Skip to content
How to safely pause GitHub autodeploy before repairing a missing App installation?
dwaynedaone
OP

a month ago

Requesting official Railway guidance on a general production-safety procedure, not permission to change or deploy any service. An existing GitHub-connected service shows “Auto deploy unavailable — Could not load branches.” Its GitHub account login is connected, but the Railway GitHub App installation is absent. Service Settings shows the source repository Disconnect button enabled, while the connected-branch Disconnect control is disabled; the documented Disable control is not available in this state. The running release must continue serving unchanged during repair. No installation, source disconnect/reconnect, trigger change, merge, test push or deployment has been performed.

Please clarify:

  1. Is there a supported way to disable ONLY GitHub automatic deployment while branch loading fails, with no build/restart, source reconnection, variable change or interruption of the running deployment? What exact UI/CLI/Public API control and scope should be used, and how is the hold verified?
  2. Does installing/authorizing Railway App for one selected repository queue, replay or backfill its existing branch HEAD or previously missed events without a new push? Does an explicit disabled autodeploy setting prevent installation-callback deployments and survive the repair?
  3. If serviceDisconnect is the supported alternative, does it preserve the running deployment, deployment rollback artifacts, variables, mounted private volume and service networking? Is it service-wide across environments? Does it cancel queued triggers, or must they be handled separately? Can the source be restored while keeping autodeploy disabled, without an immediate build of the old branch HEAD?
  4. After repair, can a reviewed exact commit be manually deployed through serviceInstanceDeployV2(commitSha) while autodeploy remains disabled, followed by separate approval to restore main autodeploy? Does enabling autodeploy itself deploy the then-current HEAD?

Please answer behavior/sequence only; do not change any account or service settings. No project/repository identifiers, secrets or private logs are included in this general question. Prepared with an AI assistant at the account owner’s request.

$10 Bounty

2 Replies

Status changed to Awaiting Railway Response Railway • about 1 month ago


Railway
BOT

a month ago

This thread has been opened as a bounty so the community can help solve it.

Status changed to Open Railway • about 1 month ago


lancetmason

a month ago

Hey, talk to me about your setup, did you deploy with AI? Are you comfortable with sharing the github repo? I think I'll be able to help with this. Id be open to doing a screenshare as well if you are, that might be easier to help resolve


jalst

a month ago

You don't need to share the repository or grant contributor access to inspect the Railway-side trigger.

There is a narrower API object worth checking before considering serviceDisconnect: DeploymentTrigger. Railway's public CLI schema exposes this read-only query:

query InspectGitHubTriggers($serviceId: String!, $after: String) {
  service(id: $serviceId) {
    repoTriggers(first: 50, after: $after) {
      edges {
        node {
          id
          projectId
          serviceId
          environmentId
          repository
          branch
        }
      }
      pageInfo { hasNextPage endCursor }
    }
  }
}

Use the affected service ID and start with after: null. If hasNextPage is true, repeat with after set to endCursor. Match the environment ID as well as the repository and branch; a service can have triggers for different environments. Keep the results and any API token private.

The same schema exposes deploymentTriggerDelete(id: String!), which takes a trigger ID, while serviceDisconnect takes a service ID. But I would not run it yet under your no-deployment requirement: its description only says it deletes a trigger. It does not document whether queued events are cancelled, whether source configuration is retained, or whether installing the App can recreate a trigger. An empty trigger list also wouldn't prove that an already-queued deployment was cancelled.

I'd ask Railway to confirm that narrower operation's behavior for your identified trigger before repairing the App. I also wouldn't use checkSuites: false as a pause: that controls waiting for CI, not whether pushes deploy.

For the later manual release, the Public API guide does document serviceInstanceDeployV2 with an explicit commitSha, validated against the connected repo. “Deploy Latest Commit” instead selects the branch's latest commit. Neither description establishes the installation/re-enable guarantees you're asking for.

Schema checked. AI-assisted; I validated the query locally against that schema. I haven't accessed your project or tested these mutations.


Welcome!

Sign in to your Railway account to join the conversation.

Loading...