Persistent Disk, Hyperdisk, Snapshots and Images

GCP Basics · Compute Engine – Machine Types, VMs, Disks and Managed Instance Groups · Lesson 3 of 5

Choose a disk type for performance and cost, grow disks online, protect them with scheduled incremental snapshots, and build custom images and image families for repeatable VMs.

VM storage is network block storage: the disk lives apart from the VM, so you can detach it, resize it and snapshot it.

flowchart LR
  VM["VM db-1"] --> PD["Persistent Disk data-1<br/>pd-balanced 200 GB"]
  PD -- "schedule daily 02:00" --> SN["Snapshots<br/>incremental, multi-regional"]
  SN --> NEW["New disk in another zone"]
  BOOT["Boot disk of web-1"] --> IMG["Custom image web-v3<br/>family web"]
  IMG --> TPL["Instance template"]
  %% step VM: db-1 keeps its data on a separate Persistent Disk, not on the boot disk.
  %% step PD: pd-balanced gives SSD-backed performance at a moderate price; performance grows with size.
  %% step SN: A snapshot schedule takes incremental snapshots and deletes old ones after the retention period.
  %% step NEW: Any snapshot can become a new disk in any zone or region, for restore or migration.
  %% step BOOT: A configured boot disk becomes a custom image, grouped in the image family web.
  %% step TPL: Instance templates reference the family, so new VMs start from the latest image.

Disk types

Type	Backed by	Use
`pd-standard`	HDD	Cheap, sequential, backups
`pd-balanced`	SSD	Default for most VMs
`pd-ssd`	SSD	Databases needing more IOPS
`pd-extreme`	SSD	Very high IOPS (older option)
Hyperdisk (Balanced, Throughput, Extreme, ML)	Next-generation	Provision IOPS and throughput independently of size; required by newer machine series such as C4 and N4
Local SSD	Physically attached	Fastest, but data is lost when the VM stops or is deleted

Zonal disks live in one zone; regional Persistent Disk / Hyperdisk Balanced HA replicate synchronously to a second zone.
Disks can be resized up while attached; then grow the file system (resize2fs, xfs_growfs). They cannot shrink.
All disks are encrypted at rest; use CMEK if required.

Snapshots

Incremental: only changed blocks after the first one; deleting a snapshot never breaks the others.
Stored multi-regionally by default (or in a chosen region).
Snapshot schedules (resource policies) automate daily/weekly snapshots with retention.
For consistent database snapshots, flush or freeze writes first (or use application-consistent snapshots on Windows/Linux guest agents).
Archive snapshots cost less for long retention.

Images and machine images

	Use
Custom image	One boot disk you configured; share across projects; group in an image family; deprecate old ones
Machine image	A whole VM: all disks, metadata, machine type; good for cloning or backup of a VM

gcloud compute resource-policies create snapshot-schedule daily-2am --region=asia-southeast1   --daily-schedule --start-time=19:00 --max-retention-days=14
gcloud compute disks add-resource-policies data-1 --resource-policies=daily-2am --zone=asia-southeast1-a
gcloud compute images create web-v3 --source-disk=web-1 --source-disk-zone=asia-southeast1-a --family=web

(--start-time is UTC: 19:00 UTC is 02:00 in Vietnam.)

Command	Meaning
`gcloud compute disks resize data-1 --size=400GB --zone=…`	Grow a disk
`gcloud compute disks snapshot data-1 --snapshot-names=data-1-manual --zone=…`	One-off snapshot
`gcloud compute images deprecate web-v2 --state=DEPRECATED --replacement=web-v3`	Retire an old image

Traps: Local SSD is not durable; resizing the disk does not resize the file system; creating an image from a running boot disk needs --force and may be inconsistent.