Skip to content

GCP Storage

Scope

Cloud Storage (classes, lifecycle, retention, dual/multi-region), Persistent Disk (pd-balanced, pd-ssd, pd-extreme), Filestore (NFS tiers), Cloud Storage FUSE, Storage Transfer Service, and Transfer Appliance.

Checklist

  • [Critical] Is the appropriate Cloud Storage class selected per data access pattern? (Standard for frequent access, Nearline for <1/month, Coldline for <1/quarter, Archive for <1/year)
  • [Recommended] Are Cloud Storage lifecycle policies configured to automatically transition objects between storage classes and delete expired objects, with age, creation date, and custom time conditions?
  • [Critical] Is uniform bucket-level access enabled on all buckets (enforced via organization policy) to prevent ACL-based access and simplify IAM management?
  • [Recommended] Is object versioning enabled on buckets requiring data protection, with lifecycle rules to limit the number of retained noncurrent versions or delete them after a set period?
  • [Critical] Are retention policies and bucket locks configured for compliance workloads, understanding that locked retention policies are irreversible and prevent object deletion?
  • [Recommended] Is dual-region or multi-region storage configured for high-availability workloads, with turbo replication (15-minute RPO) enabled for dual-region buckets requiring fast geo-redundancy?
  • [Recommended] Are signed URLs or signed policy documents used for time-limited access to private objects, with V4 signatures and appropriate expiration times (max 7 days)?
  • [Recommended] Is the correct Persistent Disk type selected for Compute Engine and GKE workloads? (pd-balanced for general use, pd-ssd for high IOPS, pd-extreme for >100K IOPS with provisioned throughput, pd-standard for cost-sensitive batch)
  • [Recommended] Is Filestore configured with the appropriate tier (Basic HDD, Basic SSD, Zonal, Regional, Enterprise) for NFS workloads, with capacity planned for performance scaling?
  • [Optional] Are Cloud Storage FUSE mounts used only for appropriate workloads (ML training data, batch processing), understanding POSIX compliance limitations and caching behavior?
  • [Optional] Are Storage Transfer Service or Transfer Appliance configured for large-scale data migrations, with bandwidth and scheduling policies to avoid network saturation?
  • [Optional] Are bucket-level notifications configured via Pub/Sub for event-driven processing of object creates, deletes, and metadata updates?
  • [Optional] Is requestor-pays enabled on shared data buckets to shift egress and operation costs to data consumers rather than the bucket owner?

Why This Matters

GCP Cloud Storage is the foundational object store with a uniquely tiered pricing model: storage class transitions are free when moving to colder tiers, but retrieval costs increase significantly (Archive retrieval is $0.05/GB vs $0 for Standard). Lifecycle policies are essential for cost control but must be designed carefully because class transitions are one-way (Standard to Nearline, never Nearline to Standard automatically). Persistent Disk performance scales linearly with disk size for pd-balanced and pd-ssd, meaning undersized disks deliver poor IOPS -- a common surprise for teams migrating from AWS EBS where performance tiers are more explicit. Filestore pricing is based on provisioned capacity, and performance scales with tier and size, making right-sizing critical.

Common Decisions (ADR Triggers)

  • Storage class strategy -- single class vs autoclass (automatic class management) vs manual lifecycle rules, cost modeling for access patterns
  • Replication model -- single-region (cheapest) vs dual-region (specific region pair, turbo replication option) vs multi-region (US/EU/ASIA)
  • Block storage type -- pd-balanced (default) vs pd-ssd vs pd-extreme (provisioned IOPS), regional persistent disk for HA
  • File storage tier -- Filestore Basic vs Zonal/Regional vs Enterprise (snapshots, replication), Filestore vs Cloud Storage FUSE vs Parallelstore for HPC
  • Access control model -- uniform bucket-level access (IAM only) vs fine-grained ACLs, signed URLs for external sharing
  • Data protection -- object versioning vs bucket retention policies vs Object Lock, soft delete (default 7-day recovery)
  • Data transfer method -- gsutil/gcloud CLI vs Storage Transfer Service vs Transfer Appliance (offline, 100TB+), interconnect transfer
  • Cost optimization -- autoclass vs manual lifecycle rules, requestor-pays for shared datasets, committed use discounts on Persistent Disk

Reference Architectures

See Also

  • general/data.md -- general data architecture patterns
  • providers/gcp/data.md -- GCP database and managed data services
  • providers/gcp/disaster-recovery.md -- Cloud Storage replication and cross-region DR
  • providers/gcp/containers.md -- GKE persistent volume integration