Skip to content

[Feature]: Durable delivery and replay for SOP run events #898

Open

Description

@apexphere

Problem

An external caller can start a LoopForge SOP and subscribe to its run, but it needs reliable delivery of decisions and outcomes after a network gap or receiver restart.

Verified on dd56585276d8ecb6e0d6865f55e255194d75f85a using two disposable script SOPs and a local approval node. After a disconnect spanning completion, the first SSE frame was the current terminal run at revision 33. An old Last-Event-ID did not replay the earlier approval states. after and logsAfter controlled telemetry and log deltas. The SSE connection stayed open and emitted its heartbeat after completion. One approval was first visible while the run was still running and then while it was waiting, with the same node attempt.

The documented run stream and WeCom notifications provide useful live views. API integrations need a durable, provider-neutral contract for the underlying run facts.

Proposed behavior

Expose durable run-fact delivery for successful completion, terminal exception, and a node entering or leaving an actionable decision state. A receiver can resume from a documented position or receive a signed outbound callback with retries. Each fact carries a stable ID, the run ID, the corresponding node/attempt when applicable, and links or identifiers for the authoritative run and artifacts.

Acceptance

  • A client disconnects before a run enters approval and reconnects after it completes. It can retrieve both facts in source order from the last acknowledged position.
  • Repeated delivery of the same approval attempt keeps its fact ID, including across running → waiting. A new approval attempt gets its own ID.
  • A receiver failure and retry preserve the fact ID. An expired cursor yields an explicit gap response.
  • Use a disposable project to verify approval, rework, failure, and completion across a receiver restart. Document authentication, delivery semantics, and retention.

Scope

Platform — UI / integrations. Related work on sandbox orchestration (#117), SSO callbacks (#215), and scheduled-task metrics (#460) covers different integration surfaces.

Run SSE reference · Run stream handler

Activity

  1. cocofhu00 commented on Nov 16, 2026

    @cocofhu00
    Contributor

    Thanks for the detailed write-up! As I understand it, external callers can start and subscribe to SOP runs, but after a disconnect or receiver restart they can't reliably recover earlier decision states and outcomes (an old Last-Event-ID doesn't replay past approvals). You're proposing durable, resumable run-fact delivery (or signed callbacks with retries) with stable fact IDs for completion, failure, and approval-state transitions.

    We've seen this and will look into it through the regular issue queue.

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

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions