Billing Dimensions

Elasticsearch Service Billing is based on your actual usage across a number of dimensions, as follows:

Read on for detail about each of these billing dimensions.

Deployment capacity refers to the cost of the nodes in your Elasticsearch deployment, plus additional node types such as Kibana, APM, and ML. Each node type is priced in terms of GB of RAM per hour (CPU and disk are scaled with RAM and included in this price). To calculate deployment capacity costs, we total up the cost of the nodes in your deployment(s) and multiply by GBs of RAM and how long they’ve been running.

Deployment capacity typically constitutes the majority of your bill, and is the easiest to understand and control.

Deployment capacity is purely a function of your current deployment configuration and time. To reduce this cost, you must configure your deployment to use fewer resources. To determine how much a particular deployment configuration will cost, try our Elasticsearch Service Pricing Calculator.

Data Transfer accounts for the volume of data (payload) going into, out of, and between the nodes in a deployment, which is summed up to a cumulative amount within a billing cycle.

We meter and bill data transfer using three dimensions:

We provide a free allowance of 100GB per month, which includes the sum of data out and data inter-node, across all deployments in the account. Once this threshold is passed, a charge is applied for any data transfer used in excess of the 100GB monthly free allowance.

Data transfer out of deployments and between nodes of the cluster is hard to control, as it is a function of the use case employed for the cluster and cannot always be tuned. Use cases such as batch queries executed at a frequent interval may be revisited to help lower transfer costs, if applicable. Watcher email alerts also count towards data transfer out of the deployment, so you may want to reduce their frequency and size.

The largest contributor to inter-node data transfer is usually shard movement between nodes in a cluster. The only way to prevent shard movement is by having a single node in a single availability zone. This solution is only possible for clusters up to 64GB RAM and is not recommended as it creates a risk of data loss. Oversharding can cause excessive shard movement. Avoiding oversharding can also help control costs and improve performance. Note that creating snapshots generates inter-node data transfer. The storage cost of snapshots is detailed later in this document.

The exact root cause of unusual data transfer is not always something we can identify as it can have many causes, some of which are out of our control and not associated with Cloud configuration changes. It may help to enable monitoring and examine index and shard activity on your cluster.

Storage costs are tied to the cost of storing the backup snapshots in the underlying IaaS object store, such as AWS S3, Google Cloud GCS or Azure Storage. These storage costs are not for the disk storage that persists the Elasticsearch indices, as that is already included in the RAM Hours.

As is common with Cloud providers, we meter and bill snapshot storage using two dimensions:

Snapshots in Elasticsearch Service save data incrementally at each snapshot event. This means that the effective snapshot size may be larger than the size of the current indices. The snapshot size increases as data is added or updated in the cluster, and deletions do not reduce the snapshot size until the snapshot containing that data is removed.

API requests are executed every time a snapshot is taken or restored, affecting usage costs. In the event that you have any automated processes that use the Elasticsearch API to create or restore snapshots, these should be set so as to avoid unexpected charges.

You can use Kibana to configure a snapshot lifecycle management (SLM) policy to automate when snapshots are created and deleted, along with other options. To learn more, refer to the Snapshot and Restore documentation.

Note that reducing either the snapshot frequency or retention period limits the availability and the recency of available data to restore from. Your snapshot policy should be configured with both costs and data availability in mind in order to minimize the potential for loss of data. Note also that reducing snapshot frequency and retention will not necessarily decrease your storage costs significantly. For example, if your dataset is only growing over time, then the total amount of data stored across all of your snapshots will be equal to your cluster size, whether that’s split across 10 snapshots or 100.

Synthetic Monitoring browser tests are charged per test run (metered in 60 second increments). Lightweight tests are charged per location per month (per deployment) for up to 1k simultaneous test run capacity (~2.6 billion tests per month). Tests executed from private locations do not incur an execution charge. All test result data is stored in your deployment and billed for under existing dimensions.