What does Snowflake attempt to do if any of the compute resources for a virtual warehouse fail to provision during start-up?

When a virtual warehouse is started or resumed, Snowflake provisions compute resources from the cloud provider. If any of these resources fail to provision successfully, Snowflake's control plane automatically initiates a recovery process. This process involves attempting to restart the warehouse. A restart effectively retries the provisioning of the required compute nodes. This automated retry mechanism is designed to handle transient issues with the underlying cloud infrastructure, ensuring the warehouse becomes available without manual intervention.

A. Repair the failed resources. Snowflake uses ephemeral, stateless compute instances from the cloud provider. It does not repair individual failed instances; it replaces them by retrying the provisioning process.
C. Queue the failed resources. Queuing the failed resources would delay the warehouse's availability. The goal is to bring the warehouse online, so Snowflake actively retries provisioning, it doesn't queue the request.
D. Provision the failed resources. While Snowflake does attempt to provision resources again, "restart" is the more accurate and specific term used in the documentation to describe the automated recovery action for a failed start-up.

1. Snowflake Documentation: Troubleshooting Virtual Warehouses.

Reference Section: Under the topic "Warehouse Fails to Start or Resume"

the documentation states: "If a warehouse fails to start

Snowflake will attempt to restart the warehouse automatically." This directly supports the concept of restarting as the recovery action.