What is this

This is a factory for Alexa skills. Read this article to get some context.

Pipeline

.gitlab-ci.yml defines the delivery pipeline. In that file, update the YAML-defined global variables:

• SERVERLESS_SERVICE_NAME for serverless to keep track of this service
• QA_INVOCATION and PROD_INVOCATION for the Alexa skill
• AWS_REGION, the AWS region where the skill is to be deployed

Configure the following project-level secret variables:

• DARK_SKY_SECRET_KEY, the Dark Sky API key
• GPS_POSITIONS, GPS position of the locations you want the weather for, in a JSON document such as:

{
    "home": "29.8,-95.5",
    "volley-ball court": "29.9,-95.6"
}

CI image

The Dockerfile for the debian-ci image is provided in the debian-ci directory. It is rather bloated, as it is used for more projects than just this skills factory.

Runner

The runner tasked with the deployment has access to the AWS and ASK secrets. Initialize AWS CLI with

aws configure

and ASK with

ask init --no-browser

If more than one person contributes code, restrict the runners for example with tags

config.toml looks like this:

[[runners]]
  name = "docker-runner"
  url = "https://<FQDN_OF_OUR_GITLAB_SERVER>"
  executor = "docker"
  environment = ["GODEBUG=netdns=cgo"]
  [runners.docker]
    tls_verify = false
    image = "docker:latest"
    privileged = false
    disable_cache = false
    volumes = ["/var/run/docker.sock:/var/run/docker.sock", "/cache", "/srv/aws/aws:/home/user/.aws", "/srv/aws/ask:/home/user/.ask" ]
    links = ["<NAME_OF_GITLAB_CONTAINER>:<FQDN_OF_OUR_GITLAB_SERVER>"]
    shm_size = 0
  [runners.cache]

Our Gitlab server does not have a publicly signed certificate, so we have to ask go to trust the certificates that the containerized OS trusts (see the debian-ci Dockerfile). That's for the

environment = ["GODEBUG=netdns=cgo"]

Accessing logs

npm run logs-tail-qa (resp. npm run logs-tail-prod) tails the log stream coming from the qa (resp. prod) Lambda function.

node_modules/.bin/sls remove removes the stack, in case you had the bad idea for example to remove Lambda functions manually ("Function not found" error message). Trigger the pipeline again to put everything back.

No CloudWatch stream error message

If npm run logs-tail-qa or its production counterpart complains of lack of CloudWatch stream, make sure that there is no other active skill defined with the same invocation.