Adding a sidecar to session pods via job-json-overrides prevents Workbench from connecting to session

[d-m]

We run node_exporter in user sessions to get additional metrics on session read/write behavior. We currently install it in each session container but would like to run it as a sidecar instead to simplify our images. When I added the following to the job-json-overrides section of the config:

config:
  profiles:
    launcher.kubernetes.profiles.conf:
      "*":
        job-json-overrides:
          - target: /spec/template/spec/containers/1
            json:
              image: "quay.io/prometheus/node-exporter:v1.3.1"
              name: "node-exporter"
              command:
                - node_exporter
                - --collector.disable-defaults
                - --collector.mountstats
                - --web.disable-exporter-metrics
            name: nodeExporterSidecar

the sidecar is created and the pod starts. However, Workbench is unable to connect to the pod and reports the following error in /var/log/rstudio/launcher/rstudio-kubernetes-launcher.log:

2022-01-19T01:24:36.087036Z [rstudio-kubernetes-launcher] ERROR system error 71 (Protocol error) [description: watchPodLogs: received non-streaming response: Bad Request - {"kind":"Status","apiVersion":"v1","metadata":{},"status":"Failure","message":"a container name must be specified for pod session-34051cfc78a31b207e05e-USERNAME---rstudio-sessio--1-vblzj, choose one of: [5964456645756e31523730702b4b57455844483063773d3d node-exporter]","reason":"BadRequest","code":400}|||]; OCCURRED AT rstudio::job_launcher::impls::kubernetes::KubernetesApi::watchPodLogs(const string&, const OnLogOutputHandler&, const ErrorHandler&)::<lambda(const rstudio::core::http::Response&)> src/cpp/job_launcher/impls/kubernetes/KubernetesApi.cpp:1092; LOGGED FROM: rstudio::job_launcher::impls::kubernetes::KubernetesOutputStream::init()::<lambda(const rstudio::core::json::Array&)>::<lambda(const rstudio::core::Error&)> src/cpp/job_launcher/impls/kubernetes/KubernetesOutputStream.cpp:85

It looks like the Kubernetes Plugin assumes that there will only be one container per session pod. There's no issue with the Helm chart, so If this isn't the right venue for this error please let me know and I'll report it via support and our sales rep.

[colearendt]

Thanks for reporting this!! I believe the new version of RStudio Workbench should resolve this issue. We will aim to have that built into a container and integrated into the helm chart next week 😄

[colearendt]

Next week became months... wowzers. In any case, we are nearing having the template integration built into the helm chart. Per #161 , if you have any examples of job-json-overrides that you are using and are able to share, it'd be great to ensure that we have commonly used values taken care of when we ship it!