Persistent Disk
Persistent Disk is the network-attached block storage service for Compute Engine VMs and GKE nodes. From the OS perspective it behaves like a physical disk. Physically, data lives on Google's distributed storage infrastructure — a disk survives VM deletion, can be reattached to a different VM, and is replicated automatically within the zone.
Disk Types
| Type | Media | Best for |
|---|---|---|
| Standard PD | HDD | Sequential batch workloads, backups, lowest cost |
| Balanced PD | SSD | General workloads — best cost/performance balance |
| SSD PD | SSD | High-IOPS databases, demanding transactional workloads |
| Extreme PD | SSD | Enterprise databases, custom IOPS provisioning |
| Hyperdisk | SSD | Decouple IOPS, throughput, and capacity provisioning |
| Local SSD | NVMe (physical) | Cache, scratch — data lost when VM stops |
Local SSD is physically attached to the server hosting the VM, delivering dramatically higher IOPS and lower latency than network-attached disks. Per GB it is actually cheaper than SSD Persistent Disk — the real trade-offs are ephemerality and the fixed 375 GiB increments it comes in. Data survives a reboot and live migration but is lost when the VM stops, is suspended, or is deleted — use only for read cache, scratch, and temporary data where loss is acceptable.
Snapshots
Snapshots are incremental, capturing only changed blocks after the first full snapshot. Automated snapshot schedules attach to disks and create snapshots on a configurable cadence with retention policies. Every production disk should have an automated snapshot schedule.
Disk Resizing
Disks can be resized online without stopping the VM. After increasing disk size, resize the partition and filesystem inside the VM. Disks can only grow — there is no shrink operation.
Regional Persistent Disks
Regional PD synchronously replicates data across two zones in a region. If the zone hosting the VM goes down, reattach the disk in the second zone. Appropriate for stateful production workloads requiring zone-level durability.
- Choosing Standard HDD for a transactional database.
- No snapshot schedule on production disks.
- Assuming disk resizing also resizes the filesystem inside the VM — that step is separate.
- Using zonal PD for stateful services where zone failure would be unacceptable.
- Keeping snapshots only in the same region as the disk — a regional outage takes both the disk and its snapshots offline. Copy critical snapshots to a second region.
- Start with Balanced PD for most workloads.
- Configure automated snapshot schedules for all production disks.
- Use Regional PD for stateful workloads requiring zone-level durability.
- Use Local SSD for cache and scratch only — never for primary persistent data.
Knowledge Check
What happens to data on a Local SSD when the VM stops?
- Data is automatically snapshotted to Cloud Storage before the VM stops
- Data persists and is available the next time the VM starts
- Data is permanently lost
- Data is migrated to a Persistent Disk automatically
Which Persistent Disk type is recommended for most general-purpose workloads?
- Standard PD — HDD-backed, lowest cost per GB
- Balanced PD — best cost-to-performance balance
- SSD PD — high IOPS for demanding database workloads
- Extreme PD — custom provisioned IOPS for enterprise databases
After resizing a Persistent Disk online, what additional step is required inside the VM?
- Nothing — the OS automatically detects and uses the additional space
- Restart the VM to apply the new disk size
- Resize the partition and filesystem inside the VM
- Detach and reattach the disk to force a rescan
What makes Regional Persistent Disk different from a standard zonal PD?
- It can be attached to many VMs in different zones at once for concurrent shared read-write access
- It synchronously replicates data across two zones; if a zone fails you reattach in the other
- It provides globally distributed storage spread across regions with no zone affinity
- It delivers roughly 10× higher IOPS than a standard zonal Persistent Disk of the same size
How do incremental Persistent Disk snapshots work after the first one?
- Each snapshot is a complete full copy of the entire disk every time
- Each snapshot stores only the blocks changed since the previous one
- Snapshots automatically compress all stored data by approximately 50 percent
- Snapshots are taken automatically every hour by default unless disabled
You got correct