Hello!
I wonder if this issue is by design or a feature/bug?
Running istio version 1.7.3.
When having a single replica of istio-ingressgateway and killing it (or if it’s being re-scheduled for whatever reason) the istiod/pilot config push results in “istio-ingressgateway.istio-system.svc.cluster.local has no endpoints”.
|2020-10-22T17:59:22.419831Z|info|ads|Incremental push, service istio-ingressgateway.istio-system.svc.cluster.local has no endpoints|
|---|---|---|---|
|2020-10-22T17:59:22.520063Z|info|ads|Push debounce stable[10] 1: 100.156086ms since last change, 100.155873ms since last push, full=false|
|2020-10-22T17:59:22.520220Z|info|ads|XDS:EDSInc Pushing:2020-10-22T17:58:42Z/3 Services:map[istio-ingressgateway.istio-system.svc.cluster.local:{}] ConnectedEndpoints:4|
|2020-10-22T17:59:23.460919Z|info|ads|Incremental push, service istio-ingressgateway.istio-system.svc.cluster.local has no endpoints|
|2020-10-22T17:59:23.562127Z|info|ads|Push debounce stable[11] 1: 100.118764ms since last change, 100.118604ms since last push, full=false|
|2020-10-22T17:59:23.562742Z|info|ads|XDS:EDSInc Pushing:2020-10-22T17:58:42Z/3 Services:map[istio-ingressgateway.istio-system.svc.cluster.local:{}] ConnectedEndpoints:4|
After this istio-ingressgateway logs are all just 503’s UH’s for 20 min after which istiod discovers the ingress gateway as a new service, does another push resulting in 200s and virtual services and pods receiving traffic again. Reproducible every time.
|2020-10-22T18:20:53.655135Z|info|ads|Full push, new service istio-ingressgateway.istio-system.svc.cluster.local|
|---|---|---|---|
|2020-10-22T18:20:53.821732Z|info|ads|Push debounce stable[3] 2: 100.126096ms since last change, 165.873549ms since last push, full=true|
|2020-10-22T18:20:53.825123Z|info|ads|XDS: Pushing:2020-10-22T18:20:53Z/1 Services:72 ConnectedEndpoints:5|
Missing endpoints is what is happening for a short while during startup, but 20 minutes downtime?