7 hours ago
Apply and dashboard don't update stored limit descriptions, so railway config plan never converges
We manage a service's settings with Railway infrastructure as code (.railway/railway.ts). After railway config apply, we expect railway config plan to report no changes, but it keeps showing a change to cpuDescription and memoryDescription in deploy.limitOverride.containers. Neither the apply nor a dashboard edit changes those stored values, so the plan for this service is never empty and every apply sends the same change again. It is still happening. Please check why these descriptions never change, and either correct them on this service or stop the plan from comparing them.
Evidence and attempts
Versions: Railway CLI 5.75.0, railway npm package 3.17.0.
The stored limits of the service held cpuDescription: "1 vCPU" and memoryDescription: "2 GB" while the service ran with cpu: 4 and memoryBytes: 4000000000. We don't know what wrote these descriptions.
What we did, in order:
- The file declared
cpu: 4andmemoryBytes: 4000000000without descriptions. After an apply, the plan still showedcpuDescription,memoryDescription,diskDescriptionanddiskByteschanging tonull(plan output). - We declared all six keys, with descriptions that match the values (snippet below), applied, and redeployed the service. The next plan showed the same two description changes (plan output).
- The dashboard shows 4 vCPU and 4 GB next to the limit sliders. It never displayed "1 vCPU" or "2 GB" (user).
- We changed the vCPU limit to 3 in the dashboard. The next plan picked up the new limit, and the descriptions still read "1 vCPU" and "2 GB" (plan output, service name replaced):
Plan: 0 to add, 1 to change, 0 to destroy
~ Update <service> deploy.limitOverride.containers.cpu, deploy.limitOverride.containers.cpuDescription, deploy.limitOverride.containers.memoryDescription
└ deploy.limitOverride.containers.cpu (3 → 4)
└ deploy.limitOverride.containers.cpuDescription ("1 vCPU" → "4 vCPU")
└ deploy.limitOverride.containers.memoryDescription ("2 GB" → "4 GB")- We set
cpuDescriptionandmemoryDescriptionto an explicitnullin the file and applied. The next plan still showed "1 vCPU" and "2 GB" as the current values (user).
A database in the same environment stores descriptions that agree with its limits. Declared the same way, it plans clean (plan output).
The declaration, equivalent to our file, which builds the same object in a helper function:
const api = service("api", {
deploy: {
limitOverride: {
containers: {
cpu: 4,
cpuDescription: "4 vCPU",
memoryBytes: 4000000000,
memoryDescription: "4 GB",
diskBytes: 100000000000,
diskDescription: "100 GB",
},
},
},
});The railway/iac types don't include the description fields, so tsc rejects this object literal. The CLI evaluates it anyway, because it strips types without checking them.
Why contact support
The AI assistant first suggested reporting this to Railway, after the dashboard test in step 4. Its reason was that the CLI plans a change that neither railway config apply nor the dashboard carries out. We found no way to change the stored descriptions ourselves. We ask Railway to check where cpuDescription, memoryDescription and diskDescription come from and why no write path updates them.
Uncertainty and constraints
- Observed: the plan reads the limit values as they are now, including the dashboard edit. It reads the descriptions unchanged across three applies and one dashboard edit.
- Hypothesis, not verified: the descriptions are left over from an older write path that nothing updates now. The dashboard labels showing the current limits instead of the stored descriptions fit this. We don't know when or how the stale values were written.
- From the CLI source (
src/iac/change_set.rs, 5.75.0), the plan reads the environment config, and Railway's API previews and applies the change set (environmentPreviewChangeSet,environmentApplyChangeSet). The CLI marks adeploychange as a redeploy, so each apply that carries this change may redeploy the service. - We can't change the descriptions through the apply, whether the file sets, omits or nulls them, or through the dashboard. We think only Railway can.
Prepared with Railway's support template (v1).
1 Reply
Status changed to Awaiting Railway Response Railway • about 7 hours ago
6 hours ago
The CPU and memory limits themselves are taking effect: the plan reads cpu and memoryBytes as they currently stand, including your dashboard change to 3 vCPU, and the dashboard reflects the live limits.
The part that does not converge is limited to cpuDescription and memoryDescription under deploy.limitOverride.containers. Those two keys are not part of the railway/iac types, which is why tsc rejects the object even though the CLI still evaluates it.
We can't yet say where the stored "1 vCPU" and "2 GB" values came from or when they were written.
Status changed to Awaiting User Response Railway • about 6 hours ago