⚙️ What Are Spot / Preemptible Instances?

Definition:

Spot (AWS) or Preemptible (GCP) instances are temporary, low-cost virtual machines offered by cloud providers. They are unused compute capacity that providers sell at a discount (often 70–90% cheaper than regular instances).

Key Characteristic:

• Can be terminated (preempted) at any time if the cloud provider needs the resources back.

---

🔄 How They Work

1. Provider has spare capacity → offers instances at low cost
2. User requests a spot/preemptible instance → gets it while available
3. Provider can reclaim the instance → gives short notice before termination (AWS: 2 min, GCP: 30 sec)

---

⚡ Pros

1. Cost Savings

   • Much cheaper than on-demand instances

2. Scalability

   • Can spin up many instances at low cost for batch jobs

3. Flexibility

   • Ideal for stateless, fault-tolerant workloads

---

⚠️ Cons

1. Not guaranteed to run

   • Can be terminated anytime → not suitable for critical workloads

2. Requires fault-tolerant design

   • Must handle instance loss gracefully (retry, checkpointing)

3. Limited maximum runtime (GCP preemptible: 24 hours max)

---

🛠️ Common Use Cases

Use Case	Reason for Using Spot/Preemptible
Batch Processing	Can restart failed tasks without affecting cost
Big Data / Analytics	Scale clusters cheaply (Hadoop, Spark)
CI/CD Pipelines	Non-critical build jobs can tolerate interruption
Rendering / Video Transcoding	Tasks are stateless → can resume if instance is lost
Fault-Tolerant Microservices	With autoscaling and redundancy

---

🔧 Best Practices

1. Checkpoint / Save State Frequently

   • So work can resume on another instance after termination

2. Use Auto Scaling Groups / Managed Instance Groups

   • Automatically replace terminated instances

3. Mix with On-Demand / Reserved Instances

   • Critical services on on-demand, batch jobs on spot → cost efficiency + reliability

4. Monitor Termination Notices

   • AWS gives 2 min warning → gracefully stop or save progress

---

🧩 Analogy

• Spot / Preemptible instance = hotel room leftover inventory

  • Very cheap, but hotel can reclaim it anytime if needed

• Regular instance = reserved hotel room

  • Always available, costs more

---

✅ Key Takeaways

1. Spot / Preemptible instances are cheap, temporary cloud VMs
2. Preemptible / spot = “use if available” → not reliable for critical workloads
3. Best for batch, stateless, or fault-tolerant workloads
4. Must design systems to handle interruptions gracefully

---