A DevOps engineer operates a Google Kubernetes Engine (GKE) Standard cluster running a monolithic batch processing job that requires 18 hours of continuous, uninterrupted execution and cannot recover from unexpected pod evictions. A team member proposes migrating the cluster's worker node pool from Standard VMs to Spot VMs to lower compute costs. Which operational decision should the DevOps engineer make to guarantee that the workload completes successfully?
- Retain standard Compute Engine instances for the node pool because Spot VMs can be reclaimed by GCP at any time with a short termination notice, making them unsuitable for non-fault-tolerant, long-running batch jobs.Answer
- BMigrate the node pool to Spot VMs and configure the Horizontal Pod Autoscaler (HPA) to immediately replace nodes whenever GCP reclaims a Spot instance.
- CMigrate the node pool to Spot VMs and configure the Horizontal Pod Autoscaler (HPA) to increase the node count whenever resource usage spikes.
- DMigrate the workload to a GKE Autopilot cluster using Spot Pods, because Autopilot guarantees immunity from node preemption for batch workloads.
Answer
Retain standard Compute Engine instances for the node pool because Spot VMs can be reclaimed by GCP at any time with a short termination notice, making them unsuitable for non-fault-tolerant, long-running batch jobs.
Spot VMs are provisioned from spare compute capacity and can be reclaimed by Google Cloud with a 30-second notice. Workloads that run for long durations without built-in fault tolerance or state checkpointing cannot run reliably on Spot instances and must use standard Compute Engine node pools.
Step-by-Step Solution
Key Concept
Selecting appropriate GKE node pool VM types based on workload fault tolerance and operational continuity requirements.