Filestore
Filestore is Google's managed NFS service. It provides a fully POSIX-compliant shared filesystem that multiple VMs, GKE pods, or Cloud Run instances can mount and access simultaneously with full read-write concurrency.
Service Tiers
- Basic HDD — legacy tier: lowest cost, single-zone, development use
- Basic SSD — legacy tier: fixed high performance, single-zone
- Zonal — the current default for production single-zone workloads
- Regional — replicates across zones, high availability
- Enterprise — regional replication, now positioned as the multishares option for GKE
In the Zonal and Regional tiers, throughput scales with provisioned capacity, and performance can also be provisioned independently of size. The legacy Basic tiers behave differently: Basic SSD delivers the same fixed performance at any size, while Basic HDD steps up in capacity bands. Google steers new instances toward Zonal and Regional.
Typical Use Cases
- Content management systems with shared media directories
- Shared home directories for compute clusters
- HPC workloads with parallel filesystem access
- Legacy applications requiring shared filesystem access
- GKE workloads using ReadWriteMany persistent volumes
When Not to Use Filestore
If Cloud Storage can meet the requirement, use it instead. Cloud Storage is dramatically cheaper, requires no capacity provisioning, and is globally accessible. Needing to store files does not mean you need Filestore — most workloads that store and retrieve files are better served by Cloud Storage. Filestore is for applications that need a mounted filesystem with shared read-write access across multiple VMs or containers simultaneously.
- Under-provisioning capacity — in the Zonal and Regional tiers, throughput scales with provisioned size.
- Using Basic HDD for production workloads.
- Not taking backups.
- Treating Cloud Storage and Filestore as interchangeable.
- Sharing one Filestore instance across very different workloads — a noisy batch job can degrade latency for interactive workloads on the same share.
- Use Regional tier for production workloads.
- Enable Filestore Backups for production shares.
- Monitor filesystem utilization to avoid full-filesystem errors.
- Use Cloud Storage instead wherever possible — it's cheaper and requires no capacity planning.
- Choose Zonal or Regional tier when you need to provision throughput independently of capacity — the Basic tiers tie the two together.
Knowledge Check
Which Filestore tier provides zone replication for high availability?
- Basic HDD
- Basic SSD
- Zonal
- Regional
When is Filestore the right choice instead of Cloud Storage?
- When you need the cheapest possible per-GB storage cost for archived data
- When you need global file access from any region worldwide with the lowest possible latency
- When applications require a mounted POSIX filesystem with shared read-write access
- When you need to store individual files that are larger than 5 TB each
For most workloads that need to store and retrieve files, what should you use instead of Filestore?
- Persistent Disk with an NFS server you install and configure manually on a VM
- Cloud Storage — cheaper, needs no capacity planning, and globally accessible
- Bigtable — optimized for storing and indexing structured file metadata at high scale
- BigQuery — purpose-built for large-scale file analytics and reporting
How does throughput relate to provisioned capacity across Filestore tiers?
- Throughput is fixed regardless of provisioned capacity in every tier
- Zonal and Regional scale with provisioned capacity; Basic SSD is fixed and Basic HDD steps in bands
- Throughput scales with the number of clients connected simultaneously
- Throughput is limited only by the VPC network bandwidth
Which protocol do clients use to mount a Filestore share?
- NFS — clients mount the share over the network with a standard client
- SMB / CIFS — the same network protocol that Windows file shares rely on
- FUSE — a userspace filesystem driver that Google provides for mounting
- iSCSI — block-level access to the share exposed as a remote raw disk
You got correct