy workloads, three key options (timeField, metaField, granularity) control how measurements are organised and bucketed, and expireAfterSeconds enables efficient automatic purging of old data at the bucket level. Next up we explore inserting and querying time series data in depth.