fix: Use instance as join label for redis-ops rates
Before we were using pod
and fqdn
as the join labels. This would
mean that when a pod
had the same name across environments, we'd
select series for both environments and display them on the graph.
Changing that to use the instance
label, which uses an
ip-adress:port
in the k8s case solves this.
Dashboards at a prolematic time:
This shows that redis-ratelimiting-node-2
is supposed to be the primary for pre
. While redis-ratelimiting-node-0
is the primary for gstg
at the beginning of that range. But in the before graph, both nodes show up cross-environment, because they have the same pod
label.
At the end of the graph there, the pre
instance switches ip, while the primary for gstg
switches from node-0
to node-2
.
For gitlab-com/gl-infra/scalability#1875 (comment 1140950779)