Skip to content

Supported surface and limits

This page lists what is supported in the current Icebreaker release. Items marked planned are on the roadmap but not available in early-adopter access. Items marked not validated are expected to work but have not been confirmed end-to-end — reach out to your Icebreaker contact if you are piloting one.

CloudStatus
Amazon Web Services (AWS)Supported
Google Cloud (GCP)Planned
Microsoft AzurePlanned

Region availability within AWS is verified per early adopter during onboarding. A public region list will be published once it is stable; for now, confirm region support with your Icebreaker contact.

DistributionStatus
Amazon EKSSupported
Other Kubernetes distributionsNot validated

Karpenter is required for Capacity Policy-based pod scheduling. Without Karpenter installed, the Icebreaker Agent runs but cannot provision capacity per query.

Icebreaker integrates with catalogs that implement the Iceberg REST Catalog specification.

CatalogStatus
LakekeeperValidated end-to-end
AWS Glue (Iceberg integration)Expected to work, not validated
Apache PolarisExpected to work, not validated
Snowflake Open Catalog (managed Polaris)Expected to work, not validated
Other Iceberg REST-compatible catalogsExpected to work, not validated

If you are piloting with a catalog marked “expected to work,” reach out — we want to help you confirm it.

FormatStatus
ParquetValidated

Icebreaker’s planning and execution path is validated against Parquet. If you need support for another Iceberg-compatible format, reach out to your Icebreaker contact.

Icebreaker executes SQL using DataFusion as the embedded query engine. The SQL surface follows DataFusion’s dialect.

Supported:

  • SELECT, FROM, WHERE, GROUP BY, ORDER BY, LIMIT, OFFSET.
  • Inner, left, right, and full outer joins.
  • Standard aggregations (COUNT, SUM, AVG, MIN, MAX).
  • Window functions.
  • Standard scalar functions: arithmetic, string, date/time.
  • Common Iceberg-specific operations: time-travel queries, snapshot inspection.
  • CTEs (WITH).

Notable gaps:

  • Statement-level transactions are not supported (Icebreaker is a query engine, not a transactional database).
  • Some DDL operations (e.g., CREATE INDEX) are not applicable to the Iceberg model.
  • A current list of unsupported functions and operators is maintained with your Icebreaker contact.

Dialect notes:

  • Identifier quoting follows DataFusion defaults (double quotes for identifiers, single quotes for strings).
  • Type coercion follows DataFusion; some Iceberg-to-DataFusion type mappings may differ from other engines.

Each Data Server exposes an Arrow Flight SQL endpoint. The Admin console surfaces it in two forms.

The raw Arrow Flight SQL endpoint, used by native Arrow Flight SQL clients. Format depends on your Icebreaker Agent’s ingress mode:

Ingress modegRPC URL format
nginx with TLSgrpc+tls://{data_server_id}.{ingress-domain}:443
nginx without TLSgrpc://{data_server_id}.{ingress-domain}:80
AWS ALBgrpc+tls://{elb-hostname}:50051

The token is passed separately by native gRPC clients, typically as an Authorization: Bearer {token} header.

Same endpoint, formatted for JDBC clients. The token is embedded as a query parameter so most JDBC clients can connect by pasting the URL alone.

jdbc:arrow-flight-sql://{host}:{port}/?useEncryption=true&token={token}
ComponentMeaning
{host}The host from the gRPC URL (no protocol prefix).
{port}The port from the gRPC URL.
useEncryption=trueAlways set for TLS-enabled Data Servers.
token={token}The Data Server token. The Arrow Flight SQL JDBC driver presents it as a bearer credential.
ResourceLimit
Concurrent queries per Data ServerGoverned by Capacity Policy capacity and node availability
Query plan sizeNo fixed hard limit; very large plans may be rejected for memory reasons
Result set sizeLimited by results bucket capacity
Capacity Policies per Icebreaker AgentNo fixed limit (governed by Kubernetes namespace and Karpenter NodePool limits)
Data Servers per Icebreaker AgentNo fixed limit
Icebreaker Agents per AccountNo fixed limit

Concrete per-environment limits depend on your Capacity Policy configuration and your AWS account quotas. Your Icebreaker contact can help validate sizing for your specific workload.

  • Spark-style distributed transformations — Icebreaker focuses on SQL workloads.
  • Direct write-back to source tables — query results are written to a results bucket; updates to source tables go through your existing ingest pipeline.
  • Self-service account provisioning — Admin console access is granted manually during the early-adopter program.
  • Non-AWS cloud accounts — GCP and Azure support is planned.