google-cloud-storage-basics

Run gcloud storage buckets create gs://my-bucket --location=us-central1 after enabling the storage.googleapis.com API. Bucket names share a single global namespace, so short or common names are usually taken; omitting the location defaults to the US multi-region.

Use Storage Transfer Service with gcloud transfer jobs create, providing an S3 credentials JSON file containing your access key ID and secret. STS runs serverless on Google infrastructure, avoiding local bandwidth use for multi-terabyte migrations.

gcloud storage is the recommended CLI; gsutil is minimally maintained and lacks newer features like soft delete and managed folders. Existing gsutil commands should be replaced with their gcloud storage equivalents.

No. gcsfuse does not support file locking, random writes, hard links, or extended attributes, and chmod/chown are silently ignored. It is optimized for high-throughput sequential reads and writes, not as a drop-in NFS replacement.

Use a Rapid Bucket for new latency-critical workloads colocated with compute in one zone, such as AI/ML training, since it offers sub-millisecond latency but no cross-zone redundancy. Standard regional or multi-region buckets remain the right default for most workloads.

Parallel composite uploads create temporary component objects that cannot be deleted until the retention period expires, increasing storage costs. By default gcloud skips PCU when the destination bucket has a retention policy or default object holds.