7 days ago
I’m trying to release a fix to an existing production Railway Function. Applying a single staged Source Code change on the dashboard repeatedly showed ‘Deploy failed’ and then ‘0 changes → production.’ The staged change disappeared, no new deployment or build log appeared, and the previous deployment continued serving.
One attempt was around 11:03 p.m. ET on November 11, 2026, from the Railway dashboard on iPhone. Before that attempt, the staged change was verified as a single source update. Afterward, the deployment history still showed the previous successful deployment and the live health check passed. A local startup test passed with a disposable database, but it did not exercise Railway’s apply step.
Can Railway help determine why the staged change was cleared without a visible deployment record, and advise how to retry safely while preserving the active deployment and customer data? I can provide internal identifiers privately if needed.
Prepared with Railway's support template (v1).
11 Replies
7 days ago
We checked the production environment, and no staged change or failed apply is recorded for it now. That means we can't see why that attempt was cleared, and we have not established the cause. The current fetiq-web deployment is still the one serving.
The Function's own encoded source is a normal, supported setup. Several fetiq-web variables have names that look like front-end files, though: some end in _JS and _CSS, and there are FETIQ_PAGE_1 through FETIQ_PAGE_3. We can only see variable names, not their values. If those variables hold code, markup or assets, that is not a supported use of variables. Each dashboard Deploy carries the whole environment's configuration, so that kind of content can stop an apply even when the staged change itself is small.
If that is the case, move the content into the Function source, a connected repo or image, a volume, or object storage before applying the source change again. Then remove those variables, along with any services you no longer use.
Your Postgres data lives on its own attached volume, and a source change to the Function does not modify that volume.
Status changed to Awaiting User Response Railway • 7 days ago
7 days ago
Thanks. The Function does load frontend assets from environment variables, so we will validate moving those assets into the release artifact in isolated staging. We understand that this is a possible cause and the failed apply has not been traced. Can you confirm whether replacing the Function source and removing the old asset variables in one changeset validates against the resulting configuration or the pre-apply configuration? Is there a documented payload limit or a way to capture an apply rejection or request ID if no deployment is created? The current production deployment remains active.
Status changed to Awaiting Railway Response Railway • 7 days ago
7 days ago
We have not established whether a single changeset that replaces the Function source and removes the asset variables is checked against the resulting configuration or the configuration before the apply, so we can't confirm that ordering for you.
What we can confirm is that a dashboard Deploy carries the whole environment's configuration, not just the staged edit. We don't publish an environment-wide payload size, so there is no figure to design against. Moving the assets out of variables and removing those variables is what reduces what each apply has to carry.
We also have not established that a rejection or request ID is shown anywhere when an apply ends without creating a deployment. Repeatedly re-staging and re-deploying the same configuration is not a safe retry, because configuration writes can land even when the apply does not finish.
Status changed to Awaiting User Response Railway • 7 days ago
7 days ago
Thanks. We will avoid repeating the failed apply. A separate staging service is already running a repo-backed release without frontend assets in environment variables. The remaining question is how to migrate the existing production Function safely, since its current deployment still needs those assets and the apply validation order is unknown. Could a Railway engineer review this failure and confirm a supported sequence that preserves the running deployment and rollback path? If you need a request trace or internal identifiers, please point us to a private channel.
Status changed to Awaiting Railway Response Railway • 7 days ago
7 days ago
Nothing is staged or mid-apply on the production environment right now, so there's no leftover failed attempt you need to clear before your next change.
We can't trace that earlier attempt any further, and we can't confirm whether a combined changeset is checked against the configuration before or after the apply. What we can tell you is what makes these applies fail: each dashboard Deploy carries the whole environment's configuration. That means removing the asset variables is what lets applies go through again. Narrower or repeated attempts don't help, and because configuration writes can land even when an apply doesn't finish, each retry can make the next one fail sooner.
The supported way to ship that frontend code is the approach your staging service already uses: deploy from a connected repo or a published image, keep large files on a volume, or fetch them from object storage. The asset variables can be removed once nothing that is serving traffic still reads them. You don't need to send internal identifiers or a request trace for any of this.
Status changed to Awaiting User Response Railway • 7 days ago
7 days ago
Update: the separate repo-backed production candidate deployed successfully around 12:04 a.m. ET on November 12. Its health check, homepage, and notifications asset responded as expected. The original Function is still serving on its existing hostname. A failure was reported around the latest attempt, but we found no new failed deployment record in this environment. Can you help distinguish a dashboard apply or approval error from a build or runtime failure here? Also, what is the supported way to preserve or transfer an existing Railway-generated hostname to the healthy replacement service with a reversible cutover? No traffic switch or asset-variable cleanup has been performed.
Status changed to Awaiting Railway Response Railway • 7 days ago
7 days ago
The production environment currently shows nothing staged and no failed or in-progress apply. A build or runtime failure would normally leave a new deployment in the service's history with its own build or deploy logs. Because the attempts left no new deployment and the previous one kept serving, they stopped at the configuration-apply step, before any build started. The dashboard doesn't show a separate rejection or request ID for that step.
For the hostname, each service has one Railway-provided domain, and it belongs to that service. We haven't verified that an existing Railway-generated hostname can be moved to another service or reclaimed after it's removed. Don't plan on releasing it as a way to switch traffic, because you may not be able to get it back.
A cutover you can reverse doesn't need to move that hostname. One option is to put a custom domain in front of the service and switch which service it points to. The other is to keep the original Function serving while clients move to the replacement service's own Railway domain. Remove the asset variables only once nothing serving traffic still reads them.
Status changed to Awaiting User Response Railway • 7 days ago
7 days ago
Thanks. We will keep the existing generated hostname attached and leave the original service running. Before choosing between keeping that URL and moving clients to the replacement URL, is an in-place conversion of the existing Function service to a repo-backed service supported while retaining its generated hostname? If supported, would rollback restore both the prior Function source and its configuration, or would configuration need to be restored separately? We are looking for a reversible sequence and have made no routing changes.
Status changed to Awaiting Railway Response Railway • 7 days ago
7 days ago
On rollback, our docs say a deployment rollback brings back both that deployment's image and its custom variables. Rolling back within your plan's image retention window (72 hours on Hobby) restores the previous image, settings and variables as a new deployment, and nothing is rebuilt. After that window the rollback option goes away, and only a redeploy is left, which rebuilds from the original source with that deployment's original variables. See Deployment Actions and the image retention policy.
We have not verified two things: whether switching an existing Function's source to a connected repo in place is a supported change, and whether a rollback would also switch the service's source setting back to the Function. Because of that, we can't confirm an in-place conversion as a reversible sequence. The Railway-provided domain is attached to the service itself, so it is not tied to any one deployment.
Your current setup already gives you a path we can confirm is reversible. The original Function keeps serving on its existing hostname, and the replacement service has its own domain. Clients can move over while the original stays available for fallback.
Status changed to Awaiting User Response Railway • 7 days ago
Railway
On rollback, our docs say a deployment rollback brings back both that deployment's image and its custom variables. Rolling back within your plan's image retention window (72 hours on Hobby) restores the previous image, settings and variables as a new deployment, and nothing is rebuilt. After that window the rollback option goes away, and only a redeploy is left, which rebuilds from the original source with that deployment's original variables. See [Deployment Actions](https://docs.railway.com/deployments/deployment-actions#rollback) and the [image retention policy](https://docs.railway.com/pricing/plans#image-retention-policy). We have not verified two things: whether switching an existing Function's source to a connected repo in place is a supported change, and whether a rollback would also switch the service's source setting back to the Function. Because of that, we can't confirm an in-place conversion as a reversible sequence. The Railway-provided domain is attached to the service itself, so it is not tied to any one deployment. Your current setup already gives you a path we can confirm is reversible. The original Function keeps serving on its existing hostname, and the replacement service has its own domain. Clients can move over while the original stays available for fallback.
7 days ago
Thanks for clarifying the deployment rollback window. If rollback restores the previous Function image but leaves the service source connected to the repo, is there a supported way to restore the Function as the source on that same service so future deploys use it again, with the generated hostname still attached? A documentation link or confirmation from a Railway engineer would help.
Status changed to Awaiting Railway Response Railway • 7 days ago
7 days ago
We haven't verified a supported way to switch a service back from a connected repo to Function source on that same service, so we can't confirm that conversion as a reversible step. For the same reason we can't say a rollback would put the Function back as the service's source for future deploys.
If you want a copy of the current Function code before changing anything, the CLI can pull a linked Function's source to your machine with railway functions pull (see railway functions). That keeps the code, but it isn't a confirmed way to restore Function source on a service that has been switched to a repo.
The path we can confirm stays reversible is the one you already have: leave the original Function serving on its existing hostname, untouched, while clients move to the replacement service.
Status changed to Awaiting User Response Railway • 7 days ago
9 hours ago
This thread has been marked as solved automatically due to a lack of recent activity. Please re-open this thread or create a new one if you require further assistance. Thank you!
Status changed to Solved Railway • about 9 hours ago