EC2 instances are volatile and can be recycled at any time without warning. Amazon recommends running them under Auto Scaling Groups to ensure overall service availability, but it's easy to forget that instances can suddenly fail until it happens in the early hours of the morning when everyone is on holiday.
Chaos Lambda increases the rate at which these failures occur during business hours, helping teams to build services that handle them gracefully.
Create the lambda function in the region you want it to target using the
cloudformation/templates/lambda_standalone.json
CloudFormation template.
There are two parameters you may want to change:
Schedule
: change if the default run times don't suit you (once per hour between 10am UTC and 4pm UTC, Monday to Friday); see http://docs.aws.amazon.com/AmazonCloudWatch/latest/events/ScheduledEvents.html for documentation on the syntax.DefaultProbability
: by default all Auto Scaling Groups in the region are targets; set this to0.0
and only ASGs with achaos-lambda-termination
tag (see below) will be affected.
By deploying the lambda_standalone.json
CloudFormation template, an SNS topic
will be created with the name ChaosLambdaTerminationTopic
. For each instance
that gets terminated, a notification will be published using this structure:
{
"event_name": "chaos_lambda.terminating",
"asg_name": "my-autoscaling-group",
"instance_id": "i-00001234"
}
By default, no subscriptions are created to this topic, so it is up to you to subscribe a queue or another lambda if you wish.
To receive notifications if the lambda function fails for any reason, create
another stack using the cloudformation/templates/alarms.json
template. This
takes the lambda function name (something similar to
chaos-lambda-ChaosLambdaFunction-EM2XNWWNZTPW
) and the email address to
send the alerts to.
Every time the lambda triggers it examines all the Auto Scaling Groups in the
region and potentially terminates one instance in each. The probability of
termination can be changed at the ASG level with a tag, and at a global level
with the DefaultProbability
stack parameter.
At the ASG level the probability can be controlled by adding a
chaos-lambda-termination
tag with a value between 0.0
(never terminate) and
1.0
(always terminate). Typically this would be used to opt out a legacy
system (0.0
).
The DefaultProbability
parameter sets the probability of termination for any
ASG without a valid chaos-lambda-termination
tag. If set to 0.0
the
system becomes "opt-in", where any ASG without this tag is ignored. The
default is 0.166
(or 1 in 6).
The lambda is triggered by a CloudWatch Events rule, the name of which can be
found from the ChaosLambdaFunctionOutput
output of the lambda stack. Locate
this rule in the AWS console under the Rules section of the CloudWatch service,
and you can disable or enable it via the Actions
button.
By default the lambda will target ASGs running in the same region. It's
generally a good idea to avoid cross-region actions, but if necessary an
alternative list of one or more region names can be specified in the Regions
stack parameter.
The value is a comma separated list of region names with optional whitespace, so the following are all valid and equivalent:
ap-south-1,eu-west-1,us-east-1
ap-south-1, eu-west-1, us-east-1
ap-south-1 , eu-west-1 , us-east-1
Chaos Lambda log lines always start with a timestamp and a word specifying the
event type. The timestamp is of the form YYYY-MM-DDThh:mm:ssZ
, eg
2015-12-11T14:00:37Z
, and the timezone will always be Z
. The different
event types are described below.
<timestamp> bad-probability [<value>] in <asg name>
Example:
2015-12-11T14:07:21Z bad-probability [not often] in test-app-ASG-7LJI5SY4VX6T
If the value of the chaos-lambda-termination
tag isn't a number between 0.0
and 1.0
inclusive then it will be logged in one of these lines. The square
brackets around the value allow CloudWatch Logs to find the full value even if
it contains spaces.
<timestamp> result <instance id> is <state>
Example:
2015-12-11T14:00:40Z result i-fe705d77 is shutting-down
After asking EC2 to terminate each of the targeted instances the new state of
each is logged with a result
line. The <state>
value is taken from the
code
property of the InstanceState
AWS type described at
http://docs.aws.amazon.com/AWSEC2/latest/APIReference/API_InstanceState.html
<timestamp> targeting <instance id> in <asg name>
Example:
2015-12-11T14:00:38Z targeting i-168f9eaf in test-app-ASG-1LOMEKEVBXXXS
The targeting
lines list all of the instances that are about to be
terminated, before the TerminateInstances
call occurs.
<timestamp> triggered <region>
Example:
2015-12-11T14:00:37Z triggered eu-west-1
Generated when the lambda is triggered, indicating the region that will be affected.