[15:47:01] How does one investigate what looks to be envoy errors when accessing an http service hosted on k8s? I'm *sometimes* seeing "upstream connect error or disconnect/reset before headers. retried and the latest reset reason: connection failure, transport failure reason: delayed connect error: 111" when accessing https://airflow-test-k8s.wikimedia.org. [15:47:02] Thanks [15:54:53] well, s/sometimes/most of the time now/ [15:59:07] brouberol: did you inspect the sidecar's logs to see if anything pops up? [15:59:38] another thing that I'd try is to use nsenter -t $PID -n $command on the host where the sidecar container runs [16:00:06] it should be easier to reproduce the error, and/or to see if there is a problem [16:01:25] when I see the error in the UI, no logs are generated in the envoy tls proxy sidecar [16:01:41] so the error might come from the ingress gateway? [16:11:43] I was indeed able to capture a log in istio-ingressgateway-4xr9n: https://phabricator.wikimedia.org/P70595 [16:15:44] the error codes UF,URX seem to indicate an upstream failure. Inspecting the airflow-production-tls-service endpoints gave me the answer: the service was resolving to pods created by airflow as well 🤦‍♂️ [16:16:20] the service selector is much too broad [16:17:43] aand fixed [16:40:27] brouberol: ahhh sorry I thought it was an error accessing airflow-test, my bad [16:40:46] no worries, you were a perfect 🦆 [16:40:47] from a pod namely [16:41:13] super :) [16:41:53] and speaking of: we currently don't have any way to specify an extra set of k/v label selectors for the mesh service. In the case of airflow, we have 4 permanent pods and many transient ones, and we'd need the tls-service to resolve to a single pod (the one labeled `component: webserver`). I could do that by specifying an extra label selector on [16:41:53] top of the `chart: x, routed_via: y`. Would it make sense to y'all to add that to the vendored modules? [16:43:00] and that extra set of label selectors could be scaffolded to {} by default [16:45:10] (I'm not sure if I'm very clear)