Spot VMs and managed instance groups
Spot VMs trade a 60-91% discount for 30-second reclaim notice — a sound trade only when the workload's own architecture already tolerates losing an instance.
Speaker notes
- A Spot VM runs on spare capacity at a steep discount in exchange for GCP being able to reclaim it with only 30 seconds' notice — always TERMINATE on preemption, never MIGRATE. It fits stateless, horizontally-scaled batch work and the elastic overflow portion of a fleet with a reliable standard-VM floor; it never fits a single stateful primary.
- Tell the real scenario: during a traffic spike, a batch of Spot instances got preempted mid-surge — but because the GPS-ingestion workers never hold an in-flight message longer than the time to process it, Pub/Sub's redelivery guarantee absorbed the loss as a brief blip, not data loss. The workload's own design made the preemption safe, not luck.
- A Managed Instance Group maintains a fleet of identical VMs from one instance template, automatically replacing unhealthy instances — the foundation everything on the next slide (autoscaling, health checks) builds directly on top of.
Deck map
01
GCP foundations in one line
02
How the ACE exam maps to this course
03
GCP's geography and the project as the unit of isolation
04
Organization, folders, and projects
05
Org policies and how inheritance works
06
Labels, network tags, and resource manager tags
07
Billing accounts, budgets, and cost control
08
gcloud CLI, Cloud Shell, and client libraries
09
Terraform on GCP and the end of Deployment Manager
10
Gemini CLI, Cloud Assist, and Application Design Center
11
Principals, roles, bindings, and policies
12
Primitive, predefined, custom roles, and IAM conditions
13
Service accounts, keys, and impersonation
14
Workload Identity Federation
15
Break-glass access and IAM auditing
16
Choosing a machine family: E2, N4, C4
17
Disks, OS Login, and VM Manager
18
Spot VMs and managed instance groups
19
Autoscaling policies, health checks, GPUs and TPUs
20
GKE Autopilot vs Standard vs the 2026 hybrid option
21
Node pools and pod autoscaling
22
Cloud Run revisions, traffic splitting, and functions
23
Choosing between GKE, Cloud Run, and Compute Engine
24
Cloud Storage classes and lifecycle management
25
Choosing a managed database
26
The managed database lineup
27
Connection pooling, read replicas, and Pub/Sub
28
Backup, regional failover, and CMEK
29
VPC networks, subnets, and Shared VPC
30
Firewall rules, tags, and Cloud NAT
31
Load balancers, Cloud DNS, and Cloud CDN
32
VPN, Interconnect, and private access
33
How a request actually reaches shipment-api
34
Cloud Monitoring: metrics, dashboards, alerts
35
Cloud Logging: router, buckets, and audit logs
36
Trace, Profiler, Error Reporting, and Managed Prometheus
37
How the five Cloud Operations products fit together
38
Readiness checklist