15 February 2026

Blog Post Hackernews

I Built a Retry Script for Oracle’s “Out of Capacity” Free Tier Instances

TL;DR: Oracle Cloud’s free tier ARM instances are amazing but almost impossible to provision due to capacity issues. I built a smart retry script that got mine deployed overnight. GitHub repo here.

The Problem

Oracle Cloud offers a genuinely generous free tier: 4 ARM CPU cores and 24GB RAM, forever free. It’s probably the best free tier out there. There’s just one catch — good luck actually getting one.

The ARM instances (VM.Standard.A1.Flex) are so popular that you’ll almost certainly hit this error:

Error: 500-InternalError
Out of host capacity.

People report trying for days, weeks, even months to provision these instances. The manual approach is to spam the “Create Instance” button in the web console every few minutes and hope you get lucky.

The Terraform Authentication Gotcha

Before I even got to the capacity issue, I hit another roadblock. Running Terraform in OCI Cloud Shell gave me this cryptic error:

Error: can not create client, bad configuration: 
did not find a proper configuration for key id

After digging through GitHub issues and Oracle docs, I found the fix. Cloud Shell uses a special authentication profile, so your provider block needs to be:

provider "oci" {
  config_file_profile = "eu-madrid-1"  # or your region
}

Not the auth = "InstancePrincipal" that most guides suggest. This took me way too long to figure out.

The Solution

I wrote a Bash script that wraps terraform apply with smart retry logic:

#!/bin/bash

MAX_RETRIES=60
RETRY_DELAY=60

until [ "$n" -ge "$MAX_RETRIES" ]; do
    OUTPUT=$(terraform apply -auto-approve 2>&1)
    
    # Success? Exit
    if [ $? -eq 0 ]; then
        echo "SUCCESS!"
        exit 0
    fi
    
    # Only retry on capacity errors
    if echo "$OUTPUT" | grep -qi "out of capacity"; then
        echo "No capacity. Retrying in 60s..."
        sleep 60
    else
        # Config error? Stop immediately
        echo "Non-retryable error. Check your config."
        exit 1
    fi
done

The key features:

Only retries on capacity errors — stops immediately if there’s a config/auth problem
Logs everything — timestamped logs for debugging
Runs in background — use nohup to keep it running when you close your terminal
Configurable — adjust retry count and delay to your needs

Results

I started the script before bed. Woke up to a fully provisioned instance. The logs showed it took 47 attempts over 46 minutes.

[2026-02-15 02:47:30] Attempt 47/60
[2026-02-15 02:48:45] ✓ Infrastructure deployed successfully!

Why This Works

Oracle’s capacity comes and goes throughout the day as people spin up and tear down instances. The script just sits there waiting for a free slot. It’s like having a bot constantly refreshing the page for you.

Some tips:

Run it overnight — less competition from other time zones
Try different regions — some regions have more capacity than others
Be patient — it might take hours, but it works

The Code

The full script with error handling, logging, and documentation is on GitHub:

https://github.com/ekadetov/oci-terraform-retry-script

It includes:

Full-featured retry script with colored output
Simple version for minimal setups
Example Terraform configs
Complete documentation

For the HN Crowd

If you’re thinking “this is a terrible user experience,” you’re absolutely right. But the hardware you get for $0/month makes it worth the hassle:

4 ARM CPU cores (Ampere Altra)
24GB RAM
200GB storage
10TB/month bandwidth
Forever free

Compare that to AWS’s t2.micro (1 vCPU, 1GB RAM) or GCP’s e2-micro (0.25-1 vCPU, 1GB RAM). Oracle’s free tier is in a different league.

Try It Yourself

git clone https://github.com/ekadetov/oci-terraform-retry-script.git
cd oci-terraform-retry-script
chmod +x terraform_retry.sh
./terraform_retry.sh

Good luck, and may the capacity be with you.

Update: Several people have asked about arm64 Docker compatibility. Yes, these instances run arm64 natively, and Docker works great. I’m running several containers on mine with no issues.

Update 2: For those asking about alternative regions — eu-frankfurt-1 and us-ashburn-1 seem to have slightly better availability based on community reports, but YMMV.

tags: