Every Container Runtime build in my project fails at the same point. Buildah
cannot initialise its storage graph driver, and it dies at buildah info —
the health check before the build — so my Dockerfile is never read.
Support confirmed this is entirely builder-side and has no ETA, and asked me to post here so product engineering can see it.
The error
→ Restoring the Buildah store cache✓ Buildah store cache: miss (no_compatible_snapshot) in 0.1s▲ container buildah: buildah version 1.43.4 (image-spec 1.1.1, runtime-spec 1.2.1) (storage-driver=storage.conf)▲ container Could not verify buildah storage via `buildah info`: `flock --shared --close /tmp/vercel-buildah-store.lock buildah --root /vercel/.containers/storage --runroot /run/containers/storage --registries-conf ... info` exited with code 125time=... level=error msg="Failed to GetDriver graph blob /vercel/.containers/storage"Error: failed to GetDriver graph blob /vercel/.containers/storage: driver not supportedtime=... level=warning msg="failed to shutdown storage: \"failed to GetDriver graph blob /vercel/.containers/storage: driver not supported\""Error: [BUILD_FAILED] `... buildah ... build --platform linux/amd64 --layers --force-rm --pull=newer --network host -t vcr.vercel.com/<team>/<project>/api:vercel-buildah-store -f /vercel/path0/Dockerfile.api /vercel/path0` exited with code 125Failing deployments: wXMweLDMK, H4cafCwVg (build region iad1, Vercel CLI 65.15.0,
buildah 1.43.4).
Ruled out
| Variable | Result |
|---|---|
| Build cache | identical failure with "Skipping build cache, deployment was triggered without cache" |
| Build machine | identical on Standard (4 core / 8 GB) and Enhanced (8 core / 16 GB) |
vercel.json | validates against https://openapi.vercel.sh/vercel.json |
| Dockerfile | never parsed — buildah exits at info |
| Application code | the Vite build in the same run succeeds (2880 modules, assets emitted); only the container step fails |
Three consecutive runs, same millisecond of the build each time. Nothing I control changes it.
GetDriver graph ... unsupported is what containers/storage reports when the host
cannot provide the configured graph driver and no usable fallback (e.g. vfs) is
selected — see https://github.com/containers/storage/issues/1043
Why there's no workaround
Support initially suggested building locally and pushing to VCR. That doesn't work for a container service:
vercel.jsonhas noimagekey — not at the top level, not insideservices, where the only build input isentrypointpointing at a Dockerfile- pre-pushing a tag does not make the build phase skip buildah
- pre-built VCR images are currently consumable only by Vercel Sandbox and Daemons
Support confirmed all three. So a project using Container Runtime services is entirely dependent on the platform builder, and right now that builder is broken for my team with no ETA.
Config
{ "regions": ["sin1"], "services": { "web": { "root": "./", "framework": "vite", "outputDirectory": "dist" }, "api": { "root": "./", "runtime": "container", "entrypoint": "Dockerfile.api", "functions": { "Dockerfile.api": { "maxDuration": 1800 } } } }, "rewrites": [ { "source": "/api/(.*)", "destination": { "service": "api" } }, { "source": "/(.*)", "destination": { "service": "web" } } ]}Dockerfile.api is a plain FROM node:24-slim build — npm ci, workspace build,
CMD node. Nothing exotic.
Questions
- Is the buildah storage driver misconfigured on some build workers in
iad1, or is this fleet-wide? - Is a
vfsfallback planned when the configured driver is unavailable? Failing atbuildah infowith no fallback turns a degraded worker into a hard outage. - Any plan to let a service reference a pre-built VCR image? That would have made this a non-event rather than a full block.
Happy to supply more deployment IDs or logs.