This feature is currently in Private Preview, and only available to customers of SELECT for Databricks.
When Databricks runs on classic compute, AWS bills you for the EC2 instances, EBS volumes and networking underneath. Once your AWS cost data is connected, SELECT shows that AWS cost next to the DBU cost of each workspace, cluster, SQL warehouse, job and query, so you can see the full cost of running Databricks.
Prerequisites
- Admin access to your AWS payer (management) account, or at least permission to manage Billing, Data Exports, S3 and IAM
- Only the payer account can manage cost allocation tags, and an export created there includes every member account in your AWS organization, so one connection covers all of your Databricks workspaces
- A member account can connect its own export instead, which covers only that account's usage. Step 1 still has to be done from the payer account
- Your Databricks workspace already connected to SELECT
Connecting your AWS bill
Step 1: Activate the Databricks cost allocation tags
Databricks automatically tags the EC2 instances and EBS volumes it launches. SELECT uses those tags to tie AWS cost back to a cluster, SQL warehouse or job. AWS only includes the tags in your cost data if they're activated as cost allocation tags.
- In the AWS console, open Billing and Cost Management → Cost allocation tags
- Open the User-defined cost allocation tags tab, then find and select these tags:
Vendor: set on all compute. Identifies resources Databricks launched (valueDatabricks)ClusterId: set on clusters and SQL warehouses. Ties instances and volumes to a clusterSqlEndpointId: set on SQL warehouses. Ties instances to a SQL warehouseName: set on all compute. Shows the workspace on AWS cost in Cost ExplorerDatabricksInstancePoolId: set on instance pools. Ties pool instances to a poolDatabricksInstancePoolCreatorId: set on instance pools. Display onlyClusterName,Creator: set on clusters. Display only
- Click Activate
Timing: a tag only appears in this list after AWS has seen it on a resource. Once it's activated, AWS can take up to another 24 hours to apply it. So allow up to two days in total before tags show up in your cost data.
Instance pools: instances that come from an instance pool only carry the pool's tags (plus your workspace's custom tags), not the cluster's. SELECT handles this by matching each instance to the clusters it served, using Databricks' own node timeline, so pool cost still reaches the right clusters and jobs. Time an instance spent idle in the pool shows up as idle pool cost. Activate the pool tags above either way.
Optionally, but recommended, backfill the tags so earlier months are included. By default, tags only apply to cost data from the moment you activate them. You can backfill them for up to the last 12 months:
- On the same Cost allocation tags page, click Backfill tags (top right)
- Pick the month to start from, then click Confirm
- You can make one backfill request every 24 hours. AWS only fills in a tag value where the resource actually had that tag at the time, and the updated history reaches your data export within about 24 hours
- AWS docs: Backfill cost allocation tags
Step 2: Create a CUR 2.0 data export
Already have a Cost and Usage Report? SELECT also reads legacy Cost and Usage Reports (CUR 1.0) and CSV exports, gzipped or zipped. You can skip to Step 3 and use that report's bucket and prefix. For a new export, use the settings below.
- Open Billing and Cost Management → Data Exports, then click Create export
- Export type: choose Standard data export
- Export name: pick a memorable name (e.g.
select-aws-cost-export), and note it down - Data table configurations:
- Data table: CUR 2.0
- Check Include resource IDs
- Time granularity: Hourly
- Leave the other options (such as Split cost allocation data and Include Capacity reservation data) at their defaults
- Column selection: select all columns, using the checkbox in the table header. If you need to limit the export, include at least the columns listed in Required columns below
- Required columns: Selecting all columns includes these. If you limit the export, keep at least:
- Billing period and time:
bill_billing_period_start_date,line_item_usage_start_date,line_item_usage_end_date - Accounts:
bill_payer_account_id,bill_payer_account_name,bill_billing_entity,line_item_usage_account_id,line_item_usage_account_name - Line item:
line_item_line_item_type,line_item_line_item_description,line_item_usage_type,line_item_usage_amount,line_item_resource_id,line_item_product_code,line_item_availability_zone - Product:
product,product_region_code,product_product_family - Cost:
line_item_unblended_cost,line_item_net_unblended_cost - Savings Plans:
savings_plan_savings_plan_a_r_n,savings_plan_savings_plan_effective_cost,savings_plan_net_savings_plan_effective_cost,savings_plan_total_commitment_to_date,savings_plan_used_commitment - Reservations:
reservation_reservation_a_r_n,reservation_effective_cost,reservation_net_effective_cost,reservation_unused_amortized_upfront_fee_for_billing_period,reservation_unused_recurring_fee,reservation_net_unused_amortized_upfront_fee_for_billing_period,reservation_net_unused_recurring_fee - Tags:
tags
- Billing period and time:
- A legacy report has different column names. SELECT checks the legacy equivalents, and names any that are missing when you add the account
- Required columns: Selecting all columns includes these. If you limit the export, keep at least:
- Data table delivery options:
- Compression type and file format: Parquet
- File versioning: Overwrite existing data export file
- Data export refresh cadence: Daily (the default)
- Report data integration: leave unset
- Data export storage settings: choose This account, then Configure
- We recommend creating a new bucket for the export (Create a bucket, then pick a name and a Region). Note down the bucket name and its region
- S3 path prefix: pick one (e.g.
cur). AWS delivers the export to a folder named after the export under it, so your prefix for SELECT is both together (e.g.cur/select-aws-cost-export). Note it down
- Click Create
AWS writes the bucket policy it needs to deliver the export. If you select an existing bucket instead of creating one, Data Exports overwrites that bucket's policy, and the console asks you to confirm it. Don't edit the policy later: changing it can stop deliveries. That includes the us-east-1 region in its SourceArn, which is correct whatever region your bucket is in.
AWS delivers the first export within 24 hours, then refreshes it at least once a day. The files land in:
For a legacy report, your prefix is its S3 path prefix followed by the report name: the folder that holds one YYYYMMDD-YYYYMMDD/ folder per month.
Want older history? A new export starts from the current month. To get earlier months, open a case with AWS Support from the payer account and ask them to backfill the export. SELECT loads up to 180 days of history.
Step 3: Copy SELECT's AWS account ID and your external ID
SELECT reads your export by assuming a role in your AWS account. It stores no AWS keys, and every session it opens expires within an hour. Your role's trust policy needs two values from SELECT.
- Navigate to: https://select.dev/app/settings/aws/add
- Under Authentication (IAM role), copy:
- SELECT AWS Account ID: the AWS account SELECT assumes your role from
- External ID: shown under Authentication (IAM role) on the same page
- Keep this page open. You'll come back to it in the last step
If the page says "Your organization isn't linked to DoiT yet", reach out to your SELECT contact. The External ID appears once it's set.
Step 4: Create a read-only IAM policy
SELECT needs read-only access to the export files, and nothing else.
- Open IAM → Policies → Create policy, then switch to the JSON editor
- Paste the policy below, and find/replace the two placeholders:
<BUCKET>: the bucket name from Step 2<PREFIX>: your prefix from Step 2 (e.g.cur/select-aws-cost-export)
- If you've set the bucket to encrypt with a KMS key you manage, also add this statement to the policy's
Statementlist, replacing<KMS_KEY_ARN>with the key's ARN:
- Click Next, give the policy a memorable name (e.g.
SelectCURReadOnly), and click Create policy
Step 5: Create a role for SELECT
- Open IAM → Roles → Create role
- Trusted entity type: choose AWS account, then Another AWS account
- Account ID: the SELECT AWS Account ID from Step 3
- Under Options, check Require external ID, and paste the External ID from Step 3
- Leave Require MFA unchecked
- Click Next. On Add permissions, select the policy from Step 4, and click Next
- Role name: pick a memorable name (e.g.
select-cur-reader), then click Create role - Open the new role and copy its ARN (e.g.
arn:aws:iam::123456789012:role/select-cur-reader)
The role's Trust relationships tab should now show this policy, with your two values in place of the placeholders:
The trust policy names SELECT's whole AWS account (:root), not a single role in it, so changes on SELECT's side never break it. The external ID is what limits it to your organization.
Add AWS account to SELECT
- Back on https://select.dev/app/settings/aws/add, fill out the details you noted in the previous steps:
- Display Name: any name you'll recognize (e.g.
Production AWS billing) - AWS Account ID: the 12-digit ID of the account that created the export in Step 2, usually your payer account. It's shown in the top-right account menu of the AWS console (e.g.
123456789012) - S3 Bucket: the bucket name only, without
s3://(e.g.my-cur-bucket) - S3 Prefix: your prefix from Step 2, the folder that contains
data/andmetadata/(e.g.cur/select-aws-cost-export). For a legacy report, the folder that contains theYYYYMMDD-YYYYMMDD/folders - AWS Region: the region your S3 bucket is in (e.g.
us-east-1) - Role ARN: the role's ARN from Step 5
- Display Name: any name you'll recognize (e.g.
- Click Add. SELECT runs these checks in order and shows the result of each:
- Assume your AWS role
- Reach the S3 bucket
- Find a CUR manifest under the prefix
- Read the latest CUR manifest
- Confirm the CUR export includes the columns SELECT needs
- When all of them pass, the connection is saved and the first sync starts. You are good to go!
Auditing SELECT's access
Every session SELECT opens on your role is named select-<connection ID>, or select-validate-<organization ID> for the checks before an account is added. In CloudTrail, the AssumeRole events and the S3 reads that follow show which SELECT connection read your files.
To stop SELECT's access at any time, delete the role, or remove SELECT's account from its trust policy.
What happens next
- The first sync loads every billing period in the export, up to 180 days back. After that, SELECT syncs once a day, so a new delivery shows up within about a day. AWS keeps restating the current month, and can update the previous month for up to two weeks after it ends, so SELECT reloads both months on every sync
- Payer and member exports together: you can connect both your payer account's export and a member account's own export. Each account's cost counts once: SELECT uses the payer's export first, then the connection added earliest
- Where you'll see AWS cost:
- On Databricks pages, next to DBU spend for each warehouse, cluster, job and query, as Cloud spend
- In Total spend, which is DBU spend plus Cloud spend
- In Cost Explorer, under the AWS platform
- Serverless compute shows
-for Cloud spend. Databricks runs serverless on its own AWS account and includes the infrastructure in the DBU price, so there's no separate AWS charge on your bill - AWS cost that doesn't belong to Databricks (other workloads, shared networking, tax) still appears in Cost Explorer under AWS, so the full bill adds up
Troubleshooting
Assume your AWS role
- Likely cause: the trust policy names the wrong account or external ID, or the Role ARN is wrong
- Fix: on the role's Trust relationships tab, check the principal is
arn:aws:iam::<SELECT_AWS_ACCOUNT_ID>:rootandsts:ExternalIdmatches the External ID shown in SELECT exactly. Re-copy the Role ARN from the role's summary page.
Assume your AWS role (SELECT could not sign in to its AWS broker role)
- Likely cause: a problem on SELECT's side, not your setup
- Fix: reach out to your SELECT contact.
Reach the S3 bucket
- Likely cause: wrong bucket name or region
- Fix: check the bucket name has no
s3://or trailing/. Check the region matches the bucket's region in S3.
Reach the S3 bucket (AccessDenied)
- Likely cause: the policy doesn't cover the bucket, or isn't attached to the role
- Fix: check the first statement's
Resourceisarn:aws:s3:::<BUCKET>, and that the policy is attached on the role's Permissions tab. If you've set the bucket to encrypt with a KMS key you manage, also allowkms:Decrypton that key.
Find a CUR manifest under the prefix
- Likely cause: the first export hasn't arrived yet, or the prefix is wrong
- Fix: AWS can take up to 24 hours to deliver. In S3, check that
<PREFIX>/metadata/exists and contains a file ending inManifest.json. For a legacy report, look for aYYYYMMDD-YYYYMMDD/folder under<PREFIX>/with one. Enter the prefix without a leading/.
Read the latest CUR manifest
- Likely cause: the policy's object path doesn't match where the files are
- Fix: check the second statement's
Resourceisarn:aws:s3:::<BUCKET>/<PREFIX>/*, matching the folders in S3 exactly, including case.
Confirm the CUR export includes the columns SELECT needs
- Likely cause: the export leaves out columns SELECT reads. The message names them.
- Fix: edit the export in Data Exports and add the missing columns (see Required columns in Step 2).
line_item_resource_idalso needs Include resource IDs checked.
Exports stopped arriving (an Invalid bucket error in Data Exports)
- Likely cause: the bucket policy or bucket owner changed after setup
- Fix: restore the bucket policy Data Exports created, or recreate the export.
Connected, but Cloud spend is - for classic compute
- Likely cause: tags not activated (Step 1), or not yet in the data
- Fix: activate the tags, and backfill them for earlier months. Allow up to two days for the tags to reach your data.
Stuck? Reach out to your SELECT contact with the failed check's message. It includes the AWS error code, which is usually all we need.



