Migrate "promotion check" from deployer to release-tools
Currently release-tools triggers a deployer pipeline, and then deployer triggers a new release-tools pipeline to get the production health status to see if it can proceed. With coordinated pipelines, this adds unnecessary indirection: release-tools should just perform the check before triggering the deployer pipeline, and the deployer pipeline should proceed under the assumption it already has "permission" to do so.
However, a single production deployer pipeline performs this check at three stages:
- gprd-migrations
- gprd-gitaly
- gprd-praefect
In theory this allows an unhealthy production state to halt the deployment partway through at any of these steps, however I'm not sure we've ever had a case where we started a deployment but then that deployment failed a health check part way through. Do we even want to stop a deploy part way through in that case? Doesn't that leave the environment in a mixed state?
Proposal
Remove all production health check calls from the deployer, and perform a single health check in release-tools immediately before triggering the downstream deployer pipeline.
Side-effects
These "check" jobs also have output, which post a message in a Slack thread about the progress of the deployment:
Removing the check at each step would drop these messages, however release managers seem to find these messages not useful unless they report a failure status.