nais / console-backend Goto Github PK

For console, we should be able to track some key metrics such as:

• internal errors
• failed external requests
• request rate
  +++

Add support for additional clusters using provided sets of:

a) cluster name
b) apiserver host
c) bearer-token with read-access to required resources

These clusters might be excluded from certain functionality such as billing in the future.

Currently we are providing console-backend with a complete kubeconfig file containing the different clusters it should connect to in order to get information about the workloads.

To avoid this manual process, we can provide console-backend with a list of Google projects. When the application starts, we make requests to the Google API to discover the required information such as apiserver IP, certificate and name.

To make console-backend able to do these requests, we need to make some additional changes to nais-terraform-modules

We need:

• Service account for console-backend
• With workload identity, connecting it to the nais-system/console-backend k8s user
• IAM permissions to get cluster info, and access to kubernetes clusters

Vi samler alt av billingdata fra både gcp og aiven i en BQ.
Vebjørn har allerede laget dashboards som gir et godt totalbilde, men dette er i metabase og dermed ikke tilgjengelig for tenants.

Ser for meg at vi inkluderer billinginformasjon i console i to steg:

Totaloversikt for tenant med breakdown på de ulike SKU'er fra aiven og google
billinginformasjon per team og applikasjon
For å ikke knekke budsjettet kan vi antakeligvis ikke gjøre direktespørringer for alle brukere mot BQ direkte. Vi bør derfor ha en form for caching på plass fra starten av.

We want to provide users with the ability to search for a application (instance) and team for quick navigation.

This should be sufficient for our initial release, and does not require full-text search.

The app-instances is readily available in our appinformer and the teams should/could be cached from team-admin.

Vi har noe deployinformasjon i console i dag, men ingen god historikk.
Det hadde vært interessant å tilgjengeliggjøre deployhistorikk fra tidenes morgen.

console-backend W0605 12:34:10.599207 1 reflector.go:347] pkg/mod/k8s.io/[email protected]/tools/cache/reflector.go:169: watch of *unstructured.Unstructured ended with: an error on the server ("unable to decode an event from the watch stream: stream error: stream ID 157; NO_ERROR; received from peer") has prevented the request from succeeding

Does this have anything to do with nginx or LB configuration when communicating with apiserver? Timeouts?

When subsystem is unavailable it should time out faster and return empty result and give a better user
experience. Now everything just hangs when e.g. Hookd is unavailable.