Lambda cold/warm start issue when DNS changes region

[geralexgr]

There is an issue in the code when the DNS is switching from one region to another.

We have noticed that when there is a failover from one region to another from the loadbalancer, although the health_active_region is changing, due to the fact that codeblock is not wrapped into a function or in the function handler, the code is not in warm state and wrong results are returned. Currently the initialisation of the config is done on the cold state. As a result when one runs the lambda every x minutes when a failover occurs it will return an error.

In the function logs we notice an error message like:

Client is configured with the deprecated endpoint: us-east-2

aws-health-aware/handler.py

Lines 24 to 38 in 6f047af

	health_dns = socket.gethostbyname_ex('global.health.amazonaws.com')
	(current_endpoint, global_endpoint, ip_endpoint) = health_dns
	health_active_list = current_endpoint.split('.')
	health_active_region = health_active_list[1]
	print("current health region: ", health_active_region)
	# create a boto3 health client w/ backoff/retry
	config = Config(
	region_name=health_active_region,
	retries=dict(
	max_attempts=10 # org view apis have a lower tps than the single
	# account apis so we need to use larger
	# backoff/retry values than than the boto defaults
	)
	)

A solution for this issue could be to wrap the initialisations inside a function.