Extract Information service to standalone Archiver repo

[gregoryfoster]

Summary

Extract the in-tree Information service (src/information/) into its own standalone repo at /home/exedev/archiver, renamed to Archiver. Follows the Notifier extraction pattern (#132): separate repo on the same VM, separate systemd unit, separate Postgres database, generated Python SDK consumed as a path dependency.

This is the rename + lift; data-model evolution toward InfoSource / SourceSpec / SourceRevision (per docs/research/2026-05-06-archiver-information-model.md) lands as a deliberate v2 in the new repo afterward.

Design doc

docs/plans/2026-05-06-archiver-extraction-design.md

Scope

• Pure rename + extract; storage tier deferred to a future Replicator service.
• Service-name-only rename: InfoItem, info_item_id, InfoSpec, info_spec_id, /info-items/, info.changes, watcher columns — all unchanged. Only the service binary, repo path, systemd unit, port-8020 binding, and SDK package name (archiver_client) change.
• Notifier-style mirror for shared content-acquisition code (fetchers/, extractors/, simhash, extraction_defaults, logging). Discipline-based sync; revisit if drift bites.
• Fetcher microservice deferred indefinitely; per-service local fetcher boundary preserved for future swap.
• Separate Postgres database for archiver (archiver / archiver_test), not just separate schema. Watcher loses direct DB access to InfoItem/InfoSpec rows; SDK is the only path.
• Big-bang migration (pre-production state, no dual-stack window).
• Pre-extraction cleanup: relocate extraction_config from src/information/core/tools/ to src/core/extraction_defaults.py in each repo (fixes existing reverse-import from src/workers/pipeline.py).
• Watcher INFORMATION_* env vars renamed to ARCHIVER_*. Hard cut; no transitional period.
• Future model evolution and Replicator stand-up are explicitly out of scope here.

Related

• chore: design + planning — Notifier adapter (Phase 1 migration) #132 — Notifier extraction (architectural pattern)
• Phase 3a: Information service authoring tools #145 — Phase 3a authoring tools (most recent Information service work, moves with the extraction)
• docs/plans/2026-05-03-information-source-specifications-design.md — original Information service design (whose "Later: extract" step this implements)
• docs/research/2026-05-06-archiver-information-model.md — Archiver future-state research (carries into the new repo)

[gregoryfoster]

Shipped. Archiver service is live and authoritative for Information Items.

What landed

Watcher repo (this repo, commits de68546..0413e0a):

• f401687 — design doc + research carried into archiver
• f3263f7 — relocate extraction_config → src/core/extraction_defaults.py (severs reverse import from src/workers/pipeline.py)
• c20133a — delete src/information/, alembic_information/, clients/python/, deploy/information.service, tests/information/; rewrite all imports information_client → archiver_client; rename INFORMATION_* env vars to ARCHIVER_*; add tests/_information_test_models.py for cross-schema FK seeding
• 5ebd984 — merge to main
• b368da2 — docs sweep (AGENTS.md, COMMANDS.md)
• 3b89ebb — round-1 CR fixes: rename leftovers in log strings + smoke script comment, replace test schema with alembic-driven fixture (subprocess-invokes archiver's own alembic — schema drift now structurally impossible)
• 0413e0a — round-2 CR fixes: AGENTS.md "tests require sibling archiver repo" callout + improved CalledProcessError reporting in test fixture

Archiver repo (/home/exedev/archiver, fresh git init, 2 commits):

• 134df53 — initial commit lifted from watcher's src/information/
• 4c6401c — post-extraction CR sweep: FastAPI title "information" → "archiver", AUTOCOMMIT bootstrap in alembic env.py, SDK regenerated against renamed app, historical plan doc annotations

VM state

• archiver.service running on port 8020 from /home/exedev/archiver
• information.service disabled, unit removed from /etc/systemd/system/
• watcher.service restarted with archiver_client SDK consumed via path dep
• New Postgres DBs: archiver (prod) + archiver_test (test); role archiver
• Prototype data preserved: 9 InfoItems + 9 InfoSpecs migrated from watcher DB → archiver DB; watcher DB's information schema dropped after migration
• /etc/archiver/.env and /etc/watcher/.env env-var prefixes updated

Verification

• 680 tests passing in watcher (was 763 pre-extraction; 87 moved to archiver)
• 87 tests passing in archiver service
• 52 tests passing in archiver SDK (52)
• Lint + format clean both repos
• End-to-end smoke: watcher → archiver GET /api/v1/info-items returns 9 items via the existing API key
• Both services up on systemd; logs show archiver client pre-warmed (watcher) and archiver db engine created (archiver)

Follow-ups

• Optimize archiver alembic test-fixture (subprocess overhead + xdist safety) #150 — Optimize archiver alembic test-fixture (subprocess overhead + pytest-xdist safety). Tracked separately; not blocking.
• Archiver repo has no GitHub remote yet — kept local on the VM, mirroring the Notifier extraction pattern. When the service relocates off-VM, push to a real remote.
• One cosmetic note in clients/python/src/archiver_client/generated/types.py:30 ("Contains information for file uploads") — auto-generated openapi-python-client boilerplate, not a rename leftover.

Out-of-scope (deliberate)

• Future InfoSource / SourceSpec / SourceRevision content-addressed evolution lives in archiver's docs/research/2026-05-06-archiver-information-model.md for future v2 work.
• Replicator service stand-up — separate future work.
• Fetcher microservice — deferred indefinitely (mechanical swap when ready).