Database performance

In Fauna, each document update stores a new version of the document, which makes counter data poorly suited for database storage.

If a frequently-updated counter is an essential part of your application, an event sourcing technique is recommended to reduce database contention and reduce unnecessary database operations. For more information, see the Write throughput scaling tutorial.

If the event sourcing pattern isn’t suitable for your application, you might be able to make other performance improvements:

Fauna processes transactions in 10-millisecond batch intervals. Each batch can contain many transactions, but only one transaction per batch can modify a specific field, so throughput could be limited by contention on field updates.

Indexes must evaluate document history when searching for matching entries. Where possible, do not include frequently-updated fields in index terms or values definitions.

Where possible, set the collection history_days field to 0. Older version removal is handled by a background task so versions accumulate until the task executes. See Collections for additional information.

Consider setting the collection ttl_days to a small value. When the ttl_days limit is reached, "old" documents and their history are removed. See Collections for additional information.

Where possible, separate ephemeral data from long-term data. For queries that need to only access or change ephemeral data, including long-term data can make those queries take longer.

Avoid indexing fields that have low cardinality. This can cause many index entries to exist for each field value, which impacts general index performance.

Internally, indexes are partitioned by term, where each index term is serviced by one database node. With low cardinality terms, fewer nodes participate in generating sets of matching index entries. This makes index performance less scalable and can make those nodes hot spots that degrade overall performance.

Using set manipulation functions, such as Union, Intersection, and Difference can produce increased scatter/gather operations to populate the resulting set, reducing query performance. Set operations with fewer indexes perform better than those that have many indexes.

When you are not using temporality on documents in a collection, document versions increase your storage costs and reduce the performance of indexes covering those documents. Reduce the collection history_days setting to reduce the time interval for persisting document versions.

Additionally, to support temporality, indexes store index entries for every version of the documents associated with the index source definition. Indexes with many entries can impact overall index performance and increases storage requirements. To address this concern: