fix(notebook): ignore stale executor callbacks

[ewen-poch]

Problem

Notebook executor callbacks can arrive after their executor or same-ID Session Aggregate has been replaced. The callback previously resolved ownership from the current registry state, so an old executor could persist run.json, mutate a replacement Aggregate, or emit onNotebookChanged for a generation it no longer owned. A teardown/recreate gap could also fall through to the legacy no-live-session persistence path.

Proposed change

Bind lifecycle callbacks to the originating Aggregate and executor generation, and close callback admission synchronously before teardown or replacement.

flowchart LR
  Executor["Executor callback"] --> Token["Aggregate + executor generation"]
  Token --> Gate{"Still current?"}
  Gate -->|yes| Queue["Lifecycle queue"]
  Queue --> Persist["Update Aggregate + run.json + notification"]
  Gate -->|no| Drop["Ignore stale callback"]
  Teardown["Teardown / replacement"] --> Close["Close admission synchronously"]
  Close --> Drain["Drain already-admitted persistence"]
  Drain --> Replace["Publish successor"]

Loading

• Generated callbacks must still match the live Session Aggregate and executor generation before any mutation, persistence, or notification.
• Teardown and executor replacement synchronously reject later callbacks, then drain already-admitted persistence before releasing the old owner.
• A generated callback received while no live Session exists is stale and is ignored; generation-less legacy rehydration keeps its existing behavior.
• In-place restart preserves the executor generation because the executor instance remains the owner.

Scope and non-goals

• Covers default and named environments, idle-shutdown and terminated callbacks, executor replacement, overlapping teardown, and same-ID recreation gaps.
• No IPC, Electron/Web/CLI/local-RPC payload, schema, Session relationship, UI, Specialist, Permission, Compute, or Issue [Feature]: Provider-neutral multi-agent research orchestration #458 orchestration change.
• No change to normal current-generation execution or persistence semantics.

Acceptance criteria and validation

All checks below ran after the final material edit and an independent fixed-SHA closure review confirmed the evidence mapping with 0 actionable findings.

Behavior	Project-owned check	Final result
Old callbacks cannot mutate/persist/notify after teardown, replacement, or same-ID recreation	npm test -- --run src/main/notebook/runtime-service.test.ts src/main/notebook/session-aggregate.test.ts	174 passed, 9 skipped
Generated no-live-session callbacks are rejected while legacy generation-less rehydration remains	Same focused command, including idle and terminated gap regressions	passed
Node and renderer types remain compatible	npm run typecheck	passed
Repository lint	npm run lint	0 errors; 23 unrelated baseline warnings
Repository regression suite	npm test	649 files passed, 15 skipped; 9,591 tests passed, 184 skipped
Patch hygiene	git diff --check	passed

Review focus

Please focus on synchronous admission closure, draining already-admitted callback persistence before owner replacement, and the distinction between generated stale callbacks and generation-less legacy rehydration.

Remaining risk: real NotebookKernelExecutor timer/shutdown timing is represented by deterministic injected-lifecycle tests rather than a non-skipped external E2E. Queue draining intentionally allows teardown to wait for already-admitted repository persistence so it cannot write across generations.

Merge by squash only after all required CI and AI review checks pass.

[github-actions]

Codex Review

Verdict: mergeable

No actionable findings.

Summary: Static inspection found no concrete merge-blocking defect introduced by this pull request.