Question about node taints with regard to doks-managed 'coredns' deployment

[digitalocean]

Question about node taints with regard to doks-managed 'coredns' deployment

[timoreimann]

Hey 👋

Most of the managed workloads that are deployed into the data plane are critical Daemonsets that must or should always run to provide core functionality. CoreDNS is in a bit of a mixed state, in the sense that it does provide core functionality but that it must also run on a worker node that is considered healthy (and that it should be moved / evicted to one should its hosting node become unhealthy).

I think we haven't revisited the current tolerations in a while, so there's possibly an opportunity to improve here. That said, I'd be hesitant to give it a blank toleration since, for instance, we wouldn't want CoreDNS to continue to run on a node that's under memory pressure.

There have also been requests to support extending the list of tolerations on CoreDNS by customers to make it better suit custom taints associated to node pools (as you did). This is something we're also considering to do.

With that background shared, I'd be curious to hear what people's preferences are (and why if non-obvious). This would help us plan for the right, next move in this context.