EC2 stop by tag

Stop EC2 instances selected by tag for a configurable duration so you can test how the workload running on those instances behaves when a tagged subset disappears.

EC2 stop by tag is an AWS chaos fault that resolves EC2 instances by tag, stops a configurable percentage of them for a configurable duration, and then starts them again. The fault is useful when you want to act on a logical group of instances (an environment, a service, a role) without pinning to specific instance IDs.

Use this fault to test how a workload behaves when a tagged subset of its capacity disappears: replicas served by other instances pick up the load, an auto-scaling group provisions replacements, or a tier behind a load balancer is taken out of rotation.

RUN YOUR FIRST EXPERIMENT

If you have not configured the chaos infrastructure yet, go to Quickstart to install the chaos infrastructure and run an experiment end to end.

Use cases

Run this fault when you want to answer concrete questions like:

Replica failover for a tagged tier: When a percentage of instances tagged for a specific service are stopped, do remaining replicas absorb the traffic without dropping requests?
Load balancer health checks: Does the target group detach stopped instances quickly and reattach them cleanly when they return?
Auto-scaling group response: Do ASGs provision replacements within the expected window?
Managed node group resilience (EKS): When tagged worker nodes are stopped, are pods rescheduled across the cluster and does capacity recover?
Blast-radius validation: Confirm that targeting by tag plus INSTANCE_AFFECTED_PERCENTAGE keeps the impact within the planned blast radius.

Prerequisites

Kubernetes version: 1.21 or later for the chaos infrastructure cluster.
Target instances exist and are reachable: At least one instance in REGION carries the tag in EC2_INSTANCE_TAG and is in the running state at the start of the experiment.
AWS credentials available: Either an AWS credentials file uploaded as a File Secret in Harness Secret Manager (see Authentication below) or an IAM role for service accounts (IRSA) bound to the chaos infrastructure service account.
IAM permissions granted: The credentials or role include the permissions listed below.
ASG awareness (only when targeting nodes in a managed node group): If targets are EKS managed node group instances, set MANAGED_NODEGROUP=enable so the fault waits for replacement nodes instead of attempting to start the originals.

Supported environments

Platform

Support status

Amazon EC2 (standalone instances)

Supported

Amazon EKS managed worker nodes

Supported (set MANAGED_NODEGROUP=enable)

Amazon EKS self-managed worker nodes

Supported

AWS regions

Supported in every commercial region; pass the region in REGION

Linux instances

Supported

Windows instances

Supported

Spot instances

Supported (the start call may fail if capacity is unavailable; see Limitations)

Instances behind ALB, NLB, or CLB

Supported

Permissions required

The IAM principal that the chaos pod uses needs the following AWS actions.

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "ec2:StartInstances",
        "ec2:StopInstances",
        "ec2:DescribeInstances",
        "ec2:DescribeInstanceStatus"
      ],
      "Resource": "*"
    },
    {
      "Effect": "Allow",
      "Action": [
        "autoscaling:DescribeAutoScalingInstances"
      ],
      "Resource": "*"
    }
  ]
}

ec2:StopInstances and ec2:StartInstances drive the fault.
ec2:DescribeInstances is used to resolve EC2_INSTANCE_TAG to a list of instance IDs and to confirm state transitions.
ec2:DescribeInstanceStatus confirms each instance reaches stopped and running states.
autoscaling:DescribeAutoScalingInstances is only required when MANAGED_NODEGROUP=enable.

Go to common policy for all AWS faults to use a single superset IAM policy across every AWS fault.

Authentication

The fault supports three credential delivery models. Pick one based on how your chaos infrastructure is deployed.

Method

When to use it

How to configure

Harness Secret Manager file secret

Chaos infrastructure runs outside EKS, or you want explicit static credentials

Upload the AWS credentials file as a File Secret in Harness Secret Manager and reference its identifier via AWS_AUTHENTICATION_SECRET

IAM Roles for Service Accounts (IRSA)

Chaos infrastructure runs in EKS and uses an OIDC-bound service account

No tunable changes; the chaos pod inherits the role automatically. Go to AWS IAM integration to set it up

Assume role

The fault needs to act in a different account or with elevated permissions

Set ASSUME_ROLE_ARN to the role ARN; the chaos pod assumes the role on top of its base credentials

When using the Harness Secret Manager method, the File Secret should contain an AWS credentials file in the standard ~/.aws/credentials format:

[default]
aws_access_key_id = REPLACE_WITH_ACCESS_KEY_ID
aws_secret_access_key = REPLACE_WITH_SECRET_ACCESS_KEY

Upload this file as a File Secret in Harness Secret Manager (Project Setup → Secrets → New File Secret), and pass the secret identifier in AWS_AUTHENTICATION_SECRET.

Fault tunables

Configure the following fault parameters when you add EC2 stop by tag to an experiment in Chaos Studio. Defaults are shown for reference.

Required parameters

Tunable

Description

Default

EC2_INSTANCE_TAG

Tag selector in key:value format (for example app:frontend).

(required)

REGION

AWS region that hosts the target instances (for example us-east-1).

(required)

INSTANCE_AFFECTED_PERCENTAGE

Percentage of matching instances to stop. 0 targets exactly one instance.

0

Chaos parameters

Tunable

Description

Default

TOTAL_CHAOS_DURATION

Duration of the fault in seconds.

60

CHAOS_INTERVAL

Delay in seconds between successive stop iterations.

60

SEQUENCE

Order in which selected instances are stopped: parallel stops them at once; serial stops them one at a time.

parallel

MANAGED_NODEGROUP

Set to enable when targets are EKS managed node group instances; the fault then waits for replacement nodes from the ASG.

disable

RAMP_TIME

Wait period in seconds before and after the fault. Go to ramp time to read how it is applied.

0

Authentication

Tunable

Description

Default

ASSUME_ROLE_ARN

ARN of an IAM role to assume on top of the base credentials.

""

AWS_AUTHENTICATION_SECRET

Identifier of the File Secret in Harness Secret Manager that contains the AWS credentials file. Not required when using IRSA.

""

Tunables that apply to every fault are documented in common tunables for all faults. AWS-specific shared tunables are documented in common AWS fault tunables.

DRAIN MANAGED NODE GROUP INSTANCES BEFORE THE FAULT

If targets are part of a managed node group, drain and cordon them before running the fault so that workload pods are not scheduled onto instances about to disappear.

Fault execution in brief

Resolves the tag in EC2_INSTANCE_TAG to a list of instance IDs in REGION, calls StopInstances on the chosen percentage (INSTANCE_AFFECTED_PERCENTAGE), waits for TOTAL_CHAOS_DURATION seconds, then calls StartInstances to bring them back (unless MANAGED_NODEGROUP=enable).

Expected behavior during fault execution

Each selected instance transitions from running to stopping to stopped within seconds of the stop call.
Instances behind a load balancer are marked unhealthy and removed from rotation according to target-group health-check settings.
Workloads on stopped instances become unreachable; replicas elsewhere should absorb the traffic if the workload is horizontally scaled.
For EKS worker nodes, pods enter Terminating then Pending; the scheduler places them on other nodes if capacity allows.
When MANAGED_NODEGROUP=enable, the ASG launches replacement nodes and the fault waits for them to join the cluster before reporting success.

WHEN THE FAULT ENDS

The chaos pod issues StartInstances and polls until each target reaches running. Long-lived TCP connections dropped during the outage need to be re-established by the caller.

Signals to watch

Attach resilience probes to assert each layer:

Application availability: Use an HTTP probe to detect downtime and measure recovery time.
Target group health: Use a Prometheus probe on aws_applicationelb_healthy_host_count to confirm the load balancer detected the outage.
EC2 instance state: Use a command probe running aws ec2 describe-instances --filters "Name=tag:<key>,Values=<value>" to assert the expected state.
CloudWatch metrics: Use a Prometheus probe on aws_ec2_status_check_failed to confirm instances went down and came back.

Verify the fault execution effect

While the experiment is running, confirm the tagged instances are stopped and then restored:

List instances by tag.

aws ec2 describe-instances \
  --region <region> \
  --filters "Name=tag:<key>,Values=<value>" \
  --query "Reservations[].Instances[].[InstanceId,State.Name]"

The expected fraction of instances should report stopping or stopped during the fault, then running after recovery.

Check load balancer target health (if applicable).

aws elbv2 describe-target-health \
  --region <region> \
  --target-group-arn <target-group-arn>

For EKS managed node groups, confirm replacements joined.
```
kubectl get nodes -o wide
```

Recovery and cleanup

End of duration: The chaos pod issues StartInstances for affected instances (unless MANAGED_NODEGROUP=enable) and polls until each reaches running.
Abort the experiment: Stopping the experiment from Chaos Studio triggers the same start call.
Manual recovery: If the fault exits with an error before issuing the start call, start affected instances manually with aws ec2 start-instances --instance-ids <ids>.

Limitations

Tag must resolve to at least one instance: If EC2_INSTANCE_TAG matches nothing in REGION, the fault fails fast with no impact.
Spot instance start failures: Start calls may fail with InsufficientInstanceCapacity; recover via your autoscaling configuration.
Stop protection: Instances with disableApiStop enabled will reject the stop call.
Instance store data loss: EC2 instance-store volumes are destroyed when an instance stops.
Cross-region targeting: One experiment targets one region.

Troubleshooting

EC2 stop by tag experiment fails with UnauthorizedOperation in Harness Chaos Engineering

The credentials supplied to the chaos pod do not have the required EC2 permissions in the target region. Confirm the IAM policy attached to the user, role, or IRSA service account includes ec2:StopInstances, ec2:StartInstances, ec2:DescribeInstances, and ec2:DescribeInstanceStatus, and that the policy condition (if any) allows the target region. When using ASSUME_ROLE_ARN, also confirm the source identity is trusted to assume the role.

EC2 stop by tag selects zero instances

The most common causes are: EC2_INSTANCE_TAG is misformatted (correct form is key:value); the tag key or value has a typo; the tag exists in a different region than REGION; or all matching instances were already stopped before the experiment started. Verify with aws ec2 describe-instances --filters 'Name=tag:,Values='.

EC2 stop by tag experiment hangs waiting for instances to start

The most common causes are: targets are spot instances and AWS could not reclaim capacity (InsufficientInstanceCapacity); targets have stop protection (disableApiStop) and the prior stop call failed silently; or MANAGED_NODEGROUP was not set to enable for ASG-managed instances. Inspect chaos pod logs and EC2 console events, then either start instances manually or rerun with MANAGED_NODEGROUP=enable.

EC2 stop by ID: Stop EC2 instances selected by instance ID instead of by tag.
EC2 process kill: Kill a specific process inside an EC2 instance without stopping the host.
SSM chaos by tag: Inject CPU, memory, or IO stress inside tagged EC2 instances via AWS Systems Manager.
Common AWS fault tunables: Shared environment variables for AWS faults.

Last updated 5 days ago

Was this helpful?

Use cases

Prerequisites

Supported environments

Permissions required

Authentication

Fault tunables

Fault execution in brief

Expected behavior during fault execution

Signals to watch

Verify the fault execution effect

Recovery and cleanup

Limitations

Troubleshooting

Related faults