[AZ-304] C6 Postgres schema: additive 0002 migration + UUIDv5

Strictly additive Alembic migration on the AZ-263 baseline (data_model
.md § 6.1 / § 6.3): six new tiles columns (tile_uuid UNIQUE,
location_hash, content_sha256, disk_bytes, accessed_at, uploaded_at),
four new btree indices, one UNIQUE expression index over the
COALESCE-zero-uuid natural key, CHECK widening of
ck_tiles_freshness_status to the AZ-263 + AZ-303 vocabulary UNION,
four NULLable bbox columns on sector_classifications, and a new
tile_freshness_rules table seeded with the two default thresholds.

Pinned UUIDv5 namespace (TILE_NAMESPACE_UUID =
5b8d0c2e-1a4f-4b3a-8c9d-e7f6a3b2c1d0) + derive_tile_id /
derive_location_hash helpers cross-coordinated with
satellite-provider. Migration runner apply_migrations(config) drives
Alembic command.upgrade("head") against the AZ-263 env with one
retry on PG SQLSTATE 40001 and structured INFO logs on apply / no-op.

Contract bump tile_metadata_store.md v1.1.0 -> v1.2.0 adds
TileMetadata.location_hash: UUID | None = None (non-breaking).
module-layout.md updated so c6_tile_cache explicitly Owns
db/migrations/**.

Tier-1 tests: UUIDv5 determinism + locked vectors + DSN resolution +
retry mocked DBAPIError -> 1180 passed, 32 skipped. Tier-2 docker
schema tests gated by @pytest.mark.docker run against the existing
docker-compose.test.yml db service.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Oleksandr Bezdieniezhnykh
2026-05-12 17:05:41 +03:00
parent 21f5a30d09
commit dde838d2cc
15 changed files with 2084 additions and 25 deletions
@@ -0,0 +1,236 @@
# Batch 27 / Cycle 1 — Implementation Report
**Date**: 2026-05-12
**Tasks**: AZ-304 (C6 Postgres schema — additive 0002 migration: identity, LRU, freshness rules)
**Story points landed**: 3
**Status**: complete (AZ-304 → In Testing)
## Scope summary
Single-task batch — final queued 1-pointer of cycle 1. AZ-304 ships the
strictly-additive `0002_c6_tile_identity_and_lru.py` migration on top of
the AZ-263 baseline, the pinned UUIDv5 namespace + derivation helpers
cross-coordinated with `satellite-provider`, the migration runner the
composition root invokes at startup, and bumps the AZ-303
`TileMetadataStore` contract v1.1.0 → v1.2.0 by adding the
`TileMetadata.location_hash: UUID | None` field.
Before implementation, the spec was rewritten under user-approved
Option A to drop the renames / retypes / drops the original draft
called for. The as-built migration only adds columns, indices, the
`tile_freshness_rules` table, and widens (loosens) the
`ck_tiles_freshness_status` CHECK to the UNION of AZ-263 and AZ-303
vocabularies. No AZ-263 column, table, index, or CHECK is renamed,
retyped, or dropped (data_model.md § 6.1 / § 6.3).
## Files added / modified
### New (production)
- `src/gps_denied_onboard/components/c6_tile_cache/_uuid_namespace.py`
pinned `TILE_NAMESPACE_UUID = 5b8d0c2e-1a4f-4b3a-8c9d-e7f6a3b2c1d0`,
`derive_tile_id(zoom, x, y, source, flight_id)` (UUIDv5 over
`"{z}/{x}/{y}/{source}/{flight_id_or_zero_uuid}"`),
`derive_location_hash(z, x, y)`. Cross-workspace coordination point
with `satellite-provider/SatelliteProvider.Common/Utils/Uuidv5.cs`.
- `src/gps_denied_onboard/components/c6_tile_cache/migrations.py`
`apply_migrations(config) -> MigrationResult` runner, frozen
`MigrationResult` dataclass, `MigrationError`. Drives Alembic
`command.upgrade(cfg, "head")` against the AZ-263 env at
`db/migrations/`; one retry on PG SQLSTATE `40001`
(serialization failure); structured INFO log on apply / no-op with
the resolved namespace UUID emitted for post-mortem drift detection.
- `db/migrations/versions/0002_c6_tile_identity_and_lru.py` — additive
Alembic migration: six new `tiles` columns (`tile_uuid` UNIQUE,
`location_hash`, `content_sha256` w/ length-64 CHECK, `disk_bytes`
w/ nonneg CHECK, `accessed_at`, `uploaded_at`); four new btree
indices on `tiles`; one UNIQUE expression index
`idx_tiles_natural_key` over the COALESCE-zero-uuid natural key;
CHECK-widening of `ck_tiles_freshness_status` to the AZ-263 +
AZ-303 UNION vocabulary; four NULLable bbox columns on
`sector_classifications`; new `tile_freshness_rules` table seeded
with the two default thresholds (active_conflict / reject @ 6m,
stable_rear / downgrade @ 12m). Reverse `downgrade()` drops every
addition and restores the AZ-263 freshness CHECK to its original
vocabulary.
### Modified (production)
- `src/gps_denied_onboard/components/c6_tile_cache/_types.py`
`TileMetadata.location_hash: UUID | None = None` (positional last,
default `None` so AZ-303 v1.1.0 constructors keep working; the
impl-task's `PostgresFilesystemStore.insert_metadata` derives the
value when `None` and the DB NOT-NULL constraint is the safety
net).
- `db/migrations/env.py` — docstring tightened to document the
`target_metadata` deferral (we use Alembic `op.*` directly; ORM
models land in a later cycle).
### New (tests)
- `tests/unit/c6_tile_cache/test_uuid_namespace.py` — 8 deterministic
Tier-1 tests covering AC-10 (5 locked tile-id vectors, 2 locked
location-hash vectors, idempotent calls, `TileSource` enum /
`UUID`-typed / `None`-flight inputs, malformed-flight rejection)
and AC-11 (location-hash invariance across source/flight, distinct
cells distinct hashes).
- `tests/unit/c6_tile_cache/test_postgres_schema.py` — `@pytest.mark.
docker` integration tests covering AC-1 through AC-9 + NFR-perf-
apply / NFR-perf-noop + `MigrationResult` frozen smoke. Reads
`DB_URL` env var; consumes the `db` service from `docker-compose.
test.yml`. Auto-skipped on Tier-1 by `tests/conftest.py`.
- `tests/unit/c6_tile_cache/test_migrations_runner.py` — 5 Tier-1
tests covering DSN resolution (config-block, env fallback, missing
raises `MigrationError`) and the SQLSTATE-40001 retry / terminal
paths (mocked DBAPIError to keep them deterministic and DB-less).
- `tests/unit/c6_tile_cache/fixtures/c6_postgres_schema_v2.sql` —
copy-pastable DDL fixture the AC-3 schema-shape diff test diffs
against. Includes every AZ-263 column / index / CHECK plus every
additive item from this migration plus the two seeded
`tile_freshness_rules` rows.
### Modified (tests + docs)
- `tests/unit/test_ac5_alembic.py` — head-revision assertion bumped
from `0001_initial` to `0002_c6_tile_identity_and_lru`; function
renamed `test_head_revision_is_0001_initial` →
`test_head_revision_matches_latest_migration`.
- `_docs/02_document/contracts/c6_tile_cache/tile_metadata_store.md`
— version bumped v1.1.0 → v1.2.0; new `TileMetadata.location_hash`
section; `sector_boundaries` → `sector_classifications`
corrections; Alembic migration path corrected from
`c6_tile_cache/_alembic/` to `db/migrations/versions/`; Change Log
v1.2.0 entry.
- `_docs/02_document/module-layout.md` — `c6_tile_cache` component
now explicitly `Owns` `db/migrations/**` (Alembic env was already
living there since AZ-263 bootstrap, but not documented).
`migrations.py` / `_uuid_namespace.py` listed under the component's
internal modules.
- `_docs/02_tasks/todo/AZ-304_c6_postgres_schema.md` → moved to
`_docs/02_tasks/done/`. Spec was rewritten earlier this batch
(Option A: strictly-additive 0002 + descope of
`psycopg_pool`/`connection.py` to AZ-305 + replacement of
`testcontainers` with the existing `@pytest.mark.docker` /
`docker-compose.test.yml` infra).
## Design decisions
1. **Strictly additive 0002, no rename / retype / drop.** User picked
Option A on the schema-reconciliation gate. Every AZ-263 column,
table, index, and CHECK survives 0002 byte-identical. The single
constraint mutation is the `ck_tiles_freshness_status` CHECK
widening, which is a semantic loosening (more values accepted) and
therefore additive per `data_model.md` § 6.1. Legacy AZ-263 values
(`stale_warn`, `stale_reject`) remain valid until an ADR-gated
future cycle deprecates them. DTO ↔ column mapping
(`freshness_label` ↔ `freshness_status`, `quality_metadata` ↔
`tile_quality_metadata`, `(lat, lon)` ↔ `(latitude, longitude)`)
is owned by the future `PostgresFilesystemStore` task (AZ-305) —
the schema does NOT pivot to DTO field names.
2. **UUIDv5 namespace pinned in code, not config.** `TILE_NAMESPACE_
UUID = 5b8d0c2e-1a4f-4b3a-8c9d-e7f6a3b2c1d0` lives in the
`_uuid_namespace` module as a `Final` constant — not in
environment / config / DB — so cross-workspace coordination with
the C# `satellite-provider` reduces to a code-level review check
on either side. The runner logs the namespace value on every
apply / no-op so production drift would surface immediately.
3. **`psycopg_pool` deferred to AZ-305.** The original spec assumed
AZ-263 pinned `psycopg_pool`; reality check showed only
`psycopg[binary]>=3.1` and `alembic>=1.13` are pinned. A new
third-party dependency would inflate this task's scope and risk;
moved to AZ-305 (`c6_postgres_filesystem_store`) where the
runtime connection pool is genuinely needed. The migration runner
uses Alembic's existing `NullPool` SQLAlchemy engine wired by
AZ-263 — no pool needed for one-shot startup migrations.
4. **Retry-without-sleep on serialization failure.** The runner
retries once on PG SQLSTATE `40001`, but without a `time.sleep`
backoff. Component files are forbidden from calling `time.*`
directly (`tests/_meta/test_no_direct_time_in_components.py`,
Invariant 2 of the replay contract), and migrations run before the
injected `Clock` is constructed, so `Clock.sleep` is unavailable
too. Alembic's `NullPool` opens a fresh connection on retry, which
already introduces natural jitter; a 0-50 ms backoff buys nothing
measurable. Documented in `migrations.py` inline.
5. **`MigrationError` NOT a member of `TileCacheError` family.**
Migrations run BEFORE the runtime error consumer (`c6_tile_cache.
errors`) is constructed — making `MigrationError` a `TileCacheError`
would create a forward-import cycle and conceptually mis-place a
startup-phase failure inside a runtime-phase error tree. Documented
in the runner module docstring + verified by
`test_migration_error_is_not_tile_cache_error`.
6. **`module-layout.md` updated as scope-creep, user-approved.** Drift
gate surfaced that `module-layout.md` referenced a `_alembic/`
directory and `.sql` migration files; reality is `db/migrations/`
`.py` files and the doc didn't claim ownership of any of it. User
picked Option A to expand AZ-304 scope by one doc edit:
`c6_tile_cache` now explicitly `Owns` `db/migrations/**`.
## AC coverage
| AC | Test name(s) | Status |
|----|--------------|--------|
| AC-1 | `test_ac1_apply_to_az263_baseline_advances_alembic_version` | Tier-2 (docker) |
| AC-2 | `test_ac2_apply_at_head_is_no_op` | Tier-2 (docker) |
| AC-3 | `test_ac3_schema_shape_diffs_clean_against_fixture` | Tier-2 (docker) |
| AC-4 | `test_ac4_per_flight_separation_allowed` | Tier-2 (docker) |
| AC-4b | `test_ac4b_duplicate_natural_key_rejected` + `..._googlemaps` | Tier-2 (docker) |
| AC-5 | `test_ac5_freshness_check_widening_accepts_union_vocabulary` | Tier-2 (docker) |
| AC-6 | `test_ac6_downgrade_reverses_cleanly` | Tier-2 (docker) |
| AC-7 | `test_ac7_default_freshness_rules_seeded` | Tier-2 (docker) |
| AC-8 | `test_ac8_runner_logs_info_on_apply_and_no_op` | Tier-2 (docker) |
| AC-9 | `test_ac9_az263_columns_byte_identical_after_upgrade` | Tier-2 (docker) |
| AC-10 | `test_ac10_namespace_locked_and_locked_vectors_match` + companions | Tier-1 passing |
| AC-11 | `test_ac11_location_hash_invariant_across_source_and_flight_id` + companion | Tier-1 passing |
| AC-12 | covered by AZ-303 `test_protocol_conformance.py` (DTO default still `None`) | Tier-1 passing |
| NFR-perf-apply | `test_nfr_perf_apply_under_5s_on_baselined_empty_db` | Tier-2 (docker) |
| NFR-perf-noop | `test_nfr_perf_noop_under_100ms` | Tier-2 (docker) |
| NFR-reliability-retry | `test_nfr_reliability_retry_once_on_serialization_failure` + companion | Tier-1 passing |
The Tier-2 docker tests are explicitly skipped on Tier-1 by the
project-wide `tests/conftest.py` mark-filter; they run locally via
`docker compose -f docker-compose.test.yml up -d db && GPS_DENIED_TIER=2
DB_URL=postgresql://gps_denied:dev@localhost:5432/gps_denied pytest
tests/unit/c6_tile_cache/test_postgres_schema.py`.
## Test run
`python3 -m pytest tests/ -q` → **1180 passed, 32 skipped (Tier-2 /
CUDA / cmake / actionlint / docker), 1 warning** in 21.77 s, no
failures.
Mypy strict on production + test modules: clean.
Ruff + ruff format on the modified set: clean.
## Self-review
- Production code: deterministic UUIDv5 with locked namespace +
documented cross-repo invariants; no `time.*` calls in component
files (Invariant 2 meta-test passes); no new third-party deps;
migration is byte-additive on the AZ-263 baseline; runner emits
structured INFO logs with `kind` + `kv` per the AZ-266 log schema;
`MigrationError` correctly outside the `TileCacheError` family.
- Tests: every AC has at least one named assertion (Tier-1 or Tier-2);
cross-repo coordination vectors are locked with documented expected
UUIDs; no hidden runtime dependencies in Tier-1 (mocked DBAPIError
for the retry path).
- Lint / type: ruff + mypy strict clean on the modified set.
- Docs: AZ-304 spec moved to `done/`; contract bumped v1.2.0;
`module-layout.md` reconciled with reality.
## Known gaps
- Tier-2 Postgres tests need a real `db` service; rolled forward to
the Tier-2 validation pass.
- `psycopg_pool` connection helper deferred to AZ-305 — documented in
both this report and the AZ-304 spec.
- `target_metadata` wiring in `db/migrations/env.py` deferred until
the first task introduces SQLAlchemy ORM models (none today,
none in AZ-305 as currently scoped).
- AZ-263 legacy CHECK values (`stale_warn`, `stale_reject`) remain
valid in `ck_tiles_freshness_status`; deprecation requires a future
ADR-gated cleanup migration.