Initial investigation (HA rolling pipeline deploy)

[allthedoll]

Description

Stub ticket.

Epic/Story

No response

Have you provided an initial effort estimate for this issue?

I have provided an initial effort estimate

[hardillb]

Quick look at the code path for a deploy to an "instance" shows the following:

• /pipelines/:pipelineId/stages/:stageId/deploy calls
• app.db.controllers.Pipeline.deploySnapshotToInstance calls
• app.containers.restartFlows()

This means that the actual restart is done by the container driver.

It currently grabs all the endpoints for the instance and sends a restart command directly to the nr-launcher instance. In HA case this is the Pod for both replicas.

So the only place things need to change is in the kube driver in the restartFlows() function.

This will need to notice that we are in HA mode (more than 1 endpoint ip address). And then trigger a restart on the first and then poll for it to finish before restarting the second.

We should do something with K8s ready flags to stop the ingress controller routing to the instance that is restarting.

[knolleary]

Research done - separate task for implementation raised #6606