Skip to content

RISK.md

Latest commit

 

History

History
127 lines (98 loc) · 5.22 KB

RISK.md

File metadata and controls

127 lines (98 loc) · 5.22 KB
Raw

Risk matrix

Last updated 2026-06-18. Re-validate before each fresh signup.

Hours-budget model (June 2026 reality)

Critical: Always Free ARM is hours-budgeted, NOT total-capacity.

ORACLE OFFICIAL (updated 2026-06-12, applied 2026-06-15):

  1,500 OCPU-hours / month   (was 3,000)
  9,000 GB-hours  / month    (was 18,000)
  = 2 OCPU / 12 GB if run 24/7 continuously

Math:

Config            | Hours/month used | Within budget?
------------------|------------------|------------------
2 OCPU / 12 GB 24/7 | 1,440 (CPU) / 8,640 (RAM) | yes (96% CPU, 96% RAM)
2 OCPU /  6 GB ×2 24/7 | 1,440 / 8,640 | yes (split)
1 OCPU /  6 GB ×2 24/7 |   720 / 4,320 | yes (48% headroom for both)
4 OCPU / 24 GB 24/7 | 2,880 / 17,280 | NO — exceeds by 1,380 CPU / 8,280 RAM hours

Consequence: No upgrade path from 2/12 → 4/24 within Always Free tier. Crossing the cap triggers two outcomes:

  • Free-Tier account → Oracle shuts down your instance (verified by Linuxiac + Reddit thread)
  • PAYG account → $10-15 USD/month billing surprise (Reddit, Upstairs-Bread-4545)

Downgrade workflow (verified by CuriosGeekMan comment, Reddit):

Instances → Actions → More Actions → Edit
  → reduce vCPU slider (e.g. to 2)
  → reduce RAM slider (e.g. to 12)
  → Save
  → Restart instance

Charge-related risks

Risk Probability Mitigation Worst case
4/24 instance not downgraded post-cut High (PAYG users caught) Downgrade via Edit immediately $10-15/month recurring
Free-Tier instance > 2/12 Instant Downgrade (or shut down imminent) VM offline
CC charge NOT refunded Very low dispute via card issuer within 30 days $94 orphan
Boot volume pricing shows cost Confirmed craeckor verified (comment Jun 14) $0 within limits
Outbound traffic crosses paid tier Medium stay under 10 TB/mo + monitor $0.0085/GB
Multi-account detection Low 1 Oracle account per person All accts terminated

Capacity risks

Risk Probability Mitigation
Home region dead High in 2026 Auto-discovery or rotate
SG-2 / IAD capacity exhausted Medium Switch to ASH or SYD
Boot volume 200 GB not enough Low Microk8s-style split
AD-1 stuck, AD-2 OK High Script rotates automatically
Egress throttle Confirmed intermittent Tailscale routing for heavy traffic

Compute-hour exhaustion

Risk Probability Mitigation
1,440 h used before month-end Low at 2/12 24/7 monitor quota dashboard
1,440 h exceeded (e.g. burst overuse) Low stop non-essential instances
Mid-month graph showing 80% used Medium downsize before month rolls over

How to monitor:

  • Console → Governance → Limits, Quotas and Usage Filter: compute.monthly.hours Or: oci limits resource-availability get --service compute --limit compute.monthly.hours

Compliance / ToS

Risk Probability Mitigation
Mining detected Instant ban No miner Docker images
Multi-account abuse Instant ban 1 account per person
Compute-hour gaming (over-use) Slow ban Stay within hourly cap
Resource abuse (CPU/network) Slow ban Stay within 2 OCPU / 12 GB / 9,000 GB-h
Public-facing prod use Allowed But Oracle uptime flaky — don't rely on it
Selling/re-selling the box Some risk Read Oracle ToS — "personal use" preferred

Edge cases observed (rssnyder comments June 2026)

  • black1linkin — instance provisioned but no public IPv4 → use IPv6 or Tailscale
  • teddyramu — speed capped to 50 Mbps after 5 days → PAYG fix
  • marcosanson — Milan home region stuck OOC → can't change home; pick wisely at signup
  • harshjdhv — Indian card rejected → try them all
  • jabuxas — every transaction declined → check CC issuer whitelist

Mitigation summary

For each realistic charge vector this toolkit ships with:

  1. Hours-budget respected — config defaults to 2 OCPU / 12 GB (perfect fit for 1,500/9,000 envelope)
  2. PAYG-$0 default — no scripts touch outside free limits
  3. Bot stops on success — no infinite provisioning loops
  4. Tailscale recommended — minimizes public-IP exposure
  5. Trusted image advice — post-provision rejection of random Docker pulls
  6. Boot volume < 100 GB — conservative default across possible split instances
  7. Single-instance default — never spawn N VMs by accident

Burst-scaling alternative (codelesscody pattern)

If you genuinely need >2/12 for short workloads (CI/CD, batch rendering, ML training), you can:

Approach: Create a custom-shape VM larger than Always Free, run for hours
          (not full month), pay only for used hours.

Example: 12 OCPU / 64 GB VM, run 10 days, then decommission:
  CPU:   12 × 24 × 10 × $0.0105/OCPU-h ≈ $30
  RAM:   64 × 24 × 10 × $0.0015/GB-h  ≈ $23
  Total: ~$53 for the burst

Compared: Hetzner CAX31 16GB ARM €17/month flat
         - half as much RAM, slightly cheaper, no burst savings

So Oracle burst is more cost-effective only for one-shot heavy workloads. Day-to-day Always Free wins.

See docs/FALLBACKS.md F7 for full burst-scaling recipe.