Permissions Setup

For Athena Querying

Requires:

• S3 Tables catalog registered in Glue (s3tablescatalog/aws-redshift)
• Athena execution permissions and a workgroup with an output location
• S3 Tables read permissions (see the least-privilege policy in ${SKILL_DIR}/references/security.md)

Encrypt the workgroup’s output location. Athena writes full result sets to S3, so query_text, user_name, and remote_host from the SYS_* tables land there in plaintext unless the workgroup enforces encryption. Configure SSE-KMS and lock it so query authors cannot override it:

EnforceWorkGroupConfiguration=true is the part that matters — without it a client can pass its own unencrypted ResultConfiguration per query. Verify with aws athena get-work-group --work-group <WORKGROUP>.

Confirm the catalog is registered:

• Returns namespaces (databases) → catalog is registered and queryable.
• EntityNotFoundException / CATALOG_NOT_FOUND → S3 Tables integration not enabled. Enable the S3 Tables integration: S3 console > Table buckets > Enable integration.

For Redshift Querying (Auto-Mounted S3 Tables Catalog)

Prerequisites:

• A Provisioned RA3 cluster. Auto-mount support depends on node type; confirm the current supported node types in the Redshift documentation (opens in a new tab) or via aws redshift describe-orderable-cluster-options rather than assuming a fixed list.
• The Glue s3tablescatalog S3 Tables catalog must exist (auto-created when the table bucket is integrated with analytics services)

Step 1: Create IAM Role with Required Permissions

Create a role (e.g., query_s3_tables) with:

Trust Policy:

All four actions (sts:AssumeRole, sts:SetContext, sts:SetSourceIdentity, sts:TagSession) are mandatory for lakeformation.amazonaws.com. Without them, Lake Formation cannot assume the role for federation.

The aws:SourceAccount conditions guard against the confused deputy problem (opens in a new tab): a bare service principal with no condition can be assumed on behalf of any account, so a caller in another account could induce the service to use this role. Restrict to the account that owns the cluster. Use aws:SourceArn with the cluster ARN instead if you want to pin to a single cluster.

Create the role and attach the read-only query policy. Pass both documents inline so the commands work unchanged through the AWS MCP server’s call_aws tool, which cannot read local files:

At a terminal you can substitute file://trust-policy.json / file://inline-policy.json for the inline strings, which avoids shell-quoting problems with long documents. Inline is the primary form because file:// silently fails wherever the executing agent has no filesystem.

Do not attach AWSLakeFormationDataAdmin to this role. The role above is attached to the cluster (Step 2) and is used to serve queries; it needs read access only. The Lake Formation setup steps below (register-resource in Step 3, put-data-lake-settings in Step 4) are data-lake-administrator operations that AdministratorAccess alone does not satisfy — Lake Formation gates them on data lake admin status rather than on IAM alone. Run those steps as the human or automation principal performing setup, not as the cluster’s role, so the cluster never holds administrative Lake Formation permissions at runtime:

AWSLakeFormationDataAdmin is a broad starting point, not a production posture: it grants administrative control over every Lake Formation resource in the account, including PutDataLakeSettings, which can rewrite the admin list. For production, replace it on the setup principal with a custom policy holding only the setup actions actually used here, and detach it once setup completes:

Once setup is complete, detach it from the setup principal too — nothing in steady-state querying needs it. The cluster’s role only ever needs the read-only inline policy below.

• Inline policy for S3 Tables, Glue, and Lake Formation access, scoped to the aws-redshift table bucket and its catalog:

lakeformation:GetDataAccess is the one action here that cannot be scoped by resource ARN — the Lake Formation documentation states that "Resource": "*" is required and “specifying any other resource for this permission is not supported.” The actual data authorization comes from the Lake Formation grants, not from this statement. The aws:ResourceAccount condition constrains it to same-account table buckets, so the role cannot be used to vend credentials for a table bucket shared in from another account. Use StringEquals, not StringLike — aws:ResourceAccount is a sensitive condition key and wildcards in it defeat the check.

Note the shape of the Glue database and table ARNs: federated S3 Tables catalogs nest under s3tablescatalog/<table-bucket>, so the resource path is database/s3tablescatalog/aws-redshift/*, not the bare database/*. The bare form would grant metadata read on every database and table in the account’s default Glue catalog — far more than querying published system tables needs.

If you hit a permission error during initial setup that the scoped policy above doesn’t cover, widen it deliberately and narrow it back down for production — do not fall back to "Action": "*" on "Resource": "*".

Step 2: Attach Role to Cluster

Step 3: Register the Table Bucket with Lake Formation

Scope the registration to just the aws-redshift table bucket so the role cannot be used to reach other table buckets in the account:

Note: VerificationStatus: NOT_VERIFIED after registration is normal and does not block functionality.

Step 4: Add Redshift SLRs as Lake Formation Read-Only Admins

WARNING: put-data-lake-settings REPLACES the entire settings object. Always include your existing DataLakeAdmins alongside the new ReadOnlyAdmins.

Step 5: Verify Auto-Mount

The cluster polls every 300 seconds. After up to 5 minutes:

Expected output includes: aws-redshift@s3tablescatalog

If it doesn’t appear after 5 minutes, a cluster reboot triggers immediate discovery.