ECS container IO stress

ECS container IO stress disrupts the state of infrastructure resources. It induces stress on the AWS ECS container using Amazon SSM Run command, which is carried out using SSM docs which is in-built into the fault.

It causes I/O stress on the containers of the ECS task using the given CLUSTER_NAME environment variable for a specific duration.
To select the Task Under Chaos (TUC), use the service name associated with the task. If you provide the service name along with the cluster name, all the tasks associated with the given service will be selected as chaos targets.
It tests the ECS task sanity (service availability) and recovery of the task containers subject to I/O stress.
This experiment induces chaos within a container and depends on an EC2 instance. Typically, these are prefixed with "ECS container" and involve direct interaction with the EC2 instances hosting the ECS containers.

ECS Container IO Stress

Use cases

ECS container IO stress:

Determines how a container recovers from a memory exhaustion.
File system read and write evicts the application (task container) and impacts its delivery. These issues are also known as noisy-neighbour problems.
Injecting a rogue process into a target container starves the main microservice process (typically pid 1) of the resources allocated to it (where the limits are defined). This slows down the application traffic or exhausts the resources leading to eviction of all task containers.

Prerequisites

Kubernetes >= 1.17
ECS container metadata is enabled (disabled by default). To enable it, refer to this docs. If your task is running from before, you may need to restart it to get the metadata directory.
You and the ECS cluster instances have a role with the required AWS access to perform the SSM and ECS operations. Refer to systems manager docs.
Create a Kubernetes secret that has the AWS access configuration(key) in the CHAOS_NAMESPACE. Below is a sample secret file:

apiVersion: v1
kind: Secret
metadata:
  name: cloud-secret
type: Opaque
stringData:
  cloud_config.yml: |-
    # Add the cloud AWS credentials respectively
    [default]
    aws_access_key_id = XXXXXXXXXXXXXXXXXXX
    aws_secret_access_key = XXXXXXXXXXXXXXX

tip

HCE recommends that you use the same secret name, that is, cloud-secret. Otherwise, you will need to update the AWS_SHARED_CREDENTIALS_FILE environment variable in the fault template with the new secret name and you won't be able to use the default health check probes.

Below is an example AWS policy to execute the fault.

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Sid": "VisualEditor0",
            "Effect": "Allow",
            "Action": [
                "ecs:UpdateContainerInstancesState",
                "ecs:RegisterContainerInstance",
                "ecs:ListContainerInstances",
                "ecs:DeregisterContainerInstance",
                "ecs:DescribeContainerInstances",
                "ecs:ListTasks",
                "ecs:DescribeClusters"

            ],
            "Resource": "*"
        },
        {
            "Effect": "Allow",
            "Action": [
                "ssm:GetDocument",
                "ssm:DescribeDocument",
                "ssm:GetParameter",
                "ssm:GetParameters",
                "ssm:SendCommand",
                "ssm:CancelCommand",
                "ssm:CreateDocument",
                "ssm:DeleteDocument",
                "ssm:GetCommandInvocation",
                "ssm:UpdateInstanceInformation",
                "ssm:DescribeInstanceInformation"
            ],
            "Resource": "*"
        },
        {
            "Effect": "Allow",
            "Action": [
                "ec2messages:AcknowledgeMessage",
                "ec2messages:DeleteMessage",
                "ec2messages:FailMessage",
                "ec2messages:GetEndpoint",
                "ec2messages:GetMessages",
                "ec2messages:SendReply"
            ],
            "Resource": "*"
        },
        {
            "Effect": "Allow",
            "Action": [
                "ec2:DescribeInstances"
            ],
            "Resource": [
                "*"
            ]
        }
    ]
}

note

The ECS container instance should be in a healthy state.
It is recommended to use the same secret name, i.e. cloud-secret. Otherwise, you will need to update the AWS_SHARED_CREDENTIALS_FILE environment variable in the fault template and you may be unable to use the default health check probes.
Refer to the superset permission/policy to execute all AWS faults.
Refer to AWS named profile for chaos to know how to use a different profile for AWS faults.
Refer to the common attributes and AWS-specific tunables to tune the common tunables for all faults and aws specific tunables.

Mandatory tunables

Variables	Description	Notes
CLUSTER_NAME	Name of the target ECS cluster.	For example, `cluster-1`.
REGION	Region name of the target ECS cluster	For example, `us-east-1`.

Optional tunables

Variables	Description	Notes
TOTAL_CHAOS_DURATION	Duration that you specify, through which chaos is injected into the target resource (in seconds).	Default: 30 s. For more information, go to duration of the chaos.
CHAOS_INTERVAL	Interval between successive instance terminations (in seconds).	Default: 30 s. For more information, go to chaos interval.
AWS_SHARED_CREDENTIALS_FILE	Path to the AWS secret credentials.	Defaults to `/tmp/cloud_config.yml`.
SERVICE_NAME	Target ECS service name.	For example, `app-svc`. For more information, go to ECS service name.
FILESYSTEM_UTILIZATION_BYTES	Memory consumed during I/O stress (in GB).	Default: 1. For more information, go to file system utilization in bytes.
FILESYSTEM_UTILIZATION_PERCENTAGE	Memory consumed during I/O stress (in percentage).	Default: 10. For more information, go to file system utilization in percentage.
VOLUME_MOUNT_PATH	Location that points to the volume mount path used in I/O stress.	Default: `/tmp`. For more information, go to volume mount path.
NUMBER_OF_WORKERS	Number of workers for memory stress.	Default: 1. For more information, go to workers.
SEQUENCE	Sequence of chaos execution for multiple instances	Default: parallel. Supports serial and parallel. For more information, go to sequence of chaos execution.
RAMP_TIME	Period to wait before and after injecting chaos (in seconds).	For example, 30 s. For more information, go to ramp time.

File system utilization percentage

Amount of free space available in the ECS container (in percentage). Tune it by using the FILESYSTEM_UTILIZATION_PERCENTAGE environment variable.