This is a simple HTTP-based MySQL replica health check, consisting of a bash script, a systemd socket and a companion service. The check not only shows whether the database service is running or not, but also how much replication is lagging behind.
HTTP as the protocol for this health check is used to provide rich status information beyond simple connectivity checks, allowing validation of replication lag.
The health check is implemented using systemd socket activation, to provide a simple way to serve HTTP requests. This approach conserves system resources by starting the service only when needed, while systemd handles all low-level socket management. For some background on socket activation, see:
As prerequisites, you need Hashicorp Packer and QEMU/KVM. This will allow you to build the image from the latest Debian locally:
packer build -var headless=false mysql.pkr.hcl
To base the build on a specific Debian version and name the resulting image like in the Github Actions pipeline, adapt the env vars in .env and run:
source .env
packer build -var headless=false $(scripts/create-tag.sh) .
For building in GCP, the source block can then be replaced with something like this:
source "googlecompute" "debian" {
project_id = "your-project-id"
zone = "your-zone"
subnetwork = "default"
source_image = "debian-12-genericcloud-amd64"
image_name = "your-image-name"
image_family = "your-image-family"
ssh_username = "builder"
}
Set up an application load balancer to regularly poll port 9876 of the MySQL Replica instance group to determine their health based on the returned HTTP status code.
It is possible to create a flag file to signal maintenance mode, e.g. when a replica needs to be replaced or updated: touch /var/tmp/replica-maintenance.