EC2 Lifecycle and Troubleshooting

Manage states and scaling, and troubleshoot connection or status-check failures.

States and effects

A common flow is pending → running → stopping → stopped. Termination goes through shutting-down → terminated and cannot be undone. Stop preserves EBS but loses Instance Store data; root EBS deletion on termination follows DeleteOnTermination.

flowchart LR
  P["pending<br/>booting"] --> R["running<br/>billed"]
  R --> SP["stopping"]
  SP --> S["stopped<br/>EBS kept, no compute bill"]
  S --> P
  R --> SD["shutting-down"]
  SD --> T["terminated<br/>cannot undo"]
  %% step P: The instance is placed on a host and boots.
  %% step R: Running instances are billed per second (Linux, after the first minute).
  %% step SP: A stop shuts the OS down cleanly.
  %% step S: Stopped instances keep their EBS volumes, which are still billed; instance store data is gone.
  %% step SD: Termination deletes the instance and, by default, its root volume.
  %% step T: A terminated instance cannot be started again.

Action	Instance store	EBS root	Public IP	Elastic IP	Host
Reboot	Kept	Kept	Kept	Kept	Same
Stop / start	Lost	Kept	New one	Kept	Usually new
Hibernate	Lost	Kept (RAM saved to it)	New one	Kept	Usually new
Terminate	Lost	Deleted if DeleteOnTermination	Released	Disassociated	—

Turn on termination protection for important instances so an accidental terminate call fails.

Scaling

Vertical scaling changes the instance type and normally requires a stop. Horizontal scaling changes the number of instances, with a Load Balancer and Auto Scaling — the cloud-native way to add capacity without downtime.

Troubleshoot by layer

SSH times out (no answer at all) — a network problem: 1. Is the instance running and are status checks passing? 2. Does it have a public IP, and does its subnet route 0.0.0.0/0 to an IGW? 3. Does the Security Group allow port 22 from your IP? 4. Does the Network ACL allow 22 inbound and ephemeral ports outbound?

Permission denied (publickey) — you reached sshd: check the username, the key file, and its 400 mode.

Status checks

System status check — AWS infrastructure: host hardware, power, network. Fix by stop and start, which moves the instance to a new host (a reboot does not).
Instance status check — the guest: OS boot, kernel, full disk, wrong network config. Read the system log or take an instance screenshot, then fix the OS.
Attached EBS status check — problems reaching an attached volume.

A CloudWatch alarm on StatusCheckFailed_System can recover an instance automatically onto new hardware, keeping its ID, IPs and EBS volumes.

Command	Purpose
`aws ec2 stop-instances --instance-ids i-0123456789abcdef0`	Stop an instance
`aws ec2 start-instances --instance-ids i-0123456789abcdef0`	Start it again
`aws ec2 describe-instance-status --include-all-instances`	View status checks
`aws ec2 get-console-output --instance-id i-0123456789abcdef0`	Read the boot log

Traps: rebooting to fix a failed system check; expecting the same public IP after stop/start; treating a timeout as a key problem.