LEARN. PRACTICE. GROW.Concept and mechanism
Compute selection starts with the execution contract: work duration, interruption tolerance, durable state, and required interfaces. A checkpointed batch can use Spot VMs if recovery avoids duplicate effects and the window tolerates unavailable capacity. An application depending on CRDs and operators needs a compatible Kubernetes platform. An HTTP application without that API requirement may have other operational options. In Cloud Run, concurrency is the number of in-flight requests per instance. Twelve instances with eight requests each represent 96 simultaneous requests in that set; they do not automatically represent 96 requests per second because request duration also matters.
Guided application
In a reconciliation project, design a small experiment: interrupt a worker in an authorized environment, retain confirmed results, and resume using the same identifiers. Compare totals, duplicates, and completion time. For a MIG, inspect the template, scaling limits, and downstream capacity; adding workers can saturate the database. Infrastructure plans require review even when generated by an AI tool. If a label change proposes database replacement, hold application and investigate the diff. Retain the decision, approved version, and recovery plan so RUN can repeat the process without depending on the author.
A checkpoint is useful only if it matches confirmed effects and survives worker loss; rehearse failures between processing and confirmation.
Common pitfalls
Spot as guaranteed capacity; concurrency as throughput; templates as automatic review; AI as change approval.
Related topics: Data, availability, and private networking · Operating releases, nodes, and recovery
Capacity must respect state, dependencies, and the change process.
Reference: Spot VMs · Standard exam guide linked 2026-09-29; edition date unconfirmed