Skip to content
Supported non-privileged cgroup delegation and isolated helper on Railway services?
avromy
OP

7 hours ago

Capability inquiry only; please do not change, restart or provision anything. Does Railway support a narrowly delegated writable cgroup-v2 subtree for a non-root service helper, with memory/CPU/PID limits and cgroup.kill plus populated=0 readback for whole-descendant cleanup? Can it also support separate UID, mount/PID/network namespaces, NoNewPrivs and proc/environment isolation for that helper, without privileged mode, SYS_ADMIN, Docker socket or disabling seccomp? A bounded metadata-only observation on an ordinary existing service showed a read-only cgroup-v2 mount, application UID1000, Seccomp2 and NoNewPrivs0. No credential or private file was inspected. The UI/API settings examined expose resource limits but no delegation control. Please confirm whether these exact controls are supported on current ordinary services, require a separately isolated service/environment or dedicated host/plan, or are unsupported. If another supported placement exists, please state the mechanism, permissions and incremental pricing before any action. Environment network isolation and sealed variables alone do not establish this process-level requirement. No new resource, paid plan, configuration change, credential access or runtime operation is authorized by this inquiry.

$20 Bounty

4 Replies

Railway
BOT

7 hours ago

The per-service controls we document are replica limits, which cap the CPU and memory available to each replica from the service's Deploy settings. We can't confirm support for a delegated writable cgroup-v2 subtree, helper-level namespaces, or the other process isolation controls you listed, on ordinary services or on any other plan or placement.

Your observation of a read-only cgroup-v2 mount from inside an existing service reflects how that service runs today. If your design depends on these controls, validate it in a separate, disposable test project before relying on it, so your existing services stay untouched.


Status changed to Awaiting User Response Railway • about 7 hours ago


Railway

The per-service controls we document are replica limits, which cap the CPU and memory available to each replica from the service's Deploy settings. We can't confirm support for a delegated writable cgroup-v2 subtree, helper-level namespaces, or the other process isolation controls you listed, on ordinary services or on any other plan or placement. Your observation of a read-only cgroup-v2 mount from inside an existing service reflects how that service runs today. If your design depends on these controls, validate it in a separate, disposable test project before relying on it, so your existing services stay untouched.

avromy
OP

7 hours ago

Please route this narrow capability inquiry to Railway engineering/support for a human answer. We will not modify the existing service or allocate another trial on an unconfirmed capability. Can ordinary services provide writable cgroup-v2 delegation and scoped helper namespaces, or is this unsupported? If a distinct supported host product can do so, please name its exact control/configuration, entitlement and price. No disposable project, paid resource, restart, credential access or configuration change is authorized. We need confirmation only.


Status changed to Awaiting Railway Response Railway • about 7 hours ago


Railway
BOT

5 hours ago

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

Status changed to Open Railway • about 5 hours ago


IIRC this is unsupported.


kuldipsaroj

an hour ago

Hello Railway Support Team,

I’m evaluating whether Railway’s current infrastructure can support a more tightly isolated execution environment for a non-root helper process running alongside an ordinary application service.

Before considering any architectural changes, I’d like to clarify a few platform-level capabilities. This is strictly an inquiry; I am not requesting any changes to my existing deployment.

1. cgroup-v2 delegation and resource enforcement

Does Railway support delegating a narrowly scoped, writable cgroup-v2 subtree to a non-root service process?

Specifically, I’m looking to establish whether the helper can independently manage:

  • Memory, CPU and PID limits.
  • Child-process containment within its delegated subtree.
  • cgroup.kill for terminating the complete descendant process tree.
  • Reliable populated=0 readback to verify that all descendant processes have exited.

2. Process and namespace isolation

Can an ordinary Railway service support a helper running with independently configured:

  • Separate, unprivileged UID.
  • Mount, PID and network namespaces.
  • NoNewPrivs enforcement.
  • Isolated /proc visibility.
  • Environment-variable and process-environment separation.

The intended implementation must work without privileged containers, CAP_SYS_ADMIN, Docker socket access, or disabling the platform's seccomp restrictions.

3. Existing environment observations

During a limited, metadata-only examination of an existing ordinary service, I observed:

  • cgroup-v2 mounted read-only.
  • Application running under UID 1000.
  • Seccomp mode 2.
  • NoNewPrivs reported as 0.

I did not inspect credentials, private files, or application secrets. The resource settings available through the UI/API appear to expose application-level resource limits, but I could not identify an option for explicit cgroup delegation.

4. Supported deployment options and pricing

Could you clarify which of these categories accurately describes Railway's current capabilities?

  1. Fully supported on ordinary services through existing configuration.
  2. Available only through a separately isolated service or environment.
  3. Requires a dedicated host, custom infrastructure arrangement, or specific plan.
  4. Not currently supported by Railway.

If another supported deployment mechanism exists, please explain its technical requirements, required permissions, operational limitations, and any incremental pricing.

One important distinction: network-level environment isolation and protected environment variables do not independently satisfy the process-level containment requirements described above.

Authorization boundary

Please treat this strictly as a technical capability and architecture inquiry.

I am not authorizing:

  • Creation or provisioning of additional resources.
  • Service restarts or deployment changes.
  • Plan upgrades or paid infrastructure.
  • Configuration or permission modifications.
  • Credential or private-file access.
  • Runtime experiments or operational changes.

I would appreciate a clear answer based on Railway's currently supported infrastructure rather than a general description of container security.

Thank you for your time and technical clarification.


Welcome!

Sign in to your Railway account to join the conversation.

Loading...