Repository navigation
agent: every MCP Events start path waits for the session's gates; restore window covers runtime-only daemons - #156
Merged
Conversation
… restore window covers runtime-only daemons Audit of #152: - A runtime subscription restored when its session starts (boot restore, and the restart after a panic) was started from the restore task spawned in `McpEventsHub::attach`, before the session's elicitation and sampling gates existed, so a question on its first poll could still be declined as "no client". One signal now gates them all: `Hub::questions_ready`, set by `McpEventsHub::start` once the gates are installed, and awaited by every subscription task (configured, restored, resubscribing) before its first request. Test: a restored runtime subscription asks on its first poll while the slow-gate seam holds the gates off; the question reaches the client. Fails (declined) without the wait. - The 503-while-restoring window was armed only when configured subscriptions existed, so a daemon whose only subscriptions are runtime ones still answered 410 to early retries. The window now tracks every session the daemon starts to restore callbacks (the events session and each runtime-subscription session) and closes when the last has reserved its tokens (still bounded at 10 min). Test: a runtime-only daemon answers an early retry 503, not 410. - Once restore completes, a token nothing holds is 410 at once, asserted in both restart tests, so a daemon stuck at 503 (restored() dropped) fails them. - release_seams: the serve_ws seams named by the audit already sit in `#[cfg(debug_assertions)]` functions (a release build carries none of the names: checked with grep on the release binary). The checker's hole was elsewhere: any attribute within three lines above counted as gating, so `#[cfg(debug_assertions)] let x = 1;` gated the call site after it. It now requires the attribute directly above the start of the statement or arm the name is in; pinned with that case. The restore seam moved into a gated function of its own. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01JimHGjsfk2Ktm5GxyZJKKk
jaredLunde
added a commit
that referenced
this pull request
Oct 7, 2026
…ore; release binaries scanned for seams Audit of #156: - A listed restoring session could hold the 503 window for its full 10 min if it failed to start, ended with an error, got no events hub (its MCP catalog now empty), or panicked on every start. `mcp_events::restored` is now called on each of those terminal paths: a failed boot pin (events session or runtime session), a session body that ends with an error, a session that starts with no hub, and every panic (a restart reserves its own tokens again). Tests: a listed session with no events hub, and one whose transcript no longer opens, each close the window, so a dead token is 410 within moments; both stay 503 without the fix. - No test seam ships, enforced on the artifact: `build (release)` now runs `mise check:release-seams`, which scans every release binary for the `BEYOND_AI_AGENT_TEST_` prefix and fails on any hit (a grep, seconds; no extra build). The source checker stays as the fast early warning. The scanner is pinned by a test: it fails on this crate's debug build (which carries the seams), on a `concat!`-assembled name, and passes a clean file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01JimHGjsfk2Ktm5GxyZJKKk
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes the four audit findings on #152. Each one has a test that fails without its fix, checked by reverting the fix.
McpEventsHub::attach, before the session's elicitation and sampling gates existed. That covers both boot restore and the restart after a panic. A question on their first poll could still be declined as "no client".Hub::questions_ready, a latch set byMcpEventsHub::start(formerlystart_configured).servecalls it once the gates are installed. Every subscription task (configured, restored, resubscribing) awaits it before its first request.mcp_events_nested::a_restored_runtime_subscription_asks_only_once_the_session_takes_questions. Run 1 makes a runtime subscription. Run 2 restores it with the server asking on its first poll whileBEYOND_AI_AGENT_TEST_SLOW_GATE_INSTALL_MSholds the gates off. Without the wait, the question is declined and the test times out.410to early retries.restored(session_id)after reserving its tokens, and the window closes when the set is empty. It is still bounded at 10 min.mcp_events_receiver::with_only_runtime_subscriptions_an_early_retry_is_told_to_retry_not_to_stopgets410without the fix.restored()survived every test, so a daemon stuck at503for the full 10 minutes would go unnoticed.410immediately after restore completes. Withrestored()removed, both fail (503).serve_ws.rscall-site seams would fail #151'srelease_seamschecker.simulated_delay("…SLOW_SESSION_END_MS"/"…SLOW_ATTACH_MS")sit inside#[cfg(debug_assertions)]functions. A release build of this branch contains none of theBEYOND_AI_AGENT_TEST_names (grep ontarget/release/beyond-ai-agent: 0; control string: 1). The hole: any attribute within three lines above counted as gating, so#[cfg(debug_assertions)] let x = 1;"gated" a call site written after it. The checker now requires the attribute directly above the start of the statement or arm that contains the name, walking back over continuation lines. The restore seam also moved into a gated function of its own.release_seams::the_seam_check_catches_a_name_at_a_call_sitegains the gated-neighbour case, which the old heuristic passes and the new one catches, and a multi-line gated statement that must still pass.Verification
-D warnings, all targets),cargo fmtanddprint checkare clean.mcp_events_*targets pass, plusrelease_seams,serve_service_session_endandserve_session_lifecycle.🤖 Generated with Claude Code
https://claude.ai/code/session_01JimHGjsfk2Ktm5GxyZJKKk