Repository navigation
Conversation
FederationImpl._startQueueInternal() began with a guard that only
checked the inbox and outbox queues. The fanout queue was added in
1.5.0, but the guard was not updated, so a federation configured with
only queue: { fanout: ... } returned before reaching the fanout
branch. ContextImpl.sendActivityInternal() still enqueued a
FanoutMessage, and because nothing called fanoutQueue.listen(), the
messages stayed in the backend unprocessed.
The failure looked intermittent because adding an inbox or outbox
queue started the fanout worker as a side effect, and non-forced sends
with fewer than five inboxes bypass the fanout queue entirely.
Nothing was logged either, because the early return preceded the
logger.
Include fanoutQueue in the guard, and add regression tests for an
explicit startQueue(..., { queue: "fanout" }), a no-selector
startQueue(), and the implicit startup after a forced fanout enqueue.
A manual-start test asserts that sending still enqueues without
starting a worker.
fedify-dev#1389
Assisted-by: OpenCode:deepseek-flash
Assisted-by: Codex:gpt-6-astra
Assisted-by: Claude Code:claude-fable-5-1
Assisted-by: OpenCode:mimo-v2.6-flash-free
|
No actionable comments were generated in the recent review. Walkthrough
Merge Risk:
|
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
Codecov Report✅ All modified and coverable lines are covered by tests.
... and 1 file with indirect coverage changes 🚀 New features to boost your workflow:
|
FederationImpl._startQueueInternal()in packages/fedify/src/federation/middleware.ts returned early when the inbox and outbox queues were unset. The fanout queue was added later without updating that guard, soqueue: { fanout: ... }enqueuedFanoutMessageobjects that nothing consumed. Adding an inbox or outbox queue started the fanout worker as a side effect, which is why the bug looked intermittent.The guard now includes
fanoutQueue. I left the per-role identity checks unchanged: the shared-instance selector defect is pre-existing, already fixed on 2.4, and unrelated to the fanout-only configuration, so it is out of scope here.The regression tests cover explicit
startQueue(..., { queue: "fanout" }), no-selectorstartQueue(), and implicit startup aftersendActivity(..., { fanout: "force" }). All three regression tests fail with the old guard. A manual-start case asserts that sending enqueues without starting a worker. The changelog fragment is materialized into CHANGES.md sosacho checkpasses.Fixes #1389.