ADR-0057 data lifecycle follow-ups: retire per-plugin sweepers, dev telemetry datasource + db:clean, Studio surface, PG rotation (tracking)

[os-zhuang]

Follow-up tracking for ADR-0057 (system data lifecycle & retention) after P1–P4 shipped in #2791 (docs/skills in #2818, tracking issue #2786 closed). Each item is independently shippable.

Items

• ① Retire the per-plugin retention sweepers — shipped in feat(lifecycle): ADR-0057 follow-ups — retire per-plugin sweepers, dev telemetry datasource + db:clean, Studio lifecycle form (#2834) #2835. JobRunRetention / NotificationRetention (+ retentionDays/retentionSweepMs options) removed; the LifecycleService is the one sweeper. Included a fix: the blanket 30d lifecycle retention on sys_automation_run (from feat(lifecycle): ADR-0057 P1–P4 — lifecycle contract, Reaper/Rotator/Archiver, datasource separation, governance (#2786) #2791) could strand suspended approval runs — removed; bounding stays with the automation store's terminal-only sweep (no status predicate in the declarative contract yet).
• ② Dev telemetry datasource + db:clean — shipped in feat(lifecycle): ADR-0057 follow-ups — retire per-plugin sweepers, dev telemetry datasource + db:clean, Studio lifecycle form (#2834) #2835. objectstack dev provisions <primary>.telemetry.<ext> as the telemetry datasource for file-backed SQLite primaries (OS_TELEMETRY_DB=0 opts out, =path opts in anywhere); new os db clean runs the one-time VACUUM (smoke-verified 48.9 MB → 0.01 MB, auto_vacuum=2 adopted).
• ③ Studio surface for lifecycle — shipped in feat(lifecycle): ADR-0057 follow-ups — retire per-plugin sweepers, dev telemetry datasource + db:clean, Studio lifecycle form (#2834) #2835. object.form.ts exposes the full lifecycle block; metadata-forms i18n bundles regenerated with curated zh-CN. NOTE: the *.objects.generated.ts bundles were deliberately not regenerated — see the extract-config drift note in the comments.
• ④ PostgreSQL native-partition Rotator — evaluated and deferred (see design sketch in the comments): no PG service in CI means the DDL would ship unverifiable. Current PG behavior is correct (age-based fallback; the bound holds). Prereq: a postgres service container in CI + sql-driver-rotation.pg.test.ts.
• ⑤ (found during ③) i18n extract-config drift for *.objects.generated.ts — the committed bundles carry curated sys_* translations (e.g. sys_audit_log) whose objects live outside packages/platform-objects/scripts/i18n-extract.config.ts's import set; a plain re-run of the extractor deletes them. Import the missing sys_* objects into the config (or split per-package bundles) before anyone regenerates object keys.

References

• ADR: docs/adr/0057-system-data-lifecycle-and-retention.md (P0–P4 implemented)
• P1–P4: feat(lifecycle): ADR-0057 P1–P4 — lifecycle contract, Reaper/Rotator/Archiver, datasource separation, governance (#2786) #2791 · docs/skills: docs+skills: document the ADR-0057 data-lifecycle surface (#2786 follow-up) #2818 · follow-ups ①–③: feat(lifecycle): ADR-0057 follow-ups — retire per-plugin sweepers, dev telemetry datasource + db:clean, Studio lifecycle form (#2834) #2835 · original tracking: ADR-0057 data lifecycle P1–P4: retention contract, rotation, separation, governance (tracking) #2786

🤖 Generated with Claude Code

[os-zhuang]

④ PostgreSQL native-partition Rotator — evaluated, deferred with a design sketch (not shipping unverified DDL)

Neither this environment nor CI has a PostgreSQL service (driver-sql tests run better-sqlite3 only), so a PARTITION BY RANGE implementation would land as code no test can ever execute — exactly the unverified-claim class ADR-0049/0054 exist to block. Current behavior on PG is already correct (rotation falls back to an equivalent age-based reap; the declared bound holds — only reclamation is row-wise instead of O(1)), so this is an optimization, not a gap.

Design sketch for whenever a PG service lands in CI:

• SqlDriver.supportsRotation becomes dialect-aware (isSqlite || isPostgres).
• PG path in ensureRotation: base table converted once to PARTITION BY RANGE (created_at) (requires a rebuild: rename → create partitioned → attach legacy as the first partition); per-period partitions <table>__r<key> created via CREATE TABLE … PARTITION OF … FOR VALUES FROM/TO; expiry = DETACH PARTITION CONCURRENTLY + DROP TABLE (O(1), lock-light).
• No view needed (unlike SQLite): the partitioned parent IS the queryable name, and writes route by the partition key natively — the shard-wise write redirects the SQLite path needs don't apply.
• Prereqs: a postgres service container in the CI dogfood/driver jobs + a sql-driver-rotation.pg.test.ts gated on OS_TEST_PG_URL.

Also noted while doing ③ (i18n regen): the *.objects.generated.ts bundles have extract-config drift. The committed bundles carry curated zh-CN translations for objects (e.g. sys_audit_log) that live outside packages/platform-objects/scripts/i18n-extract.config.ts's import set (plugin-audit etc.) — so a plain re-run of os i18n extract deletes those curated translations. ③'s PR therefore regenerated only the metadata-forms bundles and left the objects bundles untouched. Fix candidates: import the missing sys_* objects into the extract config, or split per-package bundles. Worth its own small item before anyone re-runs the extractor for object keys.

🤖 Generated with Claude Code

---

Generated by Claude Code