0008 — Terminate EC2 35.172.106.96

• Status: Proposed
• Date: 2026-04-17
• Deciders: Fede

Context

EC2 35.172.106.96 currently runs:

• property-mgmt PM2 app (the old Next.js prod server, redundant with Vercel)
• propflowai-stage PM2 app (see ADR-0007)
• Six EC2 setInterval background processors (see ADR-0006)

After ADR-0006 and ADR-0007, this host runs nothing.

Decision

Terminate EC2 35.172.106.96 once all migrations land and prod has been stable for 72 hours.

Sequence:

1. Confirm no processors running (pm2 list empty of PropFlow apps)
2. Confirm all scheduled jobs firing in EventBridge with green metrics for 72 hours
3. Stop the instance (reversible). Wait 7 days.
4. If nothing complains, terminate. Keep the EBS snapshot for 30 days.
5. Remove IAM roles, security groups, hardcoded references in deploy scripts.

EC2 34.201.51.133 (browser agent) is not affected — see ADR-0009.

Consequences

Easier

• One less EC2 to maintain, patch, pay for, SSH into during incidents.
• Eliminates SSH-dependent deploy paths entirely (Vercel doesn't need SSH, Lambda doesn't need SSH, browser agent keeps its own).
• Removes the "mobile carrier blocking port 22 during an incident" failure mode (April 6 learnings).

Harder

• If a regression requires old EC2 logic, restore from EBS snapshot. Unlikely; 7-day stopped-instance grace period is the safety net.

Follow-up work

• Update CLAUDE.md to remove EC2 references.
• Remove EC2 IP from IP allowlists (Twilio).
• Update Sentry tags.

Alternatives considered

Keep EC2 as warm standby. Rejected — a standby not exercised is a cold failure waiting. If we ever need EC2 again, stand up a fresh one with Terraform.

Downsize to t4g.nano. Considered if the 72-hour observation turns up anything unexpected. Default is terminate.