Replace Cloud Functions with Cloud Run to avoid permadiffs

[nacnudus]

Permadiffs cause terraform apply to redeploy cloud functions unnecessarily, and it is currently also unbinding service accounts from IAM permissions, which has caused the overnight batch of publishing-api to fail. It also seems to interfere with the google_bigquery_job that registers the remote function in BigQuery.

The best I can do to prevent the permadiffs is to copy an MD5 hash from a terraform plan into hard-coded config, and then run terraform apply. Certain fields must also be ignored, which gives to rise to an incorrect warning. Besides being a pain, this is hardly any easier than manually redeploying a function.

resource "google_storage_bucket_object" "parse_html" {
  name   = "sourcecode.zip"
  bucket = google_storage_bucket.cloud_functions.name
  source = data.archive_file.parse_html.output_path # Add path to the zipped function source code
  detect_md5hash = "0TPZXY07IZeI84RrW+VH2w==" # Keep up to date https://support.hashicorp.com/hc/en-us/articles/28945898467347-How-to-Avoid-Terraform-Apply-from-Forcefully-Recreating-Google-Cloud-Storage-Object
  lifecycle {
    # To prevent unnecessary updates, which will trigger a replacement of the
    # cloud function, we must ignore changes to certain attributes. Terraform
    # will give the following warning, which is a lie.
    #
    #   "The attribute crc32c is decided by the provider alone and therefore
    #   there can be no configured value to compare with. Including this
    #   attribute in ignore_changes has no effect. Remove the attribute from
    #   ignore_changes to quiet this warning."
    #
    # Including these attributes in ignore_changes _does_ have an effect, which
    # is to prevent unnecessary updates.
    ignore_changes = [crc32c, generation, md5hash]
  }
}

Now that Cloud Functions are a wrapper around Cloud Run, there would be no functional difference if we were to refactor the Cloud Functions as Cloud Run instances directly. The Dockerfile would make testing easier, and would be offset by being able to delete the double-handling in Terraform of creating both a Cloud Run instance and a Cloud Function wrapper. An existing Cloud Run instance to use as a pattern is terraform/html-to-text.tf and the corresponding source code docker/html-to-text.