feat(events): port Phase 4a per-tier policy from cueapi/cueapi#779#76
Merged
Conversation
Verbatim port of cueapi/cueapi#779 (commit a319134) — per-tier subscription dispatcher policy with p=4 debounce. Backward-compat default (`redis=None` kwarg) preserves v1 behavior so this PR is purely additive in the OSS substrate as well. **Surface (verbatim):** - New module `worker/subscription_dispatcher_policy.py` (332 lines): apply_tier_policy + stamp_dispatch_markers + 4 pure helpers. - `dispatch_subscription_events(db_engine, redis=None)` — new optional kwarg; bypassed when None. - Watermark math caveat for deferred events documented inline. - Redis-down fails OPEN (fires events anyway + logs warning). - Settings: `PRIORITY_4_DEBOUNCE_SECONDS: float = 2.0`. - `run_poller` passes `heartbeat_redis` so production gets new policy. **Behavior:** - p=5 / p=3 / p=2 / p=1: unchanged (pass through) - p=4: NEW debounce — max 1 webhook fire per recipient per 2s window **Tests** (22 verbatim port; uses FakeRedis stub): - 5 `_event_priority` branches - 1 `_debounce_key` - 5 `_is_p4_debounced` branches - 2 `_stamp_p4_fire` - 7 `apply_tier_policy` branches - 2 `stamp_dispatch_markers` **Regression**: 40 dispatcher tests pass locally (22 new policy + 18 existing dispatcher). **parity-manifest delta**: - `worker/subscription_dispatcher.py` — last_synced + ported_in bumped (now lists event-emit-primitive-port + phase-4a-policy-port) + deviation note documenting the Phase 4a additions. - NEW: `worker/subscription_dispatcher_policy.py` — `ported_in: phase-4a-policy-port`. **OSS-shape**: verbatim — no hosted-only deviations. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Parity checkThis PR modifies files tracked in
Please confirm one of the following in a reply or PR description update:
This is a soft check — it does not block merge. The goal is visibility, not friction. See HOSTED_ONLY.md for the open-core policy. |
Merged
6 tasks
mikemolinet
added a commit
that referenced
this pull request
May 11, 2026
…ed_at on list endpoint (#79) Re-port of closed [PR #48](#48) which was on a stale base ~8870 deletions behind main. Fresh against current main HEAD via direct patch from the old branch's commit (c05518c). Adds query-side enrichment to ``GET /v1/executions``: - ``worker_id=`` — filter to executions claimed by a specific worker (Execution.claimed_by_worker) - ``status__in=foo,bar,baz`` — comma-separated multi-status filter. Mutex with ``status=``; 400 conflicting_filters if both passed. - Response gains ``oldest_claimed_at`` — earliest ``claimed_at`` over the filtered set, or null when count=0. Unblocks dashboard/menubar "oldest pending: 5m" rendering without a follow-up query. ## Use cases - **Menubar pending counter**: fetch ``pending,delivering,retry_ready`` in one round trip (status__in), get total + oldest_claimed_at to render "3 in flight, oldest claimed 8m ago". - **Worker health dashboard**: filter by worker_id to scope to one worker's claims; see how stale their oldest claim is. ## Tests 3 new tests in TestListExecutions: - worker_id scoping + oldest_claimed_at value reflects earliest claimed_at over filtered set - status__in union + status/status__in mutex 400 - null oldest_claimed_at on empty filtered set 31/31 in test_execution_parity.py green. Full local suite: zero regressions. ## Re-port note Re-port of closed PR #48 (commit c05518c). Branch was ~8870 deletions behind main; fresh against current main after PR #74/#75/#76/#77/#78 merged earlier in this session. Patch applied cleanly from c05518c to current main — the patch only touches list_executions endpoint, which has remained additive (cueapi-core added outcome_state filter since the branch was cut; this PR is compatible).
7 tasks
mikemolinet
added a commit
that referenced
this pull request
May 11, 2026
…ueapi/cueapi#630) (#80) Re-port of closed [PR #47](#47) which was on a stale base ~8880 deletions behind main. Fresh against current main HEAD. Phase A of the Agent Directory productization. Eliminates the failure mode where agents had to remember 6+ fields per recipient AND had no way to discover the live roster. ## What lands - **GET /v1/agents/roster** — display-optimized snapshot for prompt- injection at session-boot. Distinct from the existing management surface (GET /v1/agents): - Always-full list (no pagination) - Drops opaque IDs / secrets / timestamps / tenancy metadata - Adds derived ``online``, ``last_seen_relative``, ``preferred_contact`` - Always excludes soft-deleted agents - Weak ETag + ``If-None-Match`` → 304 Not Modified for poll efficiency - ETag bucketed to 5-min windows so quiet periods produce stable hashes - ``Cache-Control: private, max-age=300`` matches derivation buckets - **Migration 031** (renumbered from private's 048) — adds ``agents.last_seen_at TIMESTAMPTZ NULL``. Nullable, no backfill. - **Hot-path hooks** write ``last_seen_at = now()``: - ``create_message`` — sender's agent (in same tx via touch_last_seen) - ``list_inbox`` — recipient's agent, on EVERY poll (via _bump_last_seen_stmt). Even when no queued messages exist, the poll proves activity. - **Online derivation** (server-computed in ``list_roster``): - within 5 min → ``online`` - within 30 min → ``away`` - older or NULL → ``offline`` - Caller override wins: PATCHed status=away/offline keeps that override regardless of recent activity ## Pure helpers (for unit-testability — pytest-cov + ASGI issue) - ``_build_roster_entry(agent, now)`` in agent_service.py: ORM Agent → (entry_dict, etag_part_string) - ``_compute_roster_etag(parts)`` in agent_service.py: list → weak ETag - ``_derive_online_state(now, last_seen_at, asserted_status)`` → (online_bool, derived_status) - ``_format_relative(now, last_seen_at)`` → "active now" / "5m ago" / ... - ``_bucketed_seen(last_seen_at)`` → string for ETag stability - ``_bump_last_seen_stmt(agent_id, now)`` in inbox_service.py: SQLAlchemy UPDATE statement - ``_etag_matches(if_none_match, server_etag)`` in agents router: conditional GET predicate ## Tests 27 new tests in tests/test_agent_roster.py (verbatim from private): shape verification, hot-path hooks (sender + recipient), derivation correctness across all 3 buckets, caller-asserted status override, soft-delete exclusion, preferred_contact derivation, last_seen_relative formatting, ETag 304 handling, ETag changes when roster mutates, pure-helper unit tests. 27/27 pass locally. Full local suite: 890 passed + 18 xfailed (pre-existing) + 4 skipped. Zero regressions. ## Re-port note Re-port of closed PR #47. Fresh against current main after PR #74 + #75 + #76 + #77 + #78 + #79 merged earlier in this session.
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.
Verbatim port of cueapi/cueapi#779 (commit
a319134) — Phase 4a per-tier subscription dispatcher policy + p=4 debounce.Backward-compat default (
redis=Nonekwarg) preserves v1 behavior; this PR is purely additive in the OSS substrate.Behavior
PRIORITY_4_DEBOUNCE_SECONDSwindow (default 2s)New module
worker/subscription_dispatcher_policy.py(verbatim, 332 lines):apply_tier_policy+stamp_dispatch_markers+ 4 pure-helper internals (_event_priority,_debounce_key,_is_p4_debounced,_stamp_p4_fire).Dispatcher wiring
dispatch_subscription_events(db_engine, redis=None)— new optional kwarg.redis=None(backward-compat) bypasses policy entirely.run_pollerpassesheartbeat_redisso production gets the new policy.Settings
PRIORITY_4_DEBOUNCE_SECONDS: float = 2.0— tunable via env.Tests (22 verbatim port)
5
_event_prioritybranches + 1_debounce_key+ 5_is_p4_debounced+ 2_stamp_p4_fire+ 7apply_tier_policy+ 2stamp_dispatch_markers. Uses aFakeRedisstub — no real Redis dependency.Local regression: 40 dispatcher tests pass (22 new policy + 18 existing dispatcher).
parity-manifest delta
worker/subscription_dispatcher.py—last_synced+ported_inbumped; deviation note documents the Phase 4a additions.worker/subscription_dispatcher_policy.py—ported_in: phase-4a-policy-port.OSS-shape
Verbatim — no hosted-only deviations.
Phase 4b (digest for p=1/p=2) + Phase 4c (observability) queued as follow-up PRs.
🤖 Generated with Claude Code