@rgrandl pointed out that main components were failing to push traces to jaeger, causing them to crash. I did some debugging and found that the DNS policy we were using was preventing pods from resolving service names. I have no idea why.
This PR fixes the bug by switching to the ClusterFirst DNS policy. Quite counterintuitively, ClusterFirst is the default policy, not Default.
@rgrandl, @spetrovic77 do you understand Kubernetes well enough to understand what was going wrong? I'm having a hard time understanding the different DNS policies.
@rgrandl pointed out that main components were failing to push traces to jaeger, causing them to crash. I did some debugging and found that the DNS policy we were using was preventing pods from resolving service names. I have no idea why.
This PR fixes the bug by switching to the
ClusterFirst
DNS policy. Quite counterintuitively,ClusterFirst
is the default policy, notDefault
.