Restart the frontend deployment to update site-config

[daxmc99]

@dandiem requested this to reduce CE workload

Test plan

We can do this, we should do this

Will test images

[michaellzc]

will this cause drift with the deployment manifests? e.g., kustomize, helm.

I would like to see an E2E test against a locally deployed (with kind) sourcegraph k8s deployment.

you can build images from this branch with:

sg ci build docker-images-candidates-notest

Then in your helm overrides:

storageClass:
  create: false
  name: standard

sourcegraph: 
  localDevMode: true
  image:
    # you can find the tag from buildkite logs
    defaultTag: docker-images-candidates-notest-<replace_md>
    useGlobalTagAsDefault: true

[filiphaftek]

will this cause drift with the deployment manifests? e.g., kustomize, helm.

I would like to see an E2E test against a locally deployed (with kind) sourcegraph k8s deployment.

you can build images from this branch with:

sg ci build docker-images-candidates-notest

Then in your helm overrides:

storageClass:
  create: false
  name: standard

sourcegraph: 
  localDevMode: true
  image:
    # you can find the tag from buildkite logs
    defaultTag: docker-images-candidates-notest-<replace_md>
    useGlobalTagAsDefault: true

@michaellzc I do not expect any drift, b/c I just did kubectl rollout restart deployment sourcegraph-frontend and it resulted in these annotations being added to deployment:

  template:
    metadata:
      annotations:
        checksum/auth: 5dda2c15191bf709225a311e8b732e6253d7b1c9afcf89e6ec994511cbbd2cf4
        checksum/redis: 63b58e05a2640417d599c4aee6d866cb9063e3a9aa452dc08dbfff836b7781b7
        kubectl.kubernetes.io/default-container: frontend
        kubectl.kubernetes.io/restartedAt: "2024-02-29T00:51:29+01:00" <-- new one just added by kubectl

so this action is same as normal operation done by our self-service
but +1 for test results.

[michaellzc]

got it, didn't know that's how rollout restart work underneath.

on that notes, a few suggestions:

• replace the annotation namespace from kubectl.kubernetes.io/restartedAt to something like sourcegraph.com/restartedAt. just makes it clear we're not trying to impersonate as kubectl, and leave a proper fingerprint.
• add some basic sanity checks to prevent mis-used, e.g., check if all pods are healthy, check if there's restart in progress, etc.
• record this action in audit logs
• add a flag (e.g., SRC_ENABLE_FRONTEND_SITE_RELOAD) so it's possible to disable this feature if we want, especially frontend availability affects Cloud uptime SLA.

[filiphaftek]

Cloud only - nice 👍