Usage overview
Compute
In Materialize, clusters are pools of compute resources (CPU, memory, and scratch disk space) for running your workloads, such as maintaining up-to-date results while also providing strong consistency guarantees.
You must provision at least one cluster to power your workloads. You can then use the cluster to create the objects (indexes and materialized views) that provide always-fresh results. In Materialize, both indexes and materialized views are incrementally maintained when Materialize ingests new data. That is, Materialize performs work on writes such that no work is performed when reading from these objects.
The cluster size for a workload will depend on the workload’s compute and storage requirements.
Clusters are always “on”, and you can adjust the replication factor for fault tolerance. See Compute usage factors for more information on increasing a cluster’s replication factor.
Compute usage factors
Factors that contribute to compute usage include:
| Cost factor | Details |
|---|---|
| Replication factor for a cluster. | Each replica of the cluster provisions a new pool of compute resources to perform exactly the same computations on exactly the same data. |
| Indexes and materialized views | As data changes (insert/update/delete), indexes and materialized views perform incremental updates to provide up-to-date results. |
| Sources | • Sources that use upsert logic (i.e., ENVELOPE UPSERT or ENVELOPE DEBEZIUM Kafka sources) can lead to high memory and disk utilization.• Other sources consume a negligible amount of resources in steady state. |
SELECTs and SUBSCRIBEs |
• SELECTs and SUBSCRIBEs that do not use indexes and materialized views perform work. • SELECTs and SUBSCRIBEs that use indexes and materialized views are computationally free. |
| Sinks | Only small CPU/memory costs. |
Storage
In Materialize, storage is roughly proportional to the size of your source datasets plus the size of any materialized views, with some overhead from uncompacted data and system metrics.
Most data in Materialize is continually compacted, with the exception of append-only sources. As such, the total state stored in Materialize tends to grow at a rate that is more similar to OLTP databases than cloud data warehouses.