Fault-tolerant query execution in a Managed Service for Trino cluster

Managed Service for Trino clusters support the fault-tolerant execution mechanism. By default, when resources are insufficient or a query fails with an error, the worker will stop processing the query and you will need to run it again manually. With fault-tolerant execution enabled, the query or its component task will be retried in the event of failure.

To ensure fault-tolerant query execution when Fault tolerant execution is enabled, Exchange Manager regularly exports intermediate results to an Object Storage service bucket. If a query fails, the coordinator will assign the retry to another worker, which can use the previously processed data.

Fault-tolerant execution is only supported for the following connectors:

• Delta Lake
• Hive
• Iceberg
• Oracle Preview
• PostgreSQL
• MS SQL Server Preview

To enable fault-tolerant execution, select the Retry object type parameter under Retry policy when creating or updating a cluster:

• Task: Retries the intermediate task within the query that caused worker failure.

  Warning

  For high volumes of simple queries, retrying tasks may lead to longer delay in complex batch queries. We recommend creating a dedicated Managed Service for Trino cluster with task retry settings for complex batch queries.

• Query: Retries all stages of the query where worker failure occurred.

Exchange Manager storage typesExchange Manager storage types

You can select between two bucket types to use as storage for Exchange Manager intermediate results:

• Service bucket: This bucket is created and maintained by Managed Service for Trino at no extra cost. It requires zero configuration but has limited size. This capacity may be insufficient for queries with large amounts of intermediate data.

• Custom bucket: Object Storage bucket you create yourself within your own folder. Bucket capacity is governed by Object Storage quotas, with storage and operation fees applied according to the Object Storage pricing policy.

  To allow a Managed Service for Trino cluster to write data to a custom bucket, assign the storage.editor role or a role with equivalent bucket permissions to the cluster's service account.

The service bucket is used by default. You can select the storage type when creating or updating a cluster.

For more information about fault-tolerant query execution, see the official guides.