Skip to content

Actor with DATA scope but no durable-dir volumes fail to suspend / pause #1936

Open

Description

The ActorTemplate validations accept onCommit/onPause: DATA with no durable-dirs defined. During suspend or pause, ateom (both microVM and gVisor) fail making the actor go to CRASHED state.

The error message is:

Error: failed to suspend actor: rpc error: code = Internal desc = no durable-dir volumes for DATA snapshot

Is this configuration useful? Should we validate and reject it?

Activity

  1. cloudonly commented on Sep 29, 2026

    @cloudonly
    Contributor

    I checked both checkpoint implementations and would like to add admission validation for this. The check needs to use volumes actually mounted by a container: a declared but unmounted durable-dir still fails the runtime check.

    Proposed scope: extend the existing CreateActorTemplate cross-field validator so DATA in onCommit or onPause requires a mounted durable-dir. Preserve the current microVM exception for a mounted CSI volume; gVisor currently requires a durable-dir even when CSI volumes are present. FULL snapshots remain valid without durable volumes.

    I will add validation regression tests for both scopes and runtimes, including missing/unmounted volumes and the CSI exception. The change will stay in admission validation and will not introduce empty snapshot archives or change CSI snapshot implementation.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions