Files
PaperClipAI/server
DottaandPaperclip 3421366903 fix(server): schedule decision TTL expiry from saved deadlines (#17479)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work.
> - Open decisions have a saved expiry time.
> - The main maintenance tick scans those times even when none is due.
> - Committed creation and resolution can maintain the actual next
deadline.
> - The existing scheduler can run expiry only when that deadline
arrives.
> - This pull request gives decision TTL expiry its own worker.
> - Terminal history becomes quiet while expiry and continuation
delivery remain recoverable.

## Linked Issues or Issue Description

Targets master after #17478 merged. This is slice 10 of the recurring
server work migration. It owns decision TTL expiry. Target invalidation,
interrupted effects, and retention remain separate migrations. The full
ledger is in `doc/plans/2026-10-24-recurring-server-sleep-stack.md`.

**What existing behavior does this improve?**

Time-based decision expiry in the server.

**Current behavior**

The combined decision maintenance sweep queries expired open decisions
on each tick. It mixes that query with target invalidation, effect
recovery, and continuation delivery.

**Proposed behavior**

Creation, bundle creation, and existing decision updates notify the
shared scheduler. A dedicated worker runs at the first saved open
expiry. Resolving an open decision removes its deadline. The combined
tick opts out of TTL selection while its other duties remain intact.

**Reason and benefit**

A future decision no longer requires repeated TTL queries before its
deadline. Terminal history and an empty decision table leave no
recurring TTL work. No queue table is added.

## What Changed

- Add an expiry owner for bounded due selection, next-deadline lookup,
and expiry settlement. Reuse it for expiry-on-use and the existing
target expiration transition.
- Use the central expiry-only creation notification before insert,
including bundles and caller-owned transactions. Update and terminal
transitions continue to notify expiry and continuation.
- Register the enabled-heartbeat TTL worker before decision
housekeeping. Preserve disabled-heartbeat behavior and the public
combined sweep's default batch limit.
- Honor `PAPERCLIP_DECISIONS_SWEEP_BATCH_SIZE` (default 100). Keep a
fair expiry/ID cursor across failures. Stop between candidates on
shutdown and reconstruct saved work after restart.
- Recheck open status and due TTL during settlement. Merge metadata and
persist activity atomically with expiry. Publish activity and deliver
continuation after commit. Handle exact timestamp equality.
- Continue successful overdue batches after one second through the
existing deadline adapter; preserve thirty-second failure retry and all
default callers.
- Add generated migration `0389_brave_hulk.sql`, an idempotent
concurrent partial index on open expiry and ID. Preserve five generated
snapshots.
- Add 20 real-PostgreSQL lifecycle cases and startup ownership
assertions. Document contracts and limitations in
`doc/idle-decision-expiry.md`.

## Verification

Current candidate: `aa03d2c4c25af4e6c2acf49d3160bbaf6ada030b`. This PR
targets master after #17478 merged. Creation now signals only its real
expiry obligation; terminal updates preserve continuation delivery.

- **117 current cases pass** across all 20 expiry, 54
decision/continuation, 42 startup and one scoped-deadline cases. Server
typecheck/build pass. Separate migrated PostgreSQL fixtures exercise
real transaction and notification behavior.
- Single and bundle creation each failed a regression before the
expiry-only signal repair: open rows unnecessarily notified continuation
recovery. Both now arm their saved expiry without a continuation signal.
Cancelling each row afterward still signals and delivers continuation
normally. The distinction is in the existing central notification owner;
there are no per-route flags or counters.
- **Three new real PostgreSQL regressions failed before repair:** a
configured batch of two was ignored, a successful 101-row backlog waited
for the thirty-second error retry, and an activity insert failure left
the row expired without its audit. All three pass. The audit failure
rolls back expiry; retry records one activity, preserves unrelated
metadata and wakes once.
- The shared deadline adapter has optional `continuationMs`. Only this
owner opts into one-second successful overdue continuation. Other
callers keep the previous default, and errors still retry after thirty
seconds. The archive consumer passes with the option absent.
- Migration `0389_brave_hulk.sql` is regenerated from the parent's
metadata. Parent SQL/journal entries are preserved, five generated
snapshots remain, schema drift and migration numbering/safety pass.
Fresh migration apply and replay pass.
- Lifecycle coverage includes first producer activation, future-deadline
quiet, terminal history, exact equality, commit/rollback, restart,
drain/resume, failed expiry and wake delivery, fair paging through
failed rows at PostgreSQL timestamp precision, shutdown and recovery.
- Failure accounting: the first full repair selection passed 113 cases
and failed a startup assertion that observed a pending follow-up
notification before reconciliation. The assertion now waits for the
exact saved future deadline, then still proves no early expiration.
Earlier fault-fixture and timestamp-cursor failures remain recorded in
[the behavior audit](doc/idle-decision-expiry.md). Extra upstream
embedded-only tests attempted on the parent could not initialize
locally; CI must cover that environment.

Commands: server typecheck/build and Vitest for `decision-expiry-work`,
`decisions-service`, `decision-retention-service`,
`server-startup-feedback-export` and `scoped-deadline-work`; DB
migration checks, snapshot drift and actual migration apply/replay. No
full local repository-suite or deployed sleep/cost claim is made. Fresh
current-head CI and Greptile/Ponytail remain required before merge.

## Risks

- Expiry uses `expiresAt <= now`; the old combined loop skipped exact
equality. The deadline boundary is now covered explicitly.
- TTL expiry records the TTL reason. Target-driven expiry preserves its
own reason. When both causes apply, the first terminal transition wins.
- Supported producers share the application DB signal owner. Independent
SQL writers or multiple active processes need a separate notification
contract. New deadline-edit APIs must use the existing helper.
- The index requires top-level concurrent migration execution. It is
registered with the migration owner and replayed in verification.
- Future in-process deadlines are not external Cloud alarms. Existing
sleep guards remain conservative. This PR does not claim host/database
suspension or measured savings.

## Model Used

OpenAI Codex, GPT-6 family, with reasoning and code execution. The exact
serving model identifier and context window were not exposed.

## 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>
2026-10-27 18:10:36 -05:00
..