corang
commented
6 months ago Added a separate managed node group to the eks cluster, tainted it, and forced gitaly to schedule onto it with guaranteed QOS of 15cpu and 120gb cpu. Also modified the gp3 storage class to always allocate max iops and throughput for reasonable performance at scale across the cluster
As GitLab is configured now it will likely run out of memory and crash with any significant load. We need to look into configuring gitlab to handle a larger load more consistently. Mainly increasing resource limits on pods and probably more crucially allocating gitaly to it's own k8s node with lots of memory