문제: 성숙도 리뷰 수정을 다시 읽은 재리뷰(Critical 0, 회귀 0)가 Important 27건·Minor 41건을 찾았다. 수정이 새로 만든 결함이 많았다. @SystemInvoked가 권한 규칙의 탈출구가 됐고, 데모 계정 거부가 정확히 prod일 때만 걸렸고, outbox 순서 번호가 업그레이드 때 뒤로 갔고, GraphQL union 아래 복잡도가 0이었고, 서블릿 오류 응답이 success:true로 나갔고, 호출자 deadline이 breaker를 열었다. 원래 리뷰 색인이 06 영역 발견 5건(06-10~14)을 빠뜨려 그동안 처리되지 않았다. 방법: 영역별 7묶음이 test-driven-development 절차로 실패 테스트를 먼저 쓰고 고쳤다. 결정은 재리뷰 README 4절에 적었다: 순서 번호는 V21이 기록한 전역 floor부터, 데모 계정은 local·test·demo 허용 목록, @SystemInvoked bean은 기동 실패, complete 단계 실패는 IDEMPOTENCY_COMPLETION_UNKNOWN(결과 불명), 캐시 조건부 쓰기는 별도 SCRIPT lane, strict lane 정책은 build cache 입력, 끝나지 않는 재시도 행은 원인 코드·지표·PARKED 절차. 레지스트리가 선언만 하고 아무 코드도 내보내지 않던 redis.capability.* 지표 6개는 빼고 runbook을 실제 지표(redis.lane.*) 기준으로 고쳤다. 위험·한계: V21을 V22 없이 직접 고쳤다(어느 배포에도 아직 없음). 이 브랜치의 V21을 적용한 개발 DB는 다시 만들어야 한다. 서블릿 오류 디스패치를 보안 체인이 허용한다. 발행·저장소 포트 시그니처가 바뀌어 파생 서비스 코드가 깨진다. prod에서 fixture-로 시작하는 설정값은 기동을 막는다. httpclient failure-injection lane의 main 단계 중복 실행은 이전 결정대로 남겼다. 실제 CI runner 실행은 확인하지 못했다. 검증: ci, mainCheck, :integrationCheck, -p optional-platforms ci integrationCheck performanceCheck, :performanceCheck, :releaseCheck (-PgitRevision), verify-compose-profile-contracts(18 lane) 통과. 첫 실행의 실패 2건(FindSecBugs 1, sample 동시 생성 테스트의 v4 event id)은 고친 뒤 ci·integrationCheck·mainCheck를 다시 돌려 통과했다. Refs: docs/reviews/2026-11-03-rereview/README.md 3·4·6·7절 BREAKING CHANGE: outbox 발행 포트와 저장소 포트 시그니처, 순서 번호 첫 값(V21 floor 다음), 샘플 데모 계정 프로필 허용 목록, 새 오류 코드 IDEMPOTENCY_COMPLETION_UNKNOWN, prod의 fixture- 값 거부, 기동 rule은 static·의존 없음이어야 한다.
15 KiB
Configuration reference
The notification delivery platform binds under app.notification.platform. The tree lives in
src/composition/notification-platform/src/main/resources/config/module-defaults/notification.yml,
which the support leaf's ModuleDefaultsEnvironmentPostProcessor loads at the lowest precedence,
so an application without the module carries none of it. The module is in no default composition: a
service selects it with runtimeOnly project(':composition:notification-platform') (plus the
persistence-jpa line) in app-bootstrap/build.gradle, and it is on when present. Every value carries
an inline default, so a deployment that switches the platform off supplies nothing.
Until 2026-08-15 this page named properties the binding did not have — max-retry-concurrency,
scheduler-poll-interval, callback-worker-concurrency — and omitted three it did. There was no
tree in application.yml at all, so the only way to configure the platform was to guess environment
variable names from Boot's relaxed binding. NotificationEnvironmentRegistryTest
(:composition:notification-platform:test) now fails when the shipped YAML and
docs/registries/env-keys.yaml disagree.
Master switch
| Property | Environment variable | Default | Meaning |
|---|---|---|---|
enabled |
APP_NOTIFICATION_PLATFORM_ENABLED |
true |
On while the module is present. false binds nothing at all: no runtime, no schema check, no scheduler thread, no secret required |
mode |
APP_NOTIFICATION_PLATFORM_MODE |
SERVING |
SERVING refuses to start without a working provider; INGEST_ONLY stores requests and does not dispatch: it builds no provider runtime but records each enabled profile's route, so a request accepted now is delivered later on that route by a SERVING deployment (handoff) |
Dispatch
| Property | Environment variable | Default | Bound |
|---|---|---|---|
dispatch.claim-batch-size |
APP_NOTIFICATION_PLATFORM_CLAIM_BATCH_SIZE |
50 |
1..1000 |
dispatch.lease-duration |
APP_NOTIFICATION_PLATFORM_LEASE_DURATION |
2m |
positive, finite |
dispatch.poll-interval |
APP_NOTIFICATION_PLATFORM_POLL_INTERVAL |
1s |
positive, finite |
dispatch.max-global-concurrency |
APP_NOTIFICATION_PLATFORM_MAX_CONCURRENCY |
64 |
positive |
dispatch.max-additional-attempts |
APP_NOTIFICATION_PLATFORM_MAX_ADDITIONAL_ATTEMPTS |
4 |
non-negative |
dispatch.max-queue-age |
APP_NOTIFICATION_PLATFORM_MAX_QUEUE_AGE |
24h |
positive |
dispatch.allow-ambiguous-fallback |
APP_NOTIFICATION_PLATFORM_ALLOW_AMBIGUOUS_FALLBACK |
false |
boolean |
Every value is bounded. "Unlimited" is not an accepted configuration.
The lease must outlast a provider call plus its timeout. Below that, a delivery a live worker is still waiting on gets claimed by a second worker, and the recipient receives the notification twice.
Callbacks
| Property | Environment variable | Default | Bound |
|---|---|---|---|
callbacks.enabled |
APP_NOTIFICATION_PLATFORM_CALLBACKS_ENABLED |
false |
boolean |
callbacks.max-body-bytes |
APP_NOTIFICATION_PLATFORM_CALLBACK_MAX_BODY_BYTES |
65251 |
1..65251 |
callbacks.replay-skew |
APP_NOTIFICATION_PLATFORM_CALLBACK_REPLAY_SKEW |
1h |
positive; SNS messages whose Timestamp is further than this from receipt, either way, are rejected. The default is SNS's longest HTTP/S retry period (3,600 s), so a bounce or complaint SNS redelivers after an outage is still accepted; a replay inside the window is a duplicate in the provider event ledger, not a second event (review 05-20) |
callbacks.trusted-proxies |
APP_NOTIFICATION_PLATFORM_CALLBACK_TRUSTED_PROXIES |
(empty) | CSV of peer addresses |
콜백 응답: 기록(중복 포함) 204, 거절 400(서명·topic·시각·형식), 본문 초과 413. SNS 서명 인증서를 가져오지 못하면(5xx·408·429·네트워크 실패) 거절하지 않고 503+Retry-After로 답한다 — SNS는 4xx를 처리 완료로 보고 재전송을 멈추지만 5xx는 다시 보낸다(review 05-20). 인증서 URL이 그 밖의 4xx(403·404 등)로 답하면 SNS가 발급하지 않은 인증서를 가리킨 위조 콜백이므로 서명 검증 실패로 400 거절하고 보안 감사(callbackSignatureRejected)에 남긴다(review 2026-11-03 RR-07-03). 호스트 검사를 통과한 URL이라 답하는 쪽은 SNS다.
여러 provider가 요청 URL에 서명하므로, 그 URL을 잘못 재구성하면 정상 webhook이 전부 서명 실패가 된다.
trusted-proxies가 비어 있으면 forwarded 헤더를 믿지 않고 컨테이너가 관측한 값을 쓴다. 무조건 믿으면
아무 호출자나 자기 서명이 검증될 URL을 고를 수 있어 서명 자체가 무의미해진다. 로드밸런서 뒤에 있는 배포는
그 peer를 명시한다.
65251 is not a round number by accident: it is the ciphertext column's 65536 bytes minus the largest
envelope the protector writes — a version byte, a key-id length byte, a key id of up to 255 bytes, the
AES-GCM nonce and tag. A larger configured value is refused when the settings bind, so the capability
does not start; before that check existed it passed every layer above the database and failed the
CHECK constraint after the callback had already been acknowledged to the provider. The shipped
default was 65508 (nonce and tag only) for a while after the envelope gained its key id, and every
deployment that turned the platform on failed to start until it was corrected (review 2026-11-01
r3-i, found by booting the compose lanes).
Provider profiles
Profiles are a map under providers, keyed by profile id. Arbitrary deployment-chosen ids can be
supplied as YAML or through Spring relaxed binding (APP_NOTIFICATION_PLATFORM_PROVIDERS_<ID>_<FIELD>).
The shipped reference profiles smtp and ses additionally expose explicit APP_NOTIFICATION_PLATFORM_SMTP_*
and APP_NOTIFICATION_PLATFORM_SES_* aliases so operators do not have to invent binding names.
| Field | Meaning |
|---|---|
type |
SES, SMTP — a closed enum, so an unknown value fails binding rather than assembling into nothing |
enabled |
A disabled profile is bound and validated but contributes no runtime |
primary-for-channel |
Exactly one primary per channel |
environment |
Required when enabled |
credential-profile |
Provider credential key id resolved through SecretMaterialProvider; never inline secret material |
timeout |
Positive and finite |
max-concurrency |
Positive |
rate-per-second |
Positive |
A profile pins provider type, environment, credential profile, timeouts, concurrency and rate limit. Sender identity and credential profile are separate concerns.
SES profile fields
An enabled SES profile adds a typed ses block:
| Field | Meaning |
|---|---|
ses.endpoint |
SES API base URI, used as the AWS SDK client's endpoint override |
ses.region |
AWS region the SDK signs for |
ses.sender-identity |
verified SES From identity |
ses.access-key-id |
AWS access key id; identifier only, not secret material |
ses.configuration-set |
optional SES configuration-set name |
ses.allowed-topic-arns |
SNS topic ARNs whose SES events this profile accepts (APP_NOTIFICATION_PLATFORM_SES_ALLOWED_TOPIC_ARNS, comma-separated). Required when callbacks.enabled=true: binding fails without it, because a valid SNS signature only proves AWS sent the message and anyone can own a topic |
credential-profile |
key id for the secret access-key material; generation 1 is activated at assembly |
The shipped reference profile is providers.ses and is disabled by default. It is controlled by
APP_NOTIFICATION_PLATFORM_SES_*; its secret material comes from
APP_NOTIFICATION_PLATFORM_PROVIDER_CREDENTIAL_KEY, whose active key id must match
APP_NOTIFICATION_PLATFORM_SES_CREDENTIAL_KEY_ID.
type lists only the families the shipped composition can assemble. Current shipped
reachability is:
| Provider type | Contract/adapter code | Shipped dispatch assembler | Shipped callback assembler | What happens when enabled for dispatch |
|---|---|---|---|---|
SMTP |
yes | yes | no | assembles when Spring Mail relay settings are complete |
SES |
yes | yes | yes | assembles an EMAIL runtime from the typed ses block; SNS callback verification remains available in INGEST_ONLY |
APNs, FCM, Twilio, Web Push and Webhook adapters used to ship as protocol code with no assembler, so
enabling one failed startup. They were removed; the docs/capability-status.md "제거된 코드" table
names the last commit that had them. Adding a family back means adding its adapter, a
ProviderRuntimeAssembler, its ProviderType constant and the settings/secret inputs it actually
needs, together.
SMTP relay
The one provider profile the template ships, off. A deployment that wants the common case sets
APP_NOTIFICATION_PLATFORM_SMTP_ENABLED=true and the relay address; one that wants a different
profile id or a second family declares it in its own YAML instead.
The profile and the relay are separate tables below because they answer different questions. The
profile says which provider serves EMAIL and under what limits; the relay says what the transport
is. Host, port and credentials are not here at all — they stay spring.mail.*, because Spring
already owns them and a second spelling would be a second thing to keep in step.
| Property | Environment variable | Default | Bound |
|---|---|---|---|
providers.smtp.enabled |
APP_NOTIFICATION_PLATFORM_SMTP_ENABLED |
false |
boolean |
providers.smtp.primary-for-channel |
APP_NOTIFICATION_PLATFORM_SMTP_PRIMARY |
true |
boolean; exactly one primary per channel |
providers.smtp.environment |
APP_NOTIFICATION_PLATFORM_SMTP_ENVIRONMENT |
local |
required when enabled |
providers.smtp.credential-profile |
APP_NOTIFICATION_PLATFORM_SMTP_CREDENTIAL_PROFILE |
default |
resolved through SecretMaterialProvider, never inline material |
providers.smtp.timeout |
APP_NOTIFICATION_PLATFORM_SMTP_TIMEOUT |
10s |
positive, finite |
providers.smtp.max-concurrency |
APP_NOTIFICATION_PLATFORM_SMTP_MAX_CONCURRENCY |
4 |
positive |
providers.smtp.rate-per-second |
APP_NOTIFICATION_PLATFORM_SMTP_RATE_PER_SECOND |
10 |
positive |
| Property | Environment variable | Default | Bound |
|---|---|---|---|
smtp.sender-identity |
APP_NOTIFICATION_PLATFORM_SMTP_SENDER_IDENTITY |
no-reply@example.invalid |
address |
smtp.dispatch-concurrency |
APP_NOTIFICATION_PLATFORM_SMTP_DISPATCH_CONCURRENCY |
4 |
positive |
SMTP transport configuration has one owner: Spring Mail. The actual JavaMailSenderImpl must use required STARTTLS (mail.smtp.starttls.enable=true plus mail.smtp.starttls.required=true) or implicit TLS (mail.smtp.ssl.enable=true). It must also have positive mail.smtp.connectiontimeout, mail.smtp.timeout, and mail.smtp.writetimeout values. SMTP assembly fails before serving traffic when those properties are absent or invalid.
The native environment contract is SPRING_MAIL_HOST, SPRING_MAIL_PORT, optional
SPRING_MAIL_USERNAME / secret SPRING_MAIL_PASSWORD, plus the
SPRING_MAIL_PROPERTIES_MAIL_SMTP_* TLS and timeout keys listed in .env.example. These names stay
Spring-native because Spring Boot itself owns and consumes spring.mail.*; the notification module
does not create a second alias for the same relay.
The default sender is an RFC 2606 reserved domain that resolves nowhere, so a deployment that forgot to set one produces a traceable bounce instead of mail apparently sent from an address it does not own.
Startup failures
Startup fails rather than degrading when:
- a payload or queue setting is unbounded
- a timeout is negative
- a TTL-required profile has no expiry source
- a callback signing secret is missing
- a production profile enables trust-all
- an SES profile is missing its
sesblock, or callbacks are on andses.allowed-topic-arnsis empty - two provider profiles share an id
- a route points only at disabled providers
modeisSERVINGand no provider profile is enabled- the notification schema stream is not applied and promoted
Secrets
All key material arrives through SecretMaterialProvider. Nothing is read from source, from a
committed file, or from a plaintext log. Contact point encryption and lookup HMAC keys must be
distinct, and the encryption key must be exactly 256 bits.
Seven purposes, seven keys. Each is base64 of at least 32 bytes and each must differ from every other; the platform decodes them at startup and refuses to boot if one is blank, short or shared. A blank value used to be skipped, which meant the platform started without the key and found out on the first contact point — in production, on a recipient's notification.
Every default below is unset, deliberately. Supply the values out of band, per environment. Do
not write one into this table, into application.yml, into an .env file that is tracked, or into
any example: a value that appears in the repository is a value that has been disclosed.
| Purpose | Key material | Active key id |
|---|---|---|
| Contact point encryption | APP_NOTIFICATION_PLATFORM_CONTACT_ENCRYPTION_KEY |
APP_NOTIFICATION_PLATFORM_CONTACT_ENCRYPTION_KEY_ID |
| Contact point lookup HMAC | APP_NOTIFICATION_PLATFORM_CONTACT_LOOKUP_HMAC_KEY |
APP_NOTIFICATION_PLATFORM_CONTACT_LOOKUP_HMAC_KEY_ID |
| Callback signing | APP_NOTIFICATION_PLATFORM_CALLBACK_SIGNING_KEY |
APP_NOTIFICATION_PLATFORM_CALLBACK_SIGNING_KEY_ID |
| Callback fingerprint HMAC | APP_NOTIFICATION_PLATFORM_CALLBACK_FINGERPRINT_HMAC_KEY |
APP_NOTIFICATION_PLATFORM_CALLBACK_FINGERPRINT_HMAC_KEY_ID |
| Provider credential encryption | APP_NOTIFICATION_PLATFORM_PROVIDER_CREDENTIAL_KEY |
APP_NOTIFICATION_PLATFORM_PROVIDER_CREDENTIAL_KEY_ID |
| Provider request lookup HMAC | APP_NOTIFICATION_PLATFORM_PROVIDER_REQUEST_LOOKUP_HMAC_KEY |
APP_NOTIFICATION_PLATFORM_PROVIDER_REQUEST_LOOKUP_HMAC_KEY_ID |
| Payload encryption | APP_NOTIFICATION_PLATFORM_PAYLOAD_ENCRYPTION_KEY |
APP_NOTIFICATION_PLATFORM_PAYLOAD_ENCRYPTION_KEY_ID |
A key id is not secret — an id identifies key material without revealing it — but it is required, and it has no default on purpose. A constant id makes a rotation indistinguishable from the key it replaced, so nothing could decrypt what was written before it. Change the id in the same deployment that changes the material, and keep the superseded key readable under its old id until the data it wrote has been re-encrypted. The rotation sequence is in at-rest-threat-model.md.
Readiness
The platform contributes a notifications actuator endpoint and a health indicator. It reports DOWN
when a provider's credentials were rejected, when a configured provider has no channel route, and
when the measured backlog, stuck-lease count, projection lag or reconciliation lag passes the
thresholds in NotificationServingThresholds. See operations.md.