Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
"""Authenticated, idempotent HTTP ingestion for user-engine outbox events.
|
|
|
|
|
|
|
|
|
|
Response contract (AUDIT-WP-0004-T02). Senders key their retry behaviour off
|
|
|
|
|
these, so they are part of the interface, not an implementation detail:
|
|
|
|
|
|
|
|
|
|
=== ========== =========================================================
|
|
|
|
|
202 accepted Event is durably in custody. Do not retry.
|
|
|
|
|
200 duplicate Exact resubmission of an event already in custody. Do not
|
|
|
|
|
retry; delivery already succeeded.
|
|
|
|
|
400 rejected Malformed or disallowed. Retrying will not help — dead
|
|
|
|
|
letter it.
|
|
|
|
|
401 unauthorized Credential missing or invalid. Do not retry without a
|
|
|
|
|
new credential.
|
|
|
|
|
409 conflict The event id is held with a different payload. Retrying
|
|
|
|
|
will not help; this indicates a sender bug or id reuse.
|
|
|
|
|
503 unavailable Not accepted, but retryable. Retry with backoff.
|
|
|
|
|
500 error Unexpected fault. Not accepted. Retryable with backoff.
|
|
|
|
|
=== ========== =========================================================
|
|
|
|
|
"""
|
2026-08-09 20:59:14 +02:00
|
|
|
|
|
|
|
|
from __future__ import annotations
|
|
|
|
|
|
|
|
|
|
import hashlib
|
|
|
|
|
import json
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
import logging
|
2026-08-09 20:59:14 +02:00
|
|
|
import os
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
import signal
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
import sys
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
import threading
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
from datetime import datetime, timezone
|
2026-08-09 20:59:14 +02:00
|
|
|
from http import HTTPStatus
|
|
|
|
|
from typing import Any
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
from urllib.parse import parse_qs
|
2026-08-09 20:59:14 +02:00
|
|
|
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
from audit_core.interface import (
|
|
|
|
|
AuditEvent,
|
|
|
|
|
BackendUnavailableError,
|
|
|
|
|
EventConflictError,
|
|
|
|
|
EventValidationError,
|
|
|
|
|
IdempotentAuditBackend,
|
|
|
|
|
)
|
Redact secret-shaped fields by default, countable per field path
AUDIT-WP-0004-T04, closing the workplan.
Decision (Bernd): default to redaction, allow rejection per sender. Losing an
audit record over one field is worse than storing it masked, but a
higher-assurance channel must be able to refuse rather than mask. secret_policy
is set per sender identity in AUDIT_CORE_SENDERS and defaults to redact.
Detection now covers the whole payload at any depth, including lists, rather
than only the top level of data. Under redaction the value is masked and the
key is preserved: dropping the key would hide that the sender transmitted the
field at all, which is exactly what an operator needs in order to stop it. The
stored record carries details.redaction with policy and affected paths, so a
reader never has to infer whether what they see is what was sent.
Idempotency is unaffected - the payload hash is taken over the original request
body, so redaction is deterministic and a resubmission still reconciles as a
duplicate.
Both outcomes are counted durably by sender, source, action and field path,
exposed at GET /v1/secret-findings. Per-path aggregation is the point: the
actionable unit is "stop emitting data.auth.token on membership.added", not
"there were 47 redactions". Counters survive restart because the fix they drive
lives in another service.
Contract doc updated to match. Tests 46 -> 50. WP-0004 is finished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:02:22 +02:00
|
|
|
from audit_core.redaction import (
|
|
|
|
|
POLICY_REDACT,
|
|
|
|
|
SecretFieldRejection,
|
|
|
|
|
apply_policy,
|
|
|
|
|
finding_from_path,
|
|
|
|
|
)
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
from audit_core.senders import SenderIdentity, SenderRegistry, development_registry
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
from audit_core.sqlite_backend import SQLiteAuditBackend
|
2026-08-09 20:59:14 +02:00
|
|
|
|
|
|
|
|
MAX_BODY_BYTES = 256 * 1024
|
|
|
|
|
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
log = logging.getLogger("audit_core.ingestion")
|
2026-08-09 20:59:14 +02:00
|
|
|
|
|
|
|
|
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
class Counters:
|
|
|
|
|
"""In-process request counters (AUDIT-WP-0005-T03).
|
|
|
|
|
|
|
|
|
|
Exposed as JSON at ``/v1/stats`` rather than in Prometheus exposition
|
|
|
|
|
format, because railiance01 currently runs no Prometheus, ServiceMonitor
|
|
|
|
|
CRD, or any other scrape target. Building an exposition endpoint for a
|
|
|
|
|
scrape path that does not exist would be guessing; this is usable by an
|
|
|
|
|
operator with curl today and is a small step from a /metrics endpoint when
|
|
|
|
|
a metrics stack lands.
|
|
|
|
|
|
|
|
|
|
These reset on restart, which is correct for rate signals. The counters
|
|
|
|
|
that must survive a restart — secret-shaped field findings — are persisted
|
|
|
|
|
by the backend instead.
|
|
|
|
|
"""
|
|
|
|
|
|
|
|
|
|
FIELDS = ("accepted", "duplicate", "conflict", "rejected", "unauthorized",
|
|
|
|
|
"forbidden", "unavailable", "error")
|
|
|
|
|
|
|
|
|
|
def __init__(self) -> None:
|
|
|
|
|
self._lock = threading.Lock()
|
|
|
|
|
self._counts = {name: 0 for name in self.FIELDS}
|
|
|
|
|
self._started = datetime.now(timezone.utc).replace(microsecond=0)
|
|
|
|
|
|
|
|
|
|
def hit(self, name: str) -> None:
|
|
|
|
|
with self._lock:
|
|
|
|
|
if name in self._counts:
|
|
|
|
|
self._counts[name] += 1
|
|
|
|
|
|
|
|
|
|
def snapshot(self) -> dict[str, Any]:
|
|
|
|
|
with self._lock:
|
|
|
|
|
counts = dict(self._counts)
|
|
|
|
|
return {"since": self._started.isoformat(), "counts": counts}
|
|
|
|
|
|
|
|
|
|
|
2026-08-09 20:59:14 +02:00
|
|
|
class IngestionApplication:
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
"""WSGI application accepting user-engine outbox events.
|
|
|
|
|
|
|
|
|
|
Writes through the audit backend contract rather than to storage directly,
|
|
|
|
|
so a 202 means a backend with a declared retention policy acknowledged the
|
|
|
|
|
event.
|
|
|
|
|
"""
|
|
|
|
|
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
def __init__(
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
self,
|
|
|
|
|
backend: IdempotentAuditBackend,
|
|
|
|
|
senders: SenderRegistry | str,
|
|
|
|
|
require_custody_class: str | None = None,
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
) -> None:
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
policy = backend.retention_policy
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
if require_custody_class and policy.custody_class != require_custody_class:
|
|
|
|
|
# Production sets this. Without it, losing AUDIT_CORE_DATABASE_URL
|
|
|
|
|
# silently downgrades custody to the development store instead of
|
|
|
|
|
# failing to start.
|
|
|
|
|
raise ValueError(
|
|
|
|
|
f"backend custody_class={policy.custody_class!r} does not meet the "
|
|
|
|
|
f"required {require_custody_class!r}; refusing to start"
|
|
|
|
|
)
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
if not policy.durable:
|
|
|
|
|
# The mock file backend declares durable=False. Refusing it here is
|
|
|
|
|
# what stops a development sink from silently becoming the
|
|
|
|
|
# production one (AUDIT-WP-0004-T01).
|
|
|
|
|
raise ValueError(
|
|
|
|
|
f"backend custody_class={policy.custody_class!r} is not durable; "
|
|
|
|
|
"refusing to accept audit events against it"
|
|
|
|
|
)
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
if isinstance(senders, str):
|
|
|
|
|
if not senders:
|
|
|
|
|
raise ValueError("bearer token is required")
|
|
|
|
|
senders = development_registry(senders)
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
self.backend = backend
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
self.senders = senders
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
self.counters = Counters()
|
2026-08-09 20:59:14 +02:00
|
|
|
|
|
|
|
|
def __call__(self, environ, start_response):
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
try:
|
|
|
|
|
return self._handle(environ, start_response)
|
|
|
|
|
except Exception:
|
|
|
|
|
# Nothing may escape: an unhandled exception here means
|
|
|
|
|
# start_response is never called and the sender sees a dropped
|
|
|
|
|
# connection it cannot classify.
|
|
|
|
|
log.exception("unhandled error in ingestion request")
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
self.counters.hit("error")
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
return self._json(
|
|
|
|
|
start_response, HTTPStatus.INTERNAL_SERVER_ERROR, {"error": "internal_error"}
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
def _handle(self, environ, start_response):
|
2026-08-09 20:59:14 +02:00
|
|
|
path = environ.get("PATH_INFO", "")
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
if path == "/healthz":
|
2026-08-09 20:59:14 +02:00
|
|
|
return self._json(start_response, HTTPStatus.OK, {"status": "ok"})
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
if path == "/readyz":
|
|
|
|
|
return self._readiness(start_response)
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
method = environ.get("REQUEST_METHOD")
|
|
|
|
|
identity = self.senders.authenticate(environ.get("HTTP_AUTHORIZATION"))
|
|
|
|
|
if identity is None:
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
self.counters.hit("unauthorized")
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
return self._json(
|
|
|
|
|
start_response, HTTPStatus.UNAUTHORIZED, {"error": "unauthorized"}
|
|
|
|
|
)
|
|
|
|
|
|
Redact secret-shaped fields by default, countable per field path
AUDIT-WP-0004-T04, closing the workplan.
Decision (Bernd): default to redaction, allow rejection per sender. Losing an
audit record over one field is worse than storing it masked, but a
higher-assurance channel must be able to refuse rather than mask. secret_policy
is set per sender identity in AUDIT_CORE_SENDERS and defaults to redact.
Detection now covers the whole payload at any depth, including lists, rather
than only the top level of data. Under redaction the value is masked and the
key is preserved: dropping the key would hide that the sender transmitted the
field at all, which is exactly what an operator needs in order to stop it. The
stored record carries details.redaction with policy and affected paths, so a
reader never has to infer whether what they see is what was sent.
Idempotency is unaffected - the payload hash is taken over the original request
body, so redaction is deterministic and a resubmission still reconciles as a
duplicate.
Both outcomes are counted durably by sender, source, action and field path,
exposed at GET /v1/secret-findings. Per-path aggregation is the point: the
actionable unit is "stop emitting data.auth.token on membership.added", not
"there were 47 redactions". Counters survive restart because the fix they drive
lives in another service.
Contract doc updated to match. Tests 46 -> 50. WP-0004 is finished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:02:22 +02:00
|
|
|
if method == "GET" and (
|
|
|
|
|
path.startswith("/v1/events")
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
or path in ("/v1/dead-letters", "/v1/secret-findings", "/v1/stats")
|
Redact secret-shaped fields by default, countable per field path
AUDIT-WP-0004-T04, closing the workplan.
Decision (Bernd): default to redaction, allow rejection per sender. Losing an
audit record over one field is worse than storing it masked, but a
higher-assurance channel must be able to refuse rather than mask. secret_policy
is set per sender identity in AUDIT_CORE_SENDERS and defaults to redact.
Detection now covers the whole payload at any depth, including lists, rather
than only the top level of data. Under redaction the value is masked and the
key is preserved: dropping the key would hide that the sender transmitted the
field at all, which is exactly what an operator needs in order to stop it. The
stored record carries details.redaction with policy and affected paths, so a
reader never has to infer whether what they see is what was sent.
Idempotency is unaffected - the payload hash is taken over the original request
body, so redaction is deterministic and a resubmission still reconciles as a
duplicate.
Both outcomes are counted durably by sender, source, action and field path,
exposed at GET /v1/secret-findings. Per-path aggregation is the point: the
actionable unit is "stop emitting data.auth.token on membership.added", not
"there were 47 redactions". Counters survive restart because the fix they drive
lives in another service.
Contract doc updated to match. Tests 46 -> 50. WP-0004 is finished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:02:22 +02:00
|
|
|
):
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
return self._read(start_response, environ, path, identity)
|
|
|
|
|
|
|
|
|
|
if path != "/v1/events" or method != "POST":
|
|
|
|
|
return self._json(start_response, HTTPStatus.NOT_FOUND, {"error": "not_found"})
|
|
|
|
|
|
|
|
|
|
if not identity.may_write:
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
self.counters.hit("forbidden")
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
return self._json(start_response, HTTPStatus.FORBIDDEN, {"error": "write_forbidden"})
|
|
|
|
|
|
|
|
|
|
raw = b""
|
Redact secret-shaped fields by default, countable per field path
AUDIT-WP-0004-T04, closing the workplan.
Decision (Bernd): default to redaction, allow rejection per sender. Losing an
audit record over one field is worse than storing it masked, but a
higher-assurance channel must be able to refuse rather than mask. secret_policy
is set per sender identity in AUDIT_CORE_SENDERS and defaults to redact.
Detection now covers the whole payload at any depth, including lists, rather
than only the top level of data. Under redaction the value is masked and the
key is preserved: dropping the key would hide that the sender transmitted the
field at all, which is exactly what an operator needs in order to stop it. The
stored record carries details.redaction with policy and affected paths, so a
reader never has to infer whether what they see is what was sent.
Idempotency is unaffected - the payload hash is taken over the original request
body, so redaction is deterministic and a resubmission still reconciles as a
duplicate.
Both outcomes are counted durably by sender, source, action and field path,
exposed at GET /v1/secret-findings. Per-path aggregation is the point: the
actionable unit is "stop emitting data.auth.token on membership.added", not
"there were 47 redactions". Counters survive restart because the fix they drive
lives in another service.
Contract doc updated to match. Tests 46 -> 50. WP-0004 is finished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:02:22 +02:00
|
|
|
payload: Any = {}
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
try:
|
|
|
|
|
raw = self._read_body(environ)
|
Redact secret-shaped fields by default, countable per field path
AUDIT-WP-0004-T04, closing the workplan.
Decision (Bernd): default to redaction, allow rejection per sender. Losing an
audit record over one field is worse than storing it masked, but a
higher-assurance channel must be able to refuse rather than mask. secret_policy
is set per sender identity in AUDIT_CORE_SENDERS and defaults to redact.
Detection now covers the whole payload at any depth, including lists, rather
than only the top level of data. Under redaction the value is masked and the
key is preserved: dropping the key would hide that the sender transmitted the
field at all, which is exactly what an operator needs in order to stop it. The
stored record carries details.redaction with policy and affected paths, so a
reader never has to infer whether what they see is what was sent.
Idempotency is unaffected - the payload hash is taken over the original request
body, so redaction is deterministic and a resubmission still reconciles as a
duplicate.
Both outcomes are counted durably by sender, source, action and field path,
exposed at GET /v1/secret-findings. Per-path aggregation is the point: the
actionable unit is "stop emitting data.auth.token on membership.added", not
"there were 47 redactions". Counters survive restart because the fix they drive
lives in another service.
Contract doc updated to match. Tests 46 -> 50. WP-0004 is finished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:02:22 +02:00
|
|
|
payload = json.loads(raw)
|
|
|
|
|
event = normalize(payload, environ.get("HTTP_IDEMPOTENCY_KEY"), identity)
|
|
|
|
|
except SecretFieldRejection as exc:
|
|
|
|
|
self._count_secrets(payload, identity, "rejected", exc.findings)
|
|
|
|
|
self._dead_letter(raw, str(exc), identity)
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
self.counters.hit("rejected")
|
Redact secret-shaped fields by default, countable per field path
AUDIT-WP-0004-T04, closing the workplan.
Decision (Bernd): default to redaction, allow rejection per sender. Losing an
audit record over one field is worse than storing it masked, but a
higher-assurance channel must be able to refuse rather than mask. secret_policy
is set per sender identity in AUDIT_CORE_SENDERS and defaults to redact.
Detection now covers the whole payload at any depth, including lists, rather
than only the top level of data. Under redaction the value is masked and the
key is preserved: dropping the key would hide that the sender transmitted the
field at all, which is exactly what an operator needs in order to stop it. The
stored record carries details.redaction with policy and affected paths, so a
reader never has to infer whether what they see is what was sent.
Idempotency is unaffected - the payload hash is taken over the original request
body, so redaction is deterministic and a resubmission still reconciles as a
duplicate.
Both outcomes are counted durably by sender, source, action and field path,
exposed at GET /v1/secret-findings. Per-path aggregation is the point: the
actionable unit is "stop emitting data.auth.token on membership.added", not
"there were 47 redactions". Counters survive restart because the fix they drive
lives in another service.
Contract doc updated to match. Tests 46 -> 50. WP-0004 is finished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:02:22 +02:00
|
|
|
return self._json(start_response, HTTPStatus.BAD_REQUEST, {"error": str(exc)})
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
except (ValueError, TypeError, KeyError, json.JSONDecodeError) as exc:
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
self._dead_letter(raw, str(exc), identity)
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
self.counters.hit("rejected")
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
return self._json(start_response, HTTPStatus.BAD_REQUEST, {"error": str(exc)})
|
|
|
|
|
|
Redact secret-shaped fields by default, countable per field path
AUDIT-WP-0004-T04, closing the workplan.
Decision (Bernd): default to redaction, allow rejection per sender. Losing an
audit record over one field is worse than storing it masked, but a
higher-assurance channel must be able to refuse rather than mask. secret_policy
is set per sender identity in AUDIT_CORE_SENDERS and defaults to redact.
Detection now covers the whole payload at any depth, including lists, rather
than only the top level of data. Under redaction the value is masked and the
key is preserved: dropping the key would hide that the sender transmitted the
field at all, which is exactly what an operator needs in order to stop it. The
stored record carries details.redaction with policy and affected paths, so a
reader never has to infer whether what they see is what was sent.
Idempotency is unaffected - the payload hash is taken over the original request
body, so redaction is deterministic and a resubmission still reconciles as a
duplicate.
Both outcomes are counted durably by sender, source, action and field path,
exposed at GET /v1/secret-findings. Per-path aggregation is the point: the
actionable unit is "stop emitting data.auth.token on membership.added", not
"there were 47 redactions". Counters survive restart because the fix they drive
lives in another service.
Contract doc updated to match. Tests 46 -> 50. WP-0004 is finished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:02:22 +02:00
|
|
|
redaction = event.details.get("redaction")
|
|
|
|
|
if redaction:
|
|
|
|
|
self._count_secrets(
|
|
|
|
|
payload, identity, "redacted",
|
|
|
|
|
[finding_from_path(p) for p in redaction["paths"]],
|
|
|
|
|
)
|
|
|
|
|
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
try:
|
|
|
|
|
result = self.backend.accept(event, hashlib.sha256(raw).hexdigest())
|
|
|
|
|
except EventConflictError as exc:
|
|
|
|
|
log.warning("event conflict: %s", exc)
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
self.counters.hit("conflict")
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
return self._json(start_response, HTTPStatus.CONFLICT, {"error": "event_id_conflict"})
|
|
|
|
|
except EventValidationError as exc:
|
|
|
|
|
return self._json(start_response, HTTPStatus.BAD_REQUEST, {"error": str(exc)})
|
|
|
|
|
except BackendUnavailableError as exc:
|
|
|
|
|
log.error("backend unavailable: %s", exc)
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
self.counters.hit("unavailable")
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
return self._json(
|
|
|
|
|
start_response, HTTPStatus.SERVICE_UNAVAILABLE, {"error": "backend_unavailable"}
|
|
|
|
|
)
|
|
|
|
|
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
self.counters.hit("duplicate" if result.duplicate else "accepted")
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
return self._json(
|
|
|
|
|
start_response,
|
|
|
|
|
HTTPStatus.OK if result.duplicate else HTTPStatus.ACCEPTED,
|
|
|
|
|
{
|
|
|
|
|
"status": "duplicate" if result.duplicate else "accepted",
|
|
|
|
|
"reference": result.reference,
|
|
|
|
|
},
|
|
|
|
|
)
|
|
|
|
|
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
def _read(self, start_response, environ, path: str, identity):
|
|
|
|
|
"""Operator read surface (AUDIT-WP-0004-T05).
|
|
|
|
|
|
|
|
|
|
Read is a distinct privilege from write: a sender credential must not
|
|
|
|
|
be able to read the audit trail back.
|
|
|
|
|
"""
|
|
|
|
|
if not identity.may_read:
|
|
|
|
|
return self._json(start_response, HTTPStatus.FORBIDDEN, {"error": "read_forbidden"})
|
|
|
|
|
|
|
|
|
|
query = parse_qs(environ.get("QUERY_STRING", ""))
|
|
|
|
|
try:
|
|
|
|
|
if path == "/v1/dead-letters":
|
|
|
|
|
return self._json(
|
|
|
|
|
start_response, HTTPStatus.OK,
|
|
|
|
|
{"dead_letters": self.backend.dead_letters(_limit(query))},
|
|
|
|
|
)
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
if path == "/v1/stats":
|
|
|
|
|
return self._json(start_response, HTTPStatus.OK, self.counters.snapshot())
|
Redact secret-shaped fields by default, countable per field path
AUDIT-WP-0004-T04, closing the workplan.
Decision (Bernd): default to redaction, allow rejection per sender. Losing an
audit record over one field is worse than storing it masked, but a
higher-assurance channel must be able to refuse rather than mask. secret_policy
is set per sender identity in AUDIT_CORE_SENDERS and defaults to redact.
Detection now covers the whole payload at any depth, including lists, rather
than only the top level of data. Under redaction the value is masked and the
key is preserved: dropping the key would hide that the sender transmitted the
field at all, which is exactly what an operator needs in order to stop it. The
stored record carries details.redaction with policy and affected paths, so a
reader never has to infer whether what they see is what was sent.
Idempotency is unaffected - the payload hash is taken over the original request
body, so redaction is deterministic and a resubmission still reconciles as a
duplicate.
Both outcomes are counted durably by sender, source, action and field path,
exposed at GET /v1/secret-findings. Per-path aggregation is the point: the
actionable unit is "stop emitting data.auth.token on membership.added", not
"there were 47 redactions". Counters survive restart because the fix they drive
lives in another service.
Contract doc updated to match. Tests 46 -> 50. WP-0004 is finished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:02:22 +02:00
|
|
|
if path == "/v1/secret-findings":
|
|
|
|
|
return self._json(
|
|
|
|
|
start_response, HTTPStatus.OK,
|
|
|
|
|
{"secret_findings": self.backend.secret_findings(_limit(query))},
|
|
|
|
|
)
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
if path == "/v1/events":
|
|
|
|
|
correlation = (query.get("correlation_id") or [""])[0]
|
|
|
|
|
if not correlation:
|
|
|
|
|
return self._json(
|
|
|
|
|
start_response, HTTPStatus.BAD_REQUEST,
|
|
|
|
|
{"error": "correlation_id_required"},
|
|
|
|
|
)
|
|
|
|
|
return self._json(
|
|
|
|
|
start_response, HTTPStatus.OK,
|
|
|
|
|
{"events": self.backend.by_correlation(correlation, _limit(query))},
|
|
|
|
|
)
|
|
|
|
|
event_id = path[len("/v1/events/"):]
|
|
|
|
|
record = self.backend.get(event_id) if event_id else None
|
|
|
|
|
if record is None:
|
|
|
|
|
return self._json(start_response, HTTPStatus.NOT_FOUND, {"error": "not_found"})
|
|
|
|
|
return self._json(start_response, HTTPStatus.OK, record)
|
|
|
|
|
except BackendUnavailableError as exc:
|
|
|
|
|
log.error("read failed: %s", exc)
|
|
|
|
|
return self._json(
|
|
|
|
|
start_response, HTTPStatus.SERVICE_UNAVAILABLE, {"error": "backend_unavailable"}
|
|
|
|
|
)
|
|
|
|
|
|
Redact secret-shaped fields by default, countable per field path
AUDIT-WP-0004-T04, closing the workplan.
Decision (Bernd): default to redaction, allow rejection per sender. Losing an
audit record over one field is worse than storing it masked, but a
higher-assurance channel must be able to refuse rather than mask. secret_policy
is set per sender identity in AUDIT_CORE_SENDERS and defaults to redact.
Detection now covers the whole payload at any depth, including lists, rather
than only the top level of data. Under redaction the value is masked and the
key is preserved: dropping the key would hide that the sender transmitted the
field at all, which is exactly what an operator needs in order to stop it. The
stored record carries details.redaction with policy and affected paths, so a
reader never has to infer whether what they see is what was sent.
Idempotency is unaffected - the payload hash is taken over the original request
body, so redaction is deterministic and a resubmission still reconciles as a
duplicate.
Both outcomes are counted durably by sender, source, action and field path,
exposed at GET /v1/secret-findings. Per-path aggregation is the point: the
actionable unit is "stop emitting data.auth.token on membership.added", not
"there were 47 redactions". Counters survive restart because the fix they drive
lives in another service.
Contract doc updated to match. Tests 46 -> 50. WP-0004 is finished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:02:22 +02:00
|
|
|
def _count_secrets(self, payload: Any, identity, outcome: str, findings) -> None:
|
|
|
|
|
"""Count secret-shaped fields by path, sender, source and action.
|
|
|
|
|
|
|
|
|
|
Counted so the sending service can be fixed. Aggregation is by field
|
|
|
|
|
path rather than by event, because the actionable unit is "stop
|
|
|
|
|
emitting ``data.auth.token`` on ``membership.added``", not "there were
|
|
|
|
|
47 redactions".
|
|
|
|
|
"""
|
|
|
|
|
counter = getattr(self.backend, "count_secret_findings", None)
|
|
|
|
|
if not callable(counter) or not findings:
|
|
|
|
|
return
|
|
|
|
|
source = str(payload.get("source") or "") if isinstance(payload, dict) else ""
|
|
|
|
|
action = str(payload.get("type") or "") if isinstance(payload, dict) else ""
|
|
|
|
|
try:
|
|
|
|
|
counter(
|
|
|
|
|
sender=identity.name, source=source, action=action,
|
|
|
|
|
outcome=outcome, findings=findings,
|
|
|
|
|
)
|
|
|
|
|
except BackendUnavailableError as exc:
|
|
|
|
|
# A missed counter must never change the event's outcome.
|
|
|
|
|
log.error("could not count secret findings: %s", exc)
|
|
|
|
|
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
def _dead_letter(self, raw: bytes, reason: str, identity) -> None:
|
|
|
|
|
"""Record a rejection so an operator can see what the sender dropped."""
|
|
|
|
|
recorder = getattr(self.backend, "record_rejection", None)
|
|
|
|
|
if not callable(recorder):
|
|
|
|
|
return
|
|
|
|
|
event_id = None
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
try:
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
parsed = json.loads(raw)
|
|
|
|
|
if isinstance(parsed, dict):
|
|
|
|
|
event_id = str(parsed.get("id") or "") or None
|
|
|
|
|
except (ValueError, TypeError):
|
|
|
|
|
pass
|
|
|
|
|
try:
|
|
|
|
|
recorder(
|
|
|
|
|
event_id=event_id,
|
|
|
|
|
reason=reason,
|
|
|
|
|
payload_hash=hashlib.sha256(raw).hexdigest(),
|
|
|
|
|
sender=identity.name,
|
|
|
|
|
payload=raw.decode("utf-8", "replace"),
|
|
|
|
|
)
|
|
|
|
|
except BackendUnavailableError as exc:
|
|
|
|
|
# A rejection we could not record is worth a log line, but it must
|
|
|
|
|
# not turn a 400 into a 503 — the event is still rejected.
|
|
|
|
|
log.error("could not record dead letter: %s", exc)
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
|
|
|
|
|
def _read_body(self, environ) -> bytes:
|
2026-08-09 20:59:14 +02:00
|
|
|
try:
|
|
|
|
|
length = int(environ.get("CONTENT_LENGTH") or 0)
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
except (TypeError, ValueError):
|
|
|
|
|
raise ValueError("invalid_content_length") from None
|
|
|
|
|
if length <= 0:
|
|
|
|
|
raise ValueError("empty_body")
|
|
|
|
|
if length > MAX_BODY_BYTES:
|
|
|
|
|
raise ValueError("payload_too_large")
|
|
|
|
|
raw = environ["wsgi.input"].read(length)
|
|
|
|
|
if len(raw) != length:
|
|
|
|
|
raise ValueError("truncated_body")
|
|
|
|
|
return raw
|
|
|
|
|
|
|
|
|
|
def _readiness(self, start_response):
|
|
|
|
|
try:
|
|
|
|
|
health = getattr(self.backend, "health", None)
|
|
|
|
|
if callable(health):
|
|
|
|
|
health()
|
|
|
|
|
except BackendUnavailableError as exc:
|
|
|
|
|
log.error("readiness failed: %s", exc)
|
|
|
|
|
return self._json(
|
|
|
|
|
start_response, HTTPStatus.SERVICE_UNAVAILABLE, {"status": "unavailable"}
|
2026-08-09 20:59:14 +02:00
|
|
|
)
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
policy = self.backend.retention_policy
|
|
|
|
|
return self._json(
|
|
|
|
|
start_response,
|
|
|
|
|
HTTPStatus.OK,
|
|
|
|
|
{"status": "ok", "custody_class": policy.custody_class, "durable": policy.durable},
|
|
|
|
|
)
|
2026-08-09 20:59:14 +02:00
|
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
|
def _json(start_response, status: HTTPStatus, payload: dict[str, Any]):
|
|
|
|
|
body = json.dumps(payload).encode()
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
start_response(
|
|
|
|
|
f"{status.value} {status.phrase}",
|
|
|
|
|
[("Content-Type", "application/json"), ("Content-Length", str(len(body)))],
|
|
|
|
|
)
|
2026-08-09 20:59:14 +02:00
|
|
|
return [body]
|
|
|
|
|
|
|
|
|
|
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
def normalize(
|
|
|
|
|
payload: dict[str, Any],
|
|
|
|
|
idempotency_key: str | None,
|
|
|
|
|
identity: SenderIdentity | None = None,
|
|
|
|
|
) -> AuditEvent:
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
required = (
|
|
|
|
|
"id", "type", "source", "subject", "tenant", "correlation_id", "occurred_at", "data",
|
|
|
|
|
)
|
2026-08-09 20:59:14 +02:00
|
|
|
if not isinstance(payload, dict) or any(not payload.get(key) for key in required):
|
|
|
|
|
raise ValueError("invalid_event")
|
|
|
|
|
if idempotency_key != payload["id"]:
|
|
|
|
|
raise ValueError("idempotency_key_mismatch")
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
source = str(payload["source"])
|
|
|
|
|
tenant = str(payload["tenant"])
|
|
|
|
|
# The claimed source and tenant are checked against what this credential is
|
|
|
|
|
# permitted to assert, not against a literal (AUDIT-WP-0004-T03).
|
|
|
|
|
if identity is not None:
|
|
|
|
|
if not identity.permits_source(source):
|
|
|
|
|
raise ValueError("source_not_allowed")
|
|
|
|
|
if not identity.permits_tenant(tenant):
|
|
|
|
|
raise ValueError("tenant_not_allowed")
|
|
|
|
|
elif source != "user-engine":
|
2026-08-09 20:59:14 +02:00
|
|
|
raise ValueError("source_not_allowed")
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
observed_at = _normalize_timestamp(payload["occurred_at"])
|
Redact secret-shaped fields by default, countable per field path
AUDIT-WP-0004-T04, closing the workplan.
Decision (Bernd): default to redaction, allow rejection per sender. Losing an
audit record over one field is worse than storing it masked, but a
higher-assurance channel must be able to refuse rather than mask. secret_policy
is set per sender identity in AUDIT_CORE_SENDERS and defaults to redact.
Detection now covers the whole payload at any depth, including lists, rather
than only the top level of data. Under redaction the value is masked and the
key is preserved: dropping the key would hide that the sender transmitted the
field at all, which is exactly what an operator needs in order to stop it. The
stored record carries details.redaction with policy and affected paths, so a
reader never has to infer whether what they see is what was sent.
Idempotency is unaffected - the payload hash is taken over the original request
body, so redaction is deterministic and a resubmission still reconciles as a
duplicate.
Both outcomes are counted durably by sender, source, action and field path,
exposed at GET /v1/secret-findings. Per-path aggregation is the point: the
actionable unit is "stop emitting data.auth.token on membership.added", not
"there were 47 redactions". Counters survive restart because the fix they drive
lives in another service.
Contract doc updated to match. Tests 46 -> 50. WP-0004 is finished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:02:22 +02:00
|
|
|
policy = identity.secret_policy if identity is not None else POLICY_REDACT
|
|
|
|
|
# Raises SecretFieldRejection under the reject policy; that exception
|
|
|
|
|
# carries the findings so the caller can count them.
|
|
|
|
|
data, findings = apply_policy(payload, policy)
|
|
|
|
|
details: dict[str, Any] = {
|
|
|
|
|
"correlation_id": str(payload["correlation_id"]),
|
|
|
|
|
"data": data,
|
|
|
|
|
}
|
|
|
|
|
if findings:
|
|
|
|
|
# The stored record differs from what the sender transmitted. Say so in
|
|
|
|
|
# the record itself rather than leaving it to be inferred.
|
|
|
|
|
details["redaction"] = {
|
|
|
|
|
"policy": policy,
|
|
|
|
|
"paths": [f.path for f in findings],
|
|
|
|
|
}
|
2026-08-09 20:59:14 +02:00
|
|
|
return AuditEvent(
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
event_id=str(payload["id"]),
|
|
|
|
|
observed_at=observed_at,
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
tenant=tenant,
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
scope="tenant",
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
source=source,
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
action=str(payload["type"]),
|
|
|
|
|
resource=str(payload["subject"]),
|
|
|
|
|
outcome="recorded",
|
|
|
|
|
actor=None,
|
Redact secret-shaped fields by default, countable per field path
AUDIT-WP-0004-T04, closing the workplan.
Decision (Bernd): default to redaction, allow rejection per sender. Losing an
audit record over one field is worse than storing it masked, but a
higher-assurance channel must be able to refuse rather than mask. secret_policy
is set per sender identity in AUDIT_CORE_SENDERS and defaults to redact.
Detection now covers the whole payload at any depth, including lists, rather
than only the top level of data. Under redaction the value is masked and the
key is preserved: dropping the key would hide that the sender transmitted the
field at all, which is exactly what an operator needs in order to stop it. The
stored record carries details.redaction with policy and affected paths, so a
reader never has to infer whether what they see is what was sent.
Idempotency is unaffected - the payload hash is taken over the original request
body, so redaction is deterministic and a resubmission still reconciles as a
duplicate.
Both outcomes are counted durably by sender, source, action and field path,
exposed at GET /v1/secret-findings. Per-path aggregation is the point: the
actionable unit is "stop emitting data.auth.token on membership.added", not
"there were 47 redactions". Counters survive restart because the fix they drive
lives in another service.
Contract doc updated to match. Tests 46 -> 50. WP-0004 is finished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:02:22 +02:00
|
|
|
details=details,
|
2026-08-09 20:59:14 +02:00
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
def _normalize_timestamp(value: Any) -> str:
|
|
|
|
|
"""Parse an event timestamp, requiring an explicit offset.
|
|
|
|
|
|
|
|
|
|
A naive timestamp is ambiguous by up to a day, which is not good enough for
|
|
|
|
|
an audit trail — the sender must say which offset it meant.
|
|
|
|
|
"""
|
|
|
|
|
try:
|
|
|
|
|
parsed = datetime.fromisoformat(str(value).replace("Z", "+00:00"))
|
|
|
|
|
except (TypeError, ValueError):
|
|
|
|
|
raise ValueError("invalid_timestamp") from None
|
|
|
|
|
if parsed.tzinfo is None:
|
|
|
|
|
raise ValueError("timestamp_missing_timezone")
|
|
|
|
|
return parsed.astimezone(timezone.utc).isoformat()
|
|
|
|
|
|
|
|
|
|
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
def _limit(query: dict[str, list[str]], default: int = 100, ceiling: int = 1000) -> int:
|
|
|
|
|
try:
|
|
|
|
|
return max(1, min(int((query.get("limit") or [default])[0]), ceiling))
|
|
|
|
|
except (TypeError, ValueError):
|
|
|
|
|
return default
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def serve(app, host: str, port: int, threads: int, timeout: int) -> None:
|
|
|
|
|
"""Serve ``app``, preferring a production WSGI server (T06).
|
|
|
|
|
|
|
|
|
|
waitress is the intended production server and is installed in the image
|
|
|
|
|
via the ``serve`` extra. The fallback is a threaded wsgiref server with a
|
|
|
|
|
socket timeout — bounded rather than good, and loud about which one is in
|
|
|
|
|
use so a deployment cannot quietly end up on the fallback.
|
|
|
|
|
"""
|
|
|
|
|
try:
|
|
|
|
|
from waitress import serve as waitress_serve
|
|
|
|
|
except ImportError:
|
|
|
|
|
log.warning(
|
|
|
|
|
"waitress not installed — falling back to a threaded wsgiref server. "
|
|
|
|
|
"Install the 'serve' extra for production (AUDIT-WP-0004-T06)."
|
|
|
|
|
)
|
|
|
|
|
_serve_fallback(app, host, port, timeout)
|
|
|
|
|
return
|
|
|
|
|
|
|
|
|
|
log.info("serving on waitress host=%s port=%s threads=%s", host, port, threads)
|
|
|
|
|
waitress_serve(
|
|
|
|
|
app, host=host, port=port, threads=threads,
|
|
|
|
|
channel_timeout=timeout, ident="audit-core",
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def _serve_fallback(app, host: str, port: int, timeout: int) -> None:
|
|
|
|
|
from socketserver import ThreadingMixIn
|
|
|
|
|
from wsgiref.simple_server import WSGIServer, make_server
|
|
|
|
|
|
|
|
|
|
class ThreadedWSGIServer(ThreadingMixIn, WSGIServer):
|
|
|
|
|
daemon_threads = True
|
|
|
|
|
# Without this a slow or idle client holds a worker indefinitely; the
|
|
|
|
|
# original single-threaded server let one such client block every
|
|
|
|
|
# sender.
|
|
|
|
|
timeout = timeout
|
|
|
|
|
|
|
|
|
|
with make_server(host, port, app, server_class=ThreadedWSGIServer) as server:
|
|
|
|
|
server.socket.settimeout(timeout)
|
|
|
|
|
|
|
|
|
|
def shutdown(signum, _frame):
|
|
|
|
|
log.info("received signal %s, shutting down", signum)
|
|
|
|
|
server.shutdown()
|
|
|
|
|
|
|
|
|
|
for sig in (signal.SIGTERM, signal.SIGINT):
|
|
|
|
|
signal.signal(sig, shutdown)
|
|
|
|
|
log.info("serving on threaded wsgiref host=%s port=%s", host, port)
|
|
|
|
|
server.serve_forever()
|
|
|
|
|
|
|
|
|
|
|
2026-08-10 17:09:46 +02:00
|
|
|
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)
|
|
|
|
|
|
|
|
|
|
|
2026-08-09 20:59:14 +02:00
|
|
|
def main() -> None:
|
Route ingestion through the backend contract; fix error semantics
AUDIT-WP-0004 T01, T02, T07.
T01 - ingestion wrote to SQLite directly and never called the AuditBackend
contract, so a 202 meant a row existed rather than that a backend with a
declared retention policy had accepted the event. Adds IdempotentAuditBackend
to the contract: duplicate detection lives inside the backend so custody and
idempotency state share a transaction and cannot diverge. SQLiteAuditBackend
implements it with WAL, synchronous=FULL and a busy timeout. Ingestion now
refuses any backend declaring durable=False, so the development file backend
cannot silently become the production sink.
The atomicity claim was tested rather than asserted, and the first attempt
failed: with a single shared connection, 16 racing submissions of one event
told two callers they were first. Storage was correct but the response was
not. Fixed with per-thread connections and BEGIN IMMEDIATE around the
insert/read pair, and locked in by a test.
T02 - storage errors previously escaped the handler with start_response never
called, and the auth check sat outside the try block so a non-ASCII
Authorization header crashed the request. Adds a catch-all, maps conflict to
409, backend unavailability to 503 and unexpected faults to 500, and
documents the full response contract with the retry semantics each status
implies, since senders key their behaviour off it.
T07 - ingestion tests 2 -> 23, suite 15 -> 36. accepted_at is now UTC rather
than local time, and naive timestamps are rejected instead of silently
assumed.
Remaining in WP-0004: T03 tenant/source binding, T04 redaction policy,
T05 operator read surface, T06 production serving layer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:30:20 +02:00
|
|
|
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,
|
|
|
|
|
)
|
Add deployment manifests, custody-class guard and request counters
AUDIT-WP-0005-T03 (progress). Manifests validated --dry-run=server
--validate=strict against railiance01; not applied, since deployment is gated
on RAPP-POSTGRES-WP-0002 and T02 credentials. Nothing here mutates the cluster.
Conventions read off the deployed user-engine workload rather than invented:
digest-pinned image from forgejo.coulomb.social, runAsNonRoot with
RuntimeDefault seccomp, no privilege escalation, all capabilities dropped,
readOnlyRootFilesystem, probes on a named http port, same resource envelope.
The namespace carries railiance.io/postgres-client: platform-pg, which is what
platform-pg-consumer-ingress in rapp-postgres admits; without that label the
pod cannot reach the database at all.
NetworkPolicies default-deny both directions, then permit ingress from the
user-engine namespace only, a separately labelled operator read path, and
egress to PostgreSQL in databases plus DNS.
Three decisions worth naming. Liveness is /healthz while readiness is /readyz,
so a database outage drops the pod from the Service rather than restarting it
in a loop. readOnlyRootFilesystem enforces the empty-filesystem property rather
than trusting it, so the SQLite fallback physically cannot accumulate audit
records on ephemeral storage. AUDIT_CORE_REQUIRE_CUSTODY_CLASS=archive makes a
missing database URL a startup failure instead of a silent downgrade to the
development store.
Counters deferred from WP-0004-T06 are exposed as JSON at /v1/stats behind the
read privilege, not as Prometheus exposition format: the cluster runs no
Prometheus, no ServiceMonitor CRD and no other scrape target, so an exposition
endpoint would target a scrape path that does not exist. Usable with curl now
and a small step from /metrics later.
Tests 77 -> 80.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 17:42:43 +02:00
|
|
|
app = IngestionApplication(
|
|
|
|
|
build_backend(),
|
|
|
|
|
SenderRegistry.from_env(),
|
|
|
|
|
require_custody_class=os.environ.get("AUDIT_CORE_REQUIRE_CUSTODY_CLASS") or None,
|
|
|
|
|
)
|
Bind sender identities, add operator read surface, real serving layer
AUDIT-WP-0004 T03, T05, T06.
T03 - the receiver accepted whatever tenant and source a caller sent as long
as it held the one shared token, despite WP-0003 recording tenant isolation as
delivered. audit_core.senders binds each credential to the sources and tenants
it may assert, driven by AUDIT_CORE_SENDERS rather than literals. Identities
hold a list of tokens so rotation publishes the replacement alongside the
incumbent and needs no delivery gap. Read is a separate privilege from write,
so a sender credential cannot read the audit trail back.
T05 - lookup by event id, lookup by correlation id, and a dead-letter view.
Rejections are recorded rather than silently dropped. An event rejected for
carrying secret-shaped material has its payload withheld: storing it would
write that material into the audit store, which is what the rejection exists
to prevent. Reason and payload hash are kept so it stays traceable.
Replay is deliberately not built here. Idempotent replay is a property of the
durable store and building it against SQLite would produce a second
implementation to throw away; it lands with the Postgres backend in
AUDIT-WP-0005-T01.
T06 - serving moves to waitress with configurable threads and channel timeout,
installed in the image via the serve extra. Without it the entrypoint falls
back to a threaded wsgiref server with a socket timeout and graceful shutdown
on SIGTERM, and logs a warning so a deployment cannot quietly land on the
fallback. Metric counters deferred to WP-0005-T03 to be designed against the
real scrape path.
Tests 36 -> 46, covering cross-tenant and cross-source refusal, token
rotation, read/write privilege separation, correlation lookup, and payload
withholding on secret rejection.
Remaining in WP-0004: T04 redaction policy, which needs a decision on whether
a secret-shaped field is a rejection or a redaction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:50:02 +02:00
|
|
|
serve(
|
|
|
|
|
app,
|
|
|
|
|
host=os.environ.get("AUDIT_CORE_HOST", "0.0.0.0"),
|
|
|
|
|
port=int(os.environ.get("AUDIT_CORE_HTTP_PORT", "8080")),
|
|
|
|
|
threads=int(os.environ.get("AUDIT_CORE_THREADS", "8")),
|
|
|
|
|
timeout=int(os.environ.get("AUDIT_CORE_REQUEST_TIMEOUT", "30")),
|
|
|
|
|
)
|
2026-08-10 17:09:46 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
if __name__ == "__main__": # pragma: no cover
|
|
|
|
|
main()
|