mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-10-29 12:59:13 +01:00
DottaandPaperclip
063cfd4381
<!-- entire-trail-link-start --> https://entire.com/gh/paperclipai/-/trails/204/changes/paperclip/204 <!-- entire-trail-link-end --> ## Thinking Path > - Paperclip coordinates agent work and human approvals. > - Human decisions already persist the continuation action that resumes an agent. > - A response can commit before its callback or publication finishes. > - Periodic maintenance repeatedly discovers those rows, including when idle. > - Committed writes should wake the existing owner; saved retries and claims supply real deadlines. > - Startup reconstructs durable work, and settled history becomes quiet. ## Linked Issues or Issue Description Refs #17651, which is merged. This existing continuation slice targets master and removes its duty from periodic chat maintenance. **Problem:** A saved `interaction_wakeup` action can survive a committed human response whose callback crashes. Without producer coverage, removing maintenance strands the continuation. Repeated empty maintenance also keeps the database active. **Expected behavior:** The outer transaction's settlement wakes the continuation owner. It handles ready conversation publications, dispatches under existing authority, preserves interrupted/retry work, and becomes quiet after settlement. ## What Changed - Register continuation recovery with the shared scoped deadline coordinator. Reconstruct existing issued/processing actions and retain the existing retry and stale-claim deadlines. - Notify from both enqueue producers and owner-local claim/settlement writes. Preserve outer commit/rollback and avoid moving provider or heartbeat calls into database transactions. - Give the existing publication queue an optional company/endpoint/conversation scope. The continuation owner gives ready cards in that conversation a bounded delivery opportunity before claiming the action; FIFO, capacity, ambiguity and provider authority remain with the queue. - Carry admission/cancellation through publication preparation and check before every new send. Join in-flight requests and retain pending actions/publications through drain, standby, shutdown and resume. - Remove only recurring continuation recovery; keep the manual API. Add no table. A concurrent partial index on existing pending continuation actions and matching company/id keyset paging prevent retained history from being scanned. Parent #17651's indexed confirmation discovery and current master are retained. ## Verification Published head: `a1d16f84b3fae361d5d648670a5964486370c3a3`, on actual landed terminal-confirmation parent `ee67d2ece9bda157dfad9bdf25175b1d004ed341`. - **75 distinct current cases pass:** 32 selected real PostgreSQL chat cases (14 continuation lifecycle/producer/stop cases, one new continuation query-cost case, and 17 parent confirmation/query-cost cases; 1,671 skipped), and all 43 cases in complete interaction-publication and app/startup suites. Database build, server typecheck/build, module boundaries and fresh migrations through 0399 pass. The original landed-parent matrix passed 74 cases before the additional query-cost repair; those are not added again. - The new real-SQL regression seeds 20,000 completed/unrelated actions and one pending processing claim. Before repair it examines 20,345 rows and hits 20,421 buffers. The final startup and company-scoped queries must examine at most two action rows and fewer than 20 action buffers; empty scoped recovery must also stay below 20 buffers. Planner settings are unchanged. Migration `0399_ambitious_onslaught.sql` is concurrent/idempotent and uses the existing top-level migration runner; prior SQL/journal entries and the five newest snapshots are preserved. - The matrix covers callback recovery, concurrent responses, malformed prefixes, more than one page, saved retries/stale claims, outer rollback/commit, deferred/future/paused publication states, restart, drain and stop before each new provider request. Quiet assertions inspect real selects/transactions across virtual idle time. - Normal integration `db5a84c0f8b73a0ce863f4d2fbaee0a115cf044f` preserves the complete 12-path owned binary patch and all 12 unrelated current-master paths, including managed Cloud sandbox changes. A verified semantic base resolves squash-ancestry conflicts; both normal parents and the exact tree are checked. The subsequent index/keyset repair is a deliberate additional change, tested separately; the final patch is not claimed identical to the original 12-file implementation. No force push or rebase occurs. - The same 74 pre-index cases passed on the earlier prepared and actual landed parents. The new query-cost regression failed before its repair, then passes on the repaired source. Earlier 57-case and broad acceptance retain their original revision scopes; repeated cases are not added to the current count. The audit-only follow-up leaves all owned server source/test hashes unchanged. - [The owner audit](https://github.com/paperclipai/paperclip/blob/a1d16f84b3fae361d5d648670a5964486370c3a3/doc/plans/2026-10-25-interaction-continuation-sleep-audit.md#acceptance-on-the-actual-landed-terminal-confirmation-parent) retains the original missing-signal regression, the skipped standalone publication suite, and the review's stop-before-send failures with their corrections. Skips and earlier failing revisions are not relabelled as current passes. - **Fresh exact-head CI and Greptile 5/5 without unresolved actionable findings remain required.** The former remote head's reviews are stale. No full local repository suite, live provider or deployed sleep result is claimed. ## Risks - Process-local transaction hints cover supported application writers and shared pools. Independent SQL writers and other replicas do not gain a new wake guarantee. - The publication queue's existing FIFO, capacity, retry/ambiguity and authorization contracts still govern effects. In-flight provider requests can settle after Stop; the fence prevents a subsequent request from starting. - Missing a supported producer can defer recovery until another change or startup. Producer and rollback tests cover this slice's supported paths. - Worker quiet is separate from host/database suspension. External alarm coverage and Cloud sleep gates are unchanged. ## Model Used OpenAI Codex, using repository inspection, code execution and real PostgreSQL tests. The exact model identifier and context-window size are not exposed in this session. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>