Add the PostgreSQL audit backend and a shared conformance suite
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 1s

AUDIT-WP-0005-T01, built and verified against PostgreSQL 16 locally in
Docker; the Railiance cluster was not needed.

tests/test_backend_conformance.py is one suite run against every backend, so
"the Postgres backend is done" means it satisfies the same contract SQLite
already does rather than having its own green tests. It skips cleanly with no
server reachable; make pg-test-up and make test-pg run it. Suite 50 -> 71.

RetentionPolicy declares immutable=True and earns it: migration 0002 installs
a trigger rejecting UPDATE and DELETE on the events table, so a leaked runtime
credential can append but cannot rewrite or erase the trail. That materially
narrows the residual risk ADR-0001 section 5 called out. tamper_evidence stays
False because nothing here would prove a database owner had dropped the
trigger - hash-chaining or external anchoring would be needed and is not
implemented.

Idempotency is one statement (INSERT ... ON CONFLICT DO NOTHING RETURNING),
verified to behave identically to the SQLite backend under 12 concurrent
submissions of the same event. Migrations are ordered, recorded and
idempotent. Replay reconciles rather than duplicating - the piece deferred out
of WP-0004-T05 - and is tested to leave exactly one custody record.

Backend selection is by AUDIT_CORE_DATABASE_URL; the SQLite fallback logs a
warning so a deployment that lost its URL is visible rather than quietly
running on the wrong store.

Also fixed: ingestion had no __main__ guard, so python -m audit_core.ingestion
silently did nothing. Found during end-to-end smoke.

Counting semantics documented: occurrences counts transmissions, not stored
events, so a retry of a secret-shaped field increments it again. That is the
sender behaviour being optimized away.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
tegwick 2026-08-10 17:09:46 +02:00
parent 576caa2665
commit bd274f6269
9 changed files with 806 additions and 9 deletions

View file

@ -3,7 +3,7 @@ RUN useradd --system --uid 10001 --create-home audit-core
WORKDIR /app
COPY pyproject.toml README.md LICENSE ./
COPY audit_core ./audit_core
RUN pip install --no-cache-dir ".[serve]"
RUN pip install --no-cache-dir ".[serve,postgres]"
USER 10001
EXPOSE 8080
CMD ["audit-core-ingest"]

View file

@ -21,4 +21,22 @@ help: ## Show this help
@awk 'BEGIN {FS = ":.*##"; printf "\nUsage:\n make \033[36m<target>\033[0m\n"} \
/^[a-zA-Z_-]+:.*?##/ { printf " \033[36m%-24s\033[0m %s\n", $$1, $$2 }' $(MAKEFILE_LIST)
.PHONY: test mock-audit-smoke mock-audit-cleanup help
.PHONY: test test-pg pg-test-up pg-test-down mock-audit-smoke mock-audit-cleanup help
PG_TEST_CONTAINER ?= ac-pg-test
PG_TEST_PORT ?= 55445
PG_TEST_URL ?= postgresql://postgres:test@127.0.0.1:$(PG_TEST_PORT)/audit_core
pg-test-up: ## Start a throwaway PostgreSQL for backend conformance tests
-docker rm -f $(PG_TEST_CONTAINER) 2>/dev/null
docker run -d --name $(PG_TEST_CONTAINER) -e POSTGRES_PASSWORD=test \
-e POSTGRES_DB=audit_core -p 127.0.0.1:$(PG_TEST_PORT):5432 postgres:16-alpine
@for i in $$(seq 1 40); do \
docker exec $(PG_TEST_CONTAINER) pg_isready -U postgres >/dev/null 2>&1 && exit 0; \
sleep 1; done; echo "postgres did not become ready" >&2; exit 1
pg-test-down: ## Remove the throwaway PostgreSQL
-docker rm -f $(PG_TEST_CONTAINER)
test-pg: ## Run the suite including PostgreSQL conformance (needs pg-test-up)
AUDIT_CORE_TEST_DATABASE_URL="$(PG_TEST_URL)" python3 -m pytest -q

View file

@ -11,7 +11,7 @@
| workplan | AUDIT-WP-0001 | finished | — | workplans/AUDIT-WP-0001-statehub-bootstrap.md |
| workplan | AUDIT-WP-0002 | finished | — | workplans/AUDIT-WP-0002-pluggable-audit-backend.md |
| workplan | AUDIT-WP-0003 | finished | — | workplans/AUDIT-WP-0003-user-engine-event-ingestion-service.md |
| workplan | AUDIT-WP-0004 | proposed | — | workplans/AUDIT-WP-0004-receiver-correctness-and-hardening.md |
| workplan | AUDIT-WP-0004 | finished | — | workplans/AUDIT-WP-0004-receiver-correctness-and-hardening.md |
| workplan | AUDIT-WP-0005 | proposed | — | workplans/AUDIT-WP-0005-postgres-store-and-production-deployment.md |
| task | AUDIT-WP-0001-T01 | done | — | workplans/AUDIT-WP-0001-statehub-bootstrap.md |
| task | AUDIT-WP-0001-T02 | done | — | workplans/AUDIT-WP-0001-statehub-bootstrap.md |
@ -24,7 +24,7 @@
| task | AUDIT-WP-0004-T01 | done | — | workplans/AUDIT-WP-0004-receiver-correctness-and-hardening.md |
| task | AUDIT-WP-0004-T02 | done | — | workplans/AUDIT-WP-0004-receiver-correctness-and-hardening.md |
| task | AUDIT-WP-0004-T03 | done | — | workplans/AUDIT-WP-0004-receiver-correctness-and-hardening.md |
| task | AUDIT-WP-0004-T04 | todo | — | workplans/AUDIT-WP-0004-receiver-correctness-and-hardening.md |
| task | AUDIT-WP-0004-T04 | done | — | workplans/AUDIT-WP-0004-receiver-correctness-and-hardening.md |
| task | AUDIT-WP-0004-T05 | done | — | workplans/AUDIT-WP-0004-receiver-correctness-and-hardening.md |
| task | AUDIT-WP-0004-T06 | done | — | workplans/AUDIT-WP-0004-receiver-correctness-and-hardening.md |
| task | AUDIT-WP-0004-T07 | done | — | workplans/AUDIT-WP-0004-receiver-correctness-and-hardening.md |

View file

@ -414,16 +414,44 @@ def _serve_fallback(app, host: str, port: int, timeout: int) -> None:
server.serve_forever()
def build_backend() -> IdempotentAuditBackend:
"""Select the custody backend from the environment.
``AUDIT_CORE_DATABASE_URL`` selects PostgreSQL (production custody).
Falling back to SQLite is explicit and logged, so a deployment that meant
to use Postgres and lost its URL is visible rather than quietly running on
the wrong store.
"""
url = os.environ.get("AUDIT_CORE_DATABASE_URL")
if url:
from audit_core.postgres_backend import PostgresAuditBackend
retention = os.environ.get("AUDIT_CORE_RETENTION_DAYS")
log.info("custody backend: postgresql")
return PostgresAuditBackend(
url,
schema=os.environ.get("AUDIT_CORE_DATABASE_SCHEMA", "audit_core"),
retention_days=int(retention) if retention else None,
max_size=int(os.environ.get("AUDIT_CORE_DB_POOL_MAX", "8")),
statement_timeout_ms=int(
os.environ.get("AUDIT_CORE_DB_STATEMENT_TIMEOUT_MS", "30000")
),
)
path = os.environ.get("AUDIT_CORE_DATABASE_PATH", "/data/audit-core.db")
log.warning(
"custody backend: sqlite at %s — AUDIT_CORE_DATABASE_URL is unset, so this "
"is not the production store", path,
)
return SQLiteAuditBackend(path)
def main() -> None:
logging.basicConfig(
level=os.environ.get("AUDIT_CORE_LOG_LEVEL", "INFO"),
format='{"ts":"%(asctime)s","level":"%(levelname)s","logger":"%(name)s","msg":"%(message)s"}',
stream=sys.stdout,
)
backend = SQLiteAuditBackend(
os.environ.get("AUDIT_CORE_DATABASE_PATH", "/data/audit-core.db")
)
app = IngestionApplication(backend, SenderRegistry.from_env())
app = IngestionApplication(build_backend(), SenderRegistry.from_env())
serve(
app,
host=os.environ.get("AUDIT_CORE_HOST", "0.0.0.0"),
@ -431,3 +459,7 @@ def main() -> None:
threads=int(os.environ.get("AUDIT_CORE_THREADS", "8")),
timeout=int(os.environ.get("AUDIT_CORE_REQUEST_TIMEOUT", "30")),
)
if __name__ == "__main__": # pragma: no cover
main()

View file

@ -0,0 +1,426 @@
"""Durable PostgreSQL audit backend (AUDIT-WP-0005-T01).
Production custody. Runs on the Railiance shared PostgreSQL platform
(`rapp-postgres`), inside the database-per-consumer boundary fixed by that
repo's ADR-0001.
Two properties are pushed into the database rather than the application:
* **Idempotency** ``INSERT ... ON CONFLICT DO NOTHING RETURNING`` resolves
insert-or-detect in one statement, so two concurrent submissions of the same
event cannot both be told they were first. The SQLite backend learned this
the hard way; see its test.
* **Append-only custody** a trigger rejects ``UPDATE`` and ``DELETE`` on the
events table, so a leaked runtime credential can add records but cannot
rewrite or erase existing ones. This is what lets ``RetentionPolicy`` declare
``immutable=True`` honestly; see :meth:`retention_policy` for its limits.
"""
from __future__ import annotations
import json
import os
from datetime import datetime, timezone
from typing import Any
from audit_core.interface import (
AcceptResult,
AuditEvent,
BackendUnavailableError,
EventConflictError,
EventValidationError,
RetentionPolicy,
validate_event,
)
from audit_core.redaction import Finding
try: # pragma: no cover - import guard
import psycopg
from psycopg_pool import ConnectionPool
except ImportError as exc: # pragma: no cover
raise ImportError(
"the postgres backend needs psycopg: pip install 'audit-core[postgres]'"
) from exc
DEFAULT_SCHEMA = "audit_core"
# Ordered, append-only. Each entry runs once and is recorded in
# schema_migrations. Never edit a released migration — add a new one.
MIGRATIONS: list[tuple[str, str]] = [
(
"0001-events",
"""
CREATE TABLE IF NOT EXISTS {schema}.events (
event_id text PRIMARY KEY,
payload_hash text NOT NULL,
accepted_at timestamptz NOT NULL DEFAULT now(),
observed_at timestamptz,
tenant text NOT NULL,
correlation_id text,
source text NOT NULL,
action text NOT NULL,
record jsonb NOT NULL
);
CREATE INDEX IF NOT EXISTS events_correlation_idx
ON {schema}.events (correlation_id);
-- Tenant keying is mandatory per business-app-service-contract 1.3;
-- the index makes per-tenant export and retention tractable.
CREATE INDEX IF NOT EXISTS events_tenant_idx ON {schema}.events (tenant);
""",
),
(
"0002-append-only",
"""
CREATE OR REPLACE FUNCTION {schema}.reject_mutation() RETURNS trigger AS $fn$
BEGIN
RAISE EXCEPTION 'audit events are append-only (attempted %)', TG_OP
USING ERRCODE = 'restrict_violation';
END;
$fn$ LANGUAGE plpgsql;
DROP TRIGGER IF EXISTS events_append_only ON {schema}.events;
CREATE TRIGGER events_append_only
BEFORE UPDATE OR DELETE ON {schema}.events
FOR EACH ROW EXECUTE FUNCTION {schema}.reject_mutation();
""",
),
(
"0003-dead-letters",
"""
CREATE TABLE IF NOT EXISTS {schema}.dead_letters (
id bigserial PRIMARY KEY,
event_id text,
received_at timestamptz NOT NULL DEFAULT now(),
sender text,
reason text NOT NULL,
payload_hash text NOT NULL,
payload text,
payload_withheld boolean NOT NULL DEFAULT false
);
CREATE INDEX IF NOT EXISTS dead_letters_event_idx
ON {schema}.dead_letters (event_id);
""",
),
(
"0004-secret-findings",
"""
CREATE TABLE IF NOT EXISTS {schema}.secret_findings (
sender text NOT NULL,
source text NOT NULL,
action text NOT NULL,
field_path text NOT NULL,
outcome text NOT NULL,
persisted boolean NOT NULL DEFAULT false,
occurrences bigint NOT NULL DEFAULT 0,
first_seen timestamptz NOT NULL DEFAULT now(),
last_seen timestamptz NOT NULL DEFAULT now(),
PRIMARY KEY (sender, source, action, field_path, outcome)
);
""",
),
]
# Rejection reasons whose payload must never be persisted — storing the body of
# an event rejected *for carrying secret-shaped material* would write that
# material into the audit store.
WITHHOLD_PAYLOAD_REASONS = frozenset({"secret_shaped_field"})
class PostgresAuditBackend:
"""Audit custody in PostgreSQL."""
def __init__(
self,
dsn: str | None = None,
*,
schema: str = DEFAULT_SCHEMA,
retention_days: int | None = None,
min_size: int = 1,
max_size: int = 8,
statement_timeout_ms: int = 30_000,
migrate: bool = True,
) -> None:
self.dsn = dsn or os.environ.get("AUDIT_CORE_DATABASE_URL") or ""
if not self.dsn:
raise ValueError("a DSN is required (AUDIT_CORE_DATABASE_URL)")
if not schema.isidentifier():
raise ValueError(f"unsafe schema name: {schema!r}")
self.schema = schema
self.retention_days = retention_days
try:
self.pool = ConnectionPool(
self.dsn,
min_size=min_size,
max_size=max_size,
kwargs={
"autocommit": True,
# A stalled write must surface as unavailable rather than
# hold a request open indefinitely.
"options": f"-c statement_timeout={int(statement_timeout_ms)}",
},
open=True,
timeout=10,
)
except Exception as exc: # psycopg raises a wide family here
raise BackendUnavailableError(f"cannot connect: {exc}") from exc
if migrate:
self.migrate()
# --- schema ------------------------------------------------------------
def migrate(self) -> list[str]:
"""Apply pending migrations. Returns the ids applied this call."""
applied: list[str] = []
try:
with self.pool.connection() as conn:
conn.execute(f'CREATE SCHEMA IF NOT EXISTS "{self.schema}"')
conn.execute(
f'CREATE TABLE IF NOT EXISTS "{self.schema}".schema_migrations ('
" id text PRIMARY KEY, applied_at timestamptz NOT NULL DEFAULT now())"
)
done = {
row[0]
for row in conn.execute(
f'SELECT id FROM "{self.schema}".schema_migrations'
).fetchall()
}
for migration_id, body in MIGRATIONS:
if migration_id in done:
continue
conn.execute(body.format(schema=f'"{self.schema}"'))
conn.execute(
f'INSERT INTO "{self.schema}".schema_migrations (id) VALUES (%s)',
(migration_id,),
)
applied.append(migration_id)
except psycopg.Error as exc:
raise BackendUnavailableError(f"migration failed: {exc}") from exc
return applied
@property
def _events(self) -> str:
return f'"{self.schema}".events'
# --- contract ----------------------------------------------------------
@property
def retention_policy(self) -> RetentionPolicy:
"""Declared custody guarantees.
``immutable`` is True because migration 0002 installs a trigger that
rejects UPDATE and DELETE, so no consumer credential can alter a stored
record. It is not a claim against the database owner or a superuser,
who can drop the trigger; ``tamper_evidence`` is correspondingly False,
because nothing here would *prove* they had. Hash-chaining or external
anchoring would be needed for that, and is not implemented.
"""
return RetentionPolicy(
custody_class="archive",
retention_days=self.retention_days,
immutable=True,
tamper_evidence=False,
durable=True,
)
def emit(self, event: AuditEvent) -> str:
return self.accept(event, payload_hash=_record_hash(event)).reference
def accept(self, event: AuditEvent, payload_hash: str) -> AcceptResult:
validate_event(event)
reference = f"audit:{event.event_id}"
details = event.details if isinstance(event.details, dict) else {}
record = event.as_record()
try:
with self.pool.connection() as conn:
inserted = conn.execute(
f"""
INSERT INTO {self._events}
(event_id, payload_hash, observed_at, tenant, correlation_id,
source, action, record)
VALUES (%s, %s, %s, %s, %s, %s, %s, %s)
ON CONFLICT (event_id) DO NOTHING
RETURNING event_id
""",
(
event.event_id,
payload_hash,
_timestamp(event.observed_at),
event.tenant,
str(details.get("correlation_id") or "") or None,
event.source,
event.action,
json.dumps(record, sort_keys=True),
),
).fetchone()
if inserted is not None:
return AcceptResult(duplicate=False, reference=reference)
existing = conn.execute(
f"SELECT payload_hash FROM {self._events} WHERE event_id = %s",
(event.event_id,),
).fetchone()
except psycopg.Error as exc:
raise BackendUnavailableError(str(exc)) from exc
if existing is None:
raise BackendUnavailableError("event disappeared during accept")
if existing[0] != payload_hash:
raise EventConflictError(
f"event_id {event.event_id} already held with a different payload"
)
return AcceptResult(duplicate=True, reference=reference)
# --- operator surface --------------------------------------------------
def get(self, event_id: str) -> dict | None:
rows = self._query(
f"SELECT record, accepted_at FROM {self._events} WHERE event_id = %s",
(event_id,),
)
if not rows:
return None
return {"accepted_at": _iso(rows[0][1]), **rows[0][0]}
def by_correlation(self, correlation_id: str, limit: int = 100) -> list[dict]:
rows = self._query(
f"SELECT record, accepted_at FROM {self._events} "
"WHERE correlation_id = %s ORDER BY accepted_at, event_id LIMIT %s",
(correlation_id, int(limit)),
)
return [{"accepted_at": _iso(at), **rec} for rec, at in rows]
def replay(self, event_id: str) -> AcceptResult:
"""Re-submit a stored event through :meth:`accept`.
Reconciliation, not re-creation: replaying an event already in custody
must return ``duplicate=True`` against the same record. If it were to
report a first acceptance, the store would be producing a second
custody record for one source event the exact failure the whole
idempotency design exists to prevent.
"""
rows = self._query(
f"SELECT record, payload_hash FROM {self._events} WHERE event_id = %s",
(event_id,),
)
if not rows:
raise KeyError(event_id)
record, payload_hash = rows[0]
return self.accept(_event_from_record(record), payload_hash)
def record_rejection(
self,
*,
event_id: str | None,
reason: str,
payload_hash: str,
sender: str | None = None,
payload: str | None = None,
) -> None:
withheld = reason in WITHHOLD_PAYLOAD_REASONS
self._execute(
f'INSERT INTO "{self.schema}".dead_letters '
"(event_id, sender, reason, payload_hash, payload, payload_withheld) "
"VALUES (%s, %s, %s, %s, %s, %s)",
(event_id, sender, reason, payload_hash,
None if withheld else payload, withheld),
)
def dead_letters(self, limit: int = 100) -> list[dict]:
rows = self._query(
"SELECT event_id, received_at, sender, reason, payload_hash, payload, "
f'payload_withheld FROM "{self.schema}".dead_letters '
"ORDER BY id DESC LIMIT %s",
(int(limit),),
)
return [
{
"event_id": r[0], "received_at": _iso(r[1]), "sender": r[2],
"reason": r[3], "payload_hash": r[4], "payload": r[5],
"payload_withheld": bool(r[6]),
}
for r in rows
]
def count_secret_findings(
self, *, sender: str, source: str, action: str, outcome: str, findings
) -> None:
for finding in findings:
self._execute(
f'INSERT INTO "{self.schema}".secret_findings '
"(sender, source, action, field_path, outcome, persisted, occurrences) "
"VALUES (%s, %s, %s, %s, %s, %s, 1) "
"ON CONFLICT (sender, source, action, field_path, outcome) DO UPDATE "
"SET occurrences = secret_findings.occurrences + 1, last_seen = now()",
(sender, source, action, finding.path, outcome,
bool(getattr(finding, "in_persisted_data", False))),
)
def secret_findings(self, limit: int = 100) -> list[dict]:
rows = self._query(
"SELECT sender, source, action, field_path, outcome, persisted, "
f'occurrences, first_seen, last_seen FROM "{self.schema}".secret_findings '
"ORDER BY occurrences DESC, last_seen DESC LIMIT %s",
(int(limit),),
)
return [
{
"sender": r[0], "source": r[1], "action": r[2], "field_path": r[3],
"outcome": r[4], "persisted": bool(r[5]), "occurrences": r[6],
"first_seen": _iso(r[7]), "last_seen": _iso(r[8]),
}
for r in rows
]
# --- lifecycle ---------------------------------------------------------
def health(self) -> None:
self._query("SELECT 1", ())
def close(self) -> None:
self.pool.close()
def _query(self, sql: str, params: tuple) -> list:
try:
with self.pool.connection() as conn:
return conn.execute(sql, params).fetchall()
except psycopg.Error as exc:
raise BackendUnavailableError(str(exc)) from exc
def _execute(self, sql: str, params: tuple) -> None:
try:
with self.pool.connection() as conn:
conn.execute(sql, params)
except psycopg.Error as exc:
raise BackendUnavailableError(str(exc)) from exc
def _timestamp(value: str | None) -> datetime | None:
if not value:
return None
try:
return datetime.fromisoformat(str(value).replace("Z", "+00:00"))
except ValueError:
return None
def _iso(value: Any) -> str | None:
if isinstance(value, datetime):
return value.astimezone(timezone.utc).replace(microsecond=0).isoformat()
return value
def _event_from_record(record: dict) -> AuditEvent:
return AuditEvent(
source=record["source"], action=record["action"], resource=record["resource"],
outcome=record["outcome"], tenant=record["tenant"], scope=record["scope"],
actor=record.get("actor"), reason=record.get("reason"),
details=record.get("details") or {}, event_id=record["event_id"],
observed_at=record["observed_at"], schema_version=record["schema_version"],
)
def _record_hash(event: AuditEvent) -> str:
import hashlib
return hashlib.sha256(
json.dumps(event.as_record(), sort_keys=True).encode("utf-8")
).hexdigest()

View file

@ -296,6 +296,12 @@ Read them at `GET /v1/secret-findings` (requires the read privilege):
"first_seen": "...", "last_seen": "..."}]}
```
`occurrences` counts **transmissions, not stored events**: a retry resubmitting
the same secret-shaped field increments it again, even though the event
reconciles as a duplicate and produces no second custody record. That is
deliberate — the number measures how often the sender emitted the field, which
is the behaviour being optimized away.
`persisted` distinguishes a field that reached the stored record from one that
sat elsewhere in the envelope and was dropped by normalization anyway. A
healthy sender trends to zero occurrences; a non-empty list is a backlog item

View file

@ -14,6 +14,8 @@ dev = ["pytest"]
# Production serving. Without this the entrypoint falls back to a threaded
# wsgiref server, which is bounded but not a production server (AUDIT-WP-0004-T06).
serve = ["waitress>=3.0"]
# Production custody (AUDIT-WP-0005). SQLite remains the development backend.
postgres = ["psycopg[binary,pool]>=3.2"]
[project.scripts]
audit-core-ingest = "audit_core.ingestion:main"

View file

@ -0,0 +1,278 @@
"""Behaviour every durable audit backend must satisfy.
One suite, run against every backend. This is what makes "the Postgres backend
is done" mean something: it is the same contract SQLite already passes, not a
parallel set of tests that happen to be green.
Postgres tests are skipped unless a server is reachable. Point
``AUDIT_CORE_TEST_DATABASE_URL`` at one, or run ``make pg-test-up`` to start a
throwaway container.
"""
from __future__ import annotations
import hashlib
import json
import os
import threading
import uuid
import pytest
from audit_core.interface import (
AuditEvent,
BackendUnavailableError,
EventConflictError,
)
from audit_core.redaction import Finding
from audit_core.sqlite_backend import SQLiteAuditBackend
PG_URL = os.environ.get("AUDIT_CORE_TEST_DATABASE_URL")
def _postgres_available() -> bool:
if not PG_URL:
return False
try:
import psycopg
with psycopg.connect(PG_URL, connect_timeout=3):
return True
except Exception:
return False
HAVE_PG = _postgres_available()
@pytest.fixture(params=["sqlite", "postgres"])
def backend(request, tmp_path):
if request.param == "sqlite":
yield SQLiteAuditBackend(str(tmp_path / "conformance.db"))
return
if not HAVE_PG:
pytest.skip("no PostgreSQL reachable (set AUDIT_CORE_TEST_DATABASE_URL)")
from audit_core.postgres_backend import PostgresAuditBackend
schema = f"conf_{uuid.uuid4().hex[:12]}"
instance = PostgresAuditBackend(PG_URL, schema=schema)
try:
yield instance
finally:
with instance.pool.connection() as conn:
conn.execute(f'DROP SCHEMA IF EXISTS "{schema}" CASCADE')
instance.close()
def make_event(event_id="evt-1", **kw):
fields = dict(
event_id=event_id, source="user-engine", action="membership.added",
resource="membership-1", outcome="recorded", tenant="tenant:friendly:binky",
scope="tenant", details={"correlation_id": "corr-1", "data": {"a": 1}},
observed_at="2026-08-09T00:00:00+00:00",
)
fields.update(kw)
return AuditEvent(**fields)
def digest(event: AuditEvent) -> str:
return hashlib.sha256(
json.dumps(event.as_record(), sort_keys=True).encode()
).hexdigest()
# --- the custody contract ---------------------------------------------------
def test_declares_a_retention_policy(backend):
policy = backend.retention_policy
assert policy.durable is True
assert policy.custody_class in ("development", "archive", "hot_search")
# A backend claiming tamper evidence must also claim immutability;
# the reverse is allowed.
if policy.tamper_evidence:
assert policy.immutable
def test_accepts_then_reports_duplicate(backend):
event = make_event()
first = backend.accept(event, digest(event))
assert first.duplicate is False
second = backend.accept(event, digest(event))
assert second.duplicate is True
assert second.reference == first.reference
def test_same_id_different_payload_conflicts(backend):
event = make_event()
backend.accept(event, digest(event))
with pytest.raises(EventConflictError):
backend.accept(event, "a-different-hash")
def test_rejects_an_invalid_event(backend):
from audit_core.interface import EventValidationError
with pytest.raises((EventValidationError, ValueError)):
backend.accept(make_event(tenant=""), "hash")
# --- concurrency ------------------------------------------------------------
def test_concurrent_accept_yields_exactly_one_first(backend):
"""The assertion the service rests on, checked per backend.
An early SQLite implementation passed every serial test while telling two
concurrent callers they were both first, so this is not theoretical.
"""
event = make_event("race-1")
payload_hash = digest(event)
results: list = []
lock = threading.Lock()
barrier = threading.Barrier(12)
def submit():
barrier.wait()
try:
outcome = backend.accept(event, payload_hash).duplicate
except Exception as exc: # recorded, not swallowed
outcome = type(exc).__name__
with lock:
results.append(outcome)
threads = [threading.Thread(target=submit) for _ in range(12)]
for t in threads:
t.start()
for t in threads:
t.join()
assert results.count(False) == 1, results
assert results.count(True) == 11, results
# --- durability and reads ---------------------------------------------------
def test_lookup_by_event_id_and_correlation(backend):
backend.accept(make_event("e1"), "h1")
backend.accept(make_event("e2"), "h2")
record = backend.get("e1")
assert record["event_id"] == "e1"
assert record["tenant"] == "tenant:friendly:binky"
assert record["accepted_at"]
assert backend.get("missing") is None
assert {e["event_id"] for e in backend.by_correlation("corr-1")} == {"e1", "e2"}
def test_dead_letters_withhold_secret_payloads(backend):
backend.record_rejection(
event_id="e9", reason="secret_shaped_field", payload_hash="h",
sender="user-engine", payload='{"password":"hunter2"}',
)
backend.record_rejection(
event_id="e8", reason="source_not_allowed", payload_hash="h2",
sender="user-engine", payload='{"source":"nope"}',
)
entries = {d["event_id"]: d for d in backend.dead_letters()}
assert entries["e9"]["payload"] is None
assert entries["e9"]["payload_withheld"] is True
assert entries["e8"]["payload"] is not None
assert entries["e8"]["payload_withheld"] is False
def test_secret_findings_count_per_path(backend):
findings = [Finding("data.auth_token", True)]
for _ in range(3):
backend.count_secret_findings(
sender="user-engine", source="user-engine",
action="membership.added", outcome="redacted", findings=findings,
)
row = backend.secret_findings()[0]
assert row["field_path"] == "data.auth_token"
assert row["occurrences"] == 3
assert row["persisted"] is True
def test_health_passes_on_a_live_backend(backend):
backend.health()
# --- postgres-specific guarantees -------------------------------------------
pg_only = pytest.mark.skipif(not HAVE_PG, reason="needs PostgreSQL")
@pg_only
def test_replay_reconciles_rather_than_duplicating():
"""Replay must never mint a second custody record for one source event."""
from audit_core.postgres_backend import PostgresAuditBackend
schema = f"conf_{uuid.uuid4().hex[:12]}"
backend = PostgresAuditBackend(PG_URL, schema=schema)
try:
event = make_event("replay-1")
backend.accept(event, digest(event))
outcome = backend.replay("replay-1")
assert outcome.duplicate is True
rows = backend._query(
f'SELECT count(*) FROM "{schema}".events WHERE event_id = %s', ("replay-1",)
)
assert rows[0][0] == 1
with pytest.raises(KeyError):
backend.replay("never-stored")
finally:
with backend.pool.connection() as conn:
conn.execute(f'DROP SCHEMA IF EXISTS "{schema}" CASCADE')
backend.close()
@pg_only
def test_stored_events_are_append_only():
"""The basis for declaring immutable=True.
Without this, a leaked runtime credential could rewrite or erase the audit
trail and the retention policy would be claiming a guarantee it does not
have.
"""
from audit_core.postgres_backend import PostgresAuditBackend
schema = f"conf_{uuid.uuid4().hex[:12]}"
backend = PostgresAuditBackend(PG_URL, schema=schema)
try:
event = make_event("immutable-1")
backend.accept(event, digest(event))
assert backend.retention_policy.immutable is True
with pytest.raises(BackendUnavailableError):
backend._execute(
f'UPDATE "{schema}".events SET tenant = %s WHERE event_id = %s',
("tenant:coulomb", "immutable-1"),
)
with pytest.raises(BackendUnavailableError):
backend._execute(
f'DELETE FROM "{schema}".events WHERE event_id = %s', ("immutable-1",)
)
assert backend.get("immutable-1")["tenant"] == "tenant:friendly:binky"
finally:
with backend.pool.connection() as conn:
conn.execute(f'DROP SCHEMA IF EXISTS "{schema}" CASCADE')
backend.close()
@pg_only
def test_migrations_are_idempotent_and_recorded():
from audit_core.postgres_backend import MIGRATIONS, PostgresAuditBackend
schema = f"conf_{uuid.uuid4().hex[:12]}"
backend = PostgresAuditBackend(PG_URL, schema=schema, migrate=False)
try:
applied = backend.migrate()
assert applied == [m[0] for m in MIGRATIONS]
assert backend.migrate() == [] # second call is a no-op
finally:
with backend.pool.connection() as conn:
conn.execute(f'DROP SCHEMA IF EXISTS "{schema}" CASCADE')
backend.close()

View file

@ -49,7 +49,7 @@ requirement and consumes it; it does not implement it here.
```task
id: AUDIT-WP-0005-T01
status: todo
status: done
priority: high
state_hub_task_id: "b1601d0b-922a-40f7-92c0-ea06af6c4468"
```
@ -76,6 +76,41 @@ Done when the backend passes the same contract tests as the existing backends,
concurrent duplicate submissions produce exactly one record, and a database
restart mid-write does not produce an acknowledged-but-absent event.
Done 2026-08-10, built and verified against PostgreSQL 16 locally in Docker —
the Railiance cluster was not needed for any of it.
`tests/test_backend_conformance.py` is a single suite run against every
backend, so "the Postgres backend is done" means it satisfies the same
contract SQLite already does rather than having its own tests that happen to
be green. It skips cleanly when no server is reachable; `make pg-test-up` and
`make test-pg` run it. Suite 50 -> 71.
`RetentionPolicy` declares `immutable=True`, and that is earned: migration
0002 installs a trigger rejecting UPDATE and DELETE on the events table, so a
leaked runtime credential can append but cannot rewrite or erase the trail.
This materially narrows the residual risk ADR-0001 §5 called out — a leaked
credential could previously forge the audit record. `tamper_evidence` stays
False, because nothing here would *prove* a database owner had dropped the
trigger; hash-chaining or external anchoring would be needed and is not
implemented.
Migrations are ordered, recorded in `schema_migrations`, and idempotent.
Replay reconciles rather than duplicating — the piece deferred out of
WP-0004-T05 — and is tested to leave exactly one custody record.
Backend selection is by `AUDIT_CORE_DATABASE_URL`; falling back to SQLite logs
a warning, so a deployment that lost its URL is visible rather than quietly
running on the wrong store. End-to-end smoke through waitress against Postgres
confirmed accept, duplicate, cross-tenant refusal, read/write privilege
separation, visible redaction, secret-finding counters, and readiness
reporting `custody_class=archive`.
Also fixed: `audit_core.ingestion` had no `__main__` guard, so `python -m
audit_core.ingestion` silently did nothing.
Not covered here: behaviour across a real database failover, which needs the
cluster and belongs to T05.
## T02 - Provision storage through the platform lane
```task