Francois-Xavier Beligat
12/23/2025, 9:38 AMkapa.ai
12/23/2025, 9:38 AMFailed to initialize data-plane and a 403 from the server when calling the Dataplane API. [[Multiple issues 2.0.0](https://github.com/airbytehq/airbyte/issues/68134)] From those issues, there are two configuration areas you should check: 1. Dataplane group configuration (v1 → v2 change) When upgrading or deploying v2, if global.dataplaneGroups.defaultDataplaneGroupName is left null, the platform can’t associate workloads with a dataplane group and things break in non-obvious ways. One user fixed “no jobs launching” by explicitly setting this value in their Helm values: [[airbyte services stuck](https://github.com/airbytehq/airbyte/issues/68115)] yaml global: dataplaneGroups: defaultDataplaneGroupName: "AUTO" Make sure you have this (or another valid group name) set in your values.yaml before installing/upgrading. 2. Auth / dataplane identity & secrets Several users hit dataplane initialization / auth problems when: - The airbyte-auth-secrets secret was pre-created and managed externally, and - global.auth.instanceAdmin.* values were pointing to that secret. In those cases, the secret was not updated with the dataplane client credentials, and the workload-launcher could not obtain a valid token. One workaround that resolved “Failed to obtain or add dataplane access token” was to remove those overrides so the chart can manage the secret itself: [[Could not deploy 1.6.x](https://github.com/airbytehq/airbyte/issues/58587)] Specifically, in values.yaml either: - Don’t pre-create airbyte-auth-secrets, and - Comment out / remove: yaml global: auth: instanceAdmin: secretName: airbyte-auth-secrets emailSecretKey: instance-admin-email passwordSecretKey: instance-admin-password or ensure that if you do manage the secret yourself, it is allowed to be updated and actually contains the dataplane client id/secret generated by Airbyte (otherwise the server will return 403 to the workload-launcher). Because the 403 is coming from the Airbyte server’s Dataplane API, the problem is almost certainly with these auth/dataplane settings rather than Kubernetes networking. If after: - Setting global.dataplaneGroups.defaultDataplaneGroupName (e.g. "AUTO"), and - Letting the chart manage airbyte-auth-secrets (or ensuring it’s correctly populated and updatable), you still see the 403, the knowledge sources do not provide further concrete steps for this exact scenario. In that case, you may need to share your values.yaml and server logs around the Dataplane API call in a GitHub issue or the forum for more targeted help.Francois-Xavier Beligat
12/23/2025, 11:05 AMkapa.ai
12/23/2025, 11:05 AMkapa.ai
12/23/2025, 11:05 AMairbyte-server-svc, so I can’t confirm it as a known bug in the current Helm v2 chart. What is documented is that Helm v2 introduced a number of breaking and structural changes (new naming, global.* blocks, etc.), and there were real problems in early 2.0.x chart versions that were later fixed. Users reported that 2.0.0 was broken while 2.0.18 “works properly.” [[Multiple issues 2.0.0](https://github.com/airbytehq/airbyte/issues/68134)] Given that: - Your observation (service selector matching all pods with `app.kubernetes.io/name=airbyte`) is consistent with mis‑routing. - Manually tightening the selector to a more specific label on the server pod fixes routing. - The docs don’t describe or endorse such a broad selector, nor do they document this as expected behavior. it’s reasonable to suspect a chart issue in the specific version you’re using, but the knowledge sources do not explicitly confirm this. If you’re on an early 2.0.x chart (e.g., 2.0.0–2.0.7), the safest next steps based on the available information are: 1. Upgrade to a newer Helm v2 chart version (users report success with 2.0.18 and later). [[Multiple issues 2.0.0](https://github.com/airbytehq/airbyte/issues/68134)] 2. If the problem persists on a current chart, open a GitHub issue with: - Your chart version - The generated airbyte-server-svc manifest (showing selectors) - The relevant pod labels so the maintainers can confirm whether the selector is indeed incorrect in that chart version. In short: your workaround is valid, and it likely points to a chart bug or misconfiguration, but the provided knowledge sources don’t definitively document this specific selector problem.