Opt-in durability instead of Pause vs Suspend API

Instead of explicitly requesting Pause vs Suspend, I think applications should request Suspend with an argument specifying their tolerance, naming is hard but Durable: true or something like that, opt-in.

When applications do not specify, substrate can intelligently handle suspension, e.g. in the future it might:

Suspend to local disk
Decide not to suspend immediately due to low load
Automatically move snapshots between nodes p2p to rebalance
Decide to move the snapshot to cloud storage
Live migrate sandboxes

Only if you have a hard need for an immediate backup, then we should explicitly move the snapshot to cloud storage (S3, GCS, ...).

I think the current Pause vs Suspend inverts this and doesn't give a lot of room to implement things like automatically moving snapshots between machines or live-migrate microVMs

See related discussion in #660

And previously #119

+1 DisruptionTolerance

It might not even be a property of suspension per se... maybe it is latency insensitive, we can steal CPU from the Actor etc.

Tim Hockin (@thockin)

cc Tim Hockin (@thockin) Dmitry Berkovich (@dberkov) Julian Gutierrez Oschmann (@juli4n) (also note +1 from a few contributors)

Dmitry Berkovich (@dberkov)

Dmitry Berkovich (@dberkov) are you working on this one?

Yes, I do. I will share the proposal this week.

Proposed Changes for GA: PauseActor is removed from the API and SuspendActor gains an asynchronous durability mode:

service Control {
  rpc SuspendActor(SuspendActorRequest) returns (SuspendActorResponse) {}
}

message SuspendActorRequest {
  ObjectRef actor = 1;
  SuspendActorRequestOptions options = 2;
}

message SuspendActorRequestOptions {
  UploadMode mode = 1;
}

enum UploadMode {
  UPLOAD_MODE_UNSPECIFIED = 0; // Defaults to synchronous
  UPLOAD_MODE_SYNCHRONOUS = 1;
  UPLOAD_MODE_ASYNCHRONOUS = 2;
}

message SuspendActorResponse {
  Actor actor = 1;
}

See detailed description in https://docs.google.com/document/d/19-8ZsIdF-J5QldhMyEam3nhJcOVcGkN6bmwrXmqtVq8/edit?resourcekey=0-jik6OvyqInUN5yQiJFkzHA&tab=t.wl9xs9ggvysx

Update Snapshots API to support Actor Lifecycle v2 (agent-substrate#2289

Sync upstream sandbox, snapshot, and HTTPS egress updates (#69)

SuspendActor accepts suspend_options to choose a durability mode (API…

Replace clients of PauseActor to call SuspendActor(ASYNCHRONOUS) inst…

Pause API has been removed in #2813

Navigation Menu

Repository navigation

Opt-in durability instead of Pause vs Suspend API #798

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Opt-in durability instead of Pause vs Suspend API #798

Description

Activity

bowei commented on Aug 7, 2026

BenTheElder commented on Aug 19, 2026

zlammerts-svg commented on Sep 22, 2026

dberkov commented on Sep 22, 2026

zoez7 commented on Oct 2, 2026

zoez7 commented on Oct 18, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions