A multi-membership identity (belongs to more than one tenant) used to
get a flat 501 refusal -- named as undesigned future work across
CLAUDE.md/threat-model.md/the runbook since early Phase 4. Scope for
this change was agreed via AskUserQuestion: backend protocol only,
fully verified via real HTTP round trips, not the actual picker page --
web has zero session/cookie-handling code today (confirmed while
researching this), so building that is separately-scoped, unverifiable
frontend work in this environment (no live backend, no browser).
session.Manager gains IssuePendingLogin/ValidatePendingLogin, a second
JWT token type proving identity without committing to a tenant yet
(10-minute TTL). PendingLoginClaims is deliberately a distinct Go type
from Claims, and -- caught by this change's own test suite before it
shipped -- needed a JSON field name disjoint from Claims.UserID's
"user_id" too: go-jose's unmarshal is happy to populate a struct from
any token whose claims happen to share a key, so a real session token
would otherwise have parsed successfully as a pending login. Fixed via
"pending_user_id" instead; both directions (session-as-pending,
pending-as-session) now have regression tests.
rbacstore.ListMembershipsWithTenantForUser joins tenant_memberships
with tenants, since a picker needs display names, not just IDs.
loginhandler.resolveIdentity's multiple-membership branch no longer
errors -- finishLogin routes it into startTenantSelection instead,
which issues a pending-login cookie (Path=/auth, so it's never sent on
ordinary requests) and redirects to a new configurable
SelectTenantRedirectURL (defaults to {POST_LOGIN_REDIRECT_URL}/select-
tenant). Two new routes complete the round trip: GET /auth/memberships
lists the pending identity's real tenant options, and POST
/auth/select-tenant re-derives the role for the chosen tenant
server-side (never trusts a client-supplied role, refuses a tenant_id
outside the identity's actual memberships with 403) before issuing the
real session -- responding with JSON {"redirect_url": ...}, not a
redirect, since a POST/fetch caller should control its own navigation.
Verified with the same real-fake-IdP tests the rest of this package
uses (coreos/go-oidc's oidctest, crewjam/saml's samlidp): the full
login -> pending cookie -> GET /auth/memberships -> POST
/auth/select-tenant -> real session round trip for both protocols, plus
negative paths (missing/expired pending cookie, a tenant_id outside
membership, a real session token rejected as a pending login and vice
versa). ErrMultipleMemberships is removed -- it's not an error path
anymore.
Docs updated in lockstep: CLAUDE.md, threat-model.md (including its
summary table), phase-4-runbook.md (new §12), enterprise/README.md
(new "Tenant selection" section, explicit about what's still not built
and why: no session handling in web, no CORS on enterprise-auth).
22 KiB
enterprise
Commercial license, not AGPLv3 — see /CLAUDE.md's licensing
boundary. SSO (OIDC/SAML), tenant provisioning, and RBAC. Nothing in
/agent, /ingest, /storage, /api, /web core, or /cli imports
from this module — confirmed by hack/check-tenant-boundary.sh, run in
CI. enterprise/ supplies tenant-scoped implementations of core's
already-shipped api/querylang/executor.SQLRunner/
SearchClient interfaces rather than core growing tenant awareness —
see /docs/phase-4-isolation-design.md for why.
Status
What's built and wired end-to-end. Verification status varies by
piece -- internal/audit was confirmed against a real Postgres earlier
in this phase's work; everything else below has real integration tests
written the same way (skipped unless a live database's connection
details are supplied via env var, same pattern throughout this package)
but they have not actually been run against a live database in this
environment -- see /docs/phase-4-runbook.md's verification-status
section for exactly what "not yet run" means here and why. Don't read
"has a test for this" as "this was confirmed to work."
internal/sessionissues/validates signed (HS256/JWT) tokens for both human sessions and/alerting'sRoleServicecredential.internal/authhandlerservesPOST /internal/authorize(the endpointapi/authz.HTTPAuthorizercalls) andGET /auth/features(the runtime-capability check/web's settings page reads).api's/queryand/dashboardsendpoints enforce RBAC viaauthz.RequireRole/RequireRoleOrService, nil-safe (no-op) whenENTERPRISE_AUTH_URLisn't configured -- matches Phase 0-3 behavior./alerting'squeryclientpresents aRoleServiceBearer token (API_SERVICE_TOKEN) when configured -- see/docs/phase-4-isolation-design.md'salerting↔apigap.sentryctlpresents$SENTRYCTL_TOKENas a Bearer credential on every request when set.internal/rbacstore: full CRUD overusers/tenants/tenant_memberships/data_sources(metadata/migrations/0017-0032).internal/tenantprovision: realCREATE DATABASE/CREATE USER/GRANTagainst ClickHouse. Its tests assert a tenant A user cannot read tenant B's database by fully-qualified name, and thatsystem.query_log/system.tables/SHOW DATABASESdon't leak across tenants either (task 2's finding was that the latter is version-dependent) -- not yet run against a live ClickHouse in this environment, see the note above.internal/chrunner: the tenant-scopedSQLRunner-- a per-tenant connection registry that resolves which tenant's ClickHouse connection to use from the authenticated identity in request context, never a parameter. Same adversarial probe, now through the actual production code path (chrunner.Registry.RunSQL, not just tenantprovision's raw grants).internal/audit.QueryAPILogger: the realapi/queryapi.AuditLoggerimplementation -- wired intoenterprise-api, no longernil.internal/loginhandler:GET /auth/oidc/login+GET /auth/oidc/callback-- the actual human login flow, previously entirely missing. Redirects to the configured IdP with CSRF-protection state in a short-lived cookie, exchanges the code, verifies the ID token viainternal/oidc, upserts ausersrow, resolves tenant/role fromtenant_memberships(refuses with a clear error on zero rows; more than one starts a real tenant-selection round trip --GET /auth/memberships+POST /auth/select-tenant, backed by a short-livedsession.Manager.IssuePendingLogintoken -- rather than guessing; see "Tenant selection" below), and issues a session cookie. This one genuinely is verified, unlike the ClickHouse pieces above:loginhandler_test.goruns the full flow against a real fake IdP (coreos/go-oidc's ownoidctestpackage, real RS256 signing and verification, no live database or Docker needed) and every test passes. Not yet tried against a real external IdP or a runningenterprise-authcontainer.internal/searchclient: the Tantivy-side sibling ofchrunner-- implementsapi/querylang/executor.SearchClient, resolvingSearchRequest.tenant_id(new field,proto/sentry/search/v1/ search.proto) from the authenticated request identity, same fail-closed shape aschrunner.Registry.RunSQL. Paired withsearch/src/registry.rs'sIndexRegistry(Rust, opens a per-tenant Tantivy index on demand). Both genuinely verified -- unlike the ClickHouse pieces, Tantivy is an embedded library, so the isolation probe (three tenants, shared search term, scoped search returns only that tenant's document) actually ran:search'scargo test/cargo clippy --all-targets -- -D warningsand this package'sgo testboth pass clean, no Docker or live database needed for either.Clientalso carries aTenantChecker(backed byrbacstore.TenantIsActive) sincesearch/src/registry.rs'sIndexRegistryopens-or-creates an index for any syntactically-validtenant_id-- a real gap found while closing/docs/phase-4-isolation-design.md's verification-plan item 4: a mid-provisioning tenant would otherwise get a silently-empty search result instead of a refusal. Verified the same Docker-free way.cmd/enterprise-api: a second binary (alongsideapi/cmd/api, unchanged) importing bothapi's handler packages and the tenant-aware implementations above -- see its own doc comment for why this shape exists (enterprise → apiis the allowed import direction;apican never importenterprise/).-provision-tenant=<id>is the operator action that provisions ClickHouse and marks a tenant active, same "offline action, not a network endpoint" shape asenterprise-auth -mint-service-token.
OIDC and SAML login are both now fully wired: internal/loginhandler
serves GET /auth/oidc/login+GET /auth/oidc/callback and
GET /auth/saml/login+POST /auth/saml/acs, converging on the same
upsert-user/resolve-tenant/issue-session path. Both are verified the
same way -- a real fake IdP with genuine cryptographic signing and
verification (coreos/go-oidc's oidctest for OIDC,
crewjam/saml/samlidp for SAML), no Docker needed, every test in
loginhandler_test.go/saml_test.go passing including the full login
round trip and negative paths (bad state/InResponseTo, expired/missing
credential, no/multiple tenant memberships). Writing the SAML test
caught two real bugs in internal/saml.ParseResponse, both fixed:
missing r.ParseForm() before reading the POSTed SAMLResponse field,
and email-attribute matching that missed the standard LDAP "mail" OID
(urn:oid:0.9.2342.19200300.100.1.3) that IdPs send by default absent
an explicit AttributeConsumingService request for "email" -- exactly
what samlidp's own default assertion builder does. Neither protocol
has been tried against a real external IdP or a running
enterprise-auth container -- see /docs/phase-4-runbook.md §3a/§3b.
dashboard_permissions is now wired end to end: rbacstore/ dashboard_permissions.go is the raw CRUD, rbacstore/ dashboards_adapter.go's DashboardPermissions implements
api/dashboards.PermissionStore (the interface core defines and
carries as a nil-by-default field, same shape as
queryapi.AuditLogger), and enterprise-api's main.go wires it in.
api/dashboards' handler now enforces the matrix's "(own/granted)"
qualifier: an Editor may edit/delete a dashboard (or its panels) they
created, or one where a grant raises their effective role to Editor;
managing grants themselves is stricter still -- creator or Admin/Owner
only, closing a self-escalation path where a granted-but-not-creator
Editor could otherwise extend their own access. metadata/migrations/ 0033_restrict_dashboard_permissions_role.sql fixes a divergence
between 0024's actual CHECK constraint (allowed role='admin', nullable
granted_by) and the design doc's schema (viewer/editor only,
granted_by required) found while wiring this up. Verified against a
fake PermissionStore in api/dashboards/handler_test.go (the full
own/granted/admin/creator matrix, including the granted-editor-cannot-
manage-grants regression); real integration tests exist in
rbacstore_test.go but haven't run against a live Postgres in this
environment, same disclosed gap as the rest of this package's
Postgres-backed pieces.
Tenant selection (multi-membership identities)
The backend protocol for choosing a tenant is built and tested; the
frontend page that would actually call it is deliberately not (see
"Deliberately deferred" below). When resolveIdentity finds more than
one tenant_memberships row for a logged-in identity, finishLogin
issues a session.Manager.IssuePendingLogin token (a distinct Go/JWT
type from a real session -- see that type's doc comment for a real bug
this design caught in its own tests: a shared JSON key would have let a
full session token double as a pending login) as a
sentry_pending_login cookie (Path=/auth) and redirects to
SELECT_TENANT_REDIRECT_URL (defaults to
{POST_LOGIN_REDIRECT_URL}/select-tenant) instead of completing the
login. From there:
GET /auth/memberships-- lists the pending identity's tenants (tenant_id,tenant_display_name,role) to choose between.POST /auth/select-tenantwith{"tenant_id": "..."}-- re-derives the role for that specific tenant server-side (never trusts a client-supplied role, only checks the claimedtenant_idagainst the identity's actual memberships, refusing with 403 otherwise), issues the real session cookie, and responds with{"redirect_url": "..."}for the caller to navigate to -- JSON, not a redirect, since this is a POST/fetch call a frontend page should control the navigation for itself.
Verified with the same real-fake-IdP tests as the rest of this package
(loginhandler_test.go/saml_test.go): the full login -> pending
cookie -> GET /auth/memberships -> POST /auth/select-tenant -> real
session round trip, plus negative paths (missing/expired/wrong-type
pending cookie, a tenant_id outside the identity's actual
memberships, a real session token rejected when presented as a pending
login).
Deliberately deferred, not half-built -- named explicitly rather than silently left out:
- The actual tenant-picker page -- nothing in
web/calls the endpoints above yet. Building it is genuinely different, larger scope than the backend protocol:webhas zero session/cookie-handling code today (confirmed by reading it end to end while designing this), so a real picker page means adding that from scratch, plus CORS wiring (enterprise-authhas no CORS middleware at all right now -- a cross-originfetchwith credentials fromweb's origin needs it), neither of which is verifiable in this environment without a live backend and a browser session to exercise. - Ingest tenant-awareness, for either storage engine --
chrunner/searchclientprove read isolation given tenant-scoped data exists, but nothing writes it: every recordingestproduces still lands in the single shared ClickHouse database and the single shared Tantivy index. A newly-provisioned tenant's storage is real and isolated, and permanently empty. Undesigned, not just unbuilt -- see/docs/security/threat-model.md. - Any deployment-topology mechanism that actually routes traffic to
enterprise-apiinstead ofapi-- both binaries exist,docker-compose.ymlincludesenterprise-apiavailable but not wired intoweb's default base URL, and the Helm chart has no service for it at all yet. This is now the single largest gap -- both storage engines' isolation mechanisms themselves are built.
Package layout
cmd/enterprise-auth/ config loading, OIDC discovery at startup, health/authorize/features endpoints, -mint-service-token, -create-tenant, -grant-membership-*, -revoke-membership-*, -list-memberships-tenant
cmd/enterprise-api/ multi-tenant-aware alternative to api/cmd/api -- see its own doc comment
internal/tenant/ the ID type -- see its package doc comment before touching it
internal/oidc/ coreos/go-oidc wiring: discovery, login redirect, code exchange + ID token verification
internal/saml/ crewjam/saml wiring: SP setup, login redirect, response parsing/validation
internal/session/ issues/validates signed session + RoleService tokens
internal/authhandler/ POST /internal/authorize, GET /auth/features
internal/loginhandler/ GET /auth/oidc/{login,callback} + GET /auth/saml/login + POST /auth/saml/acs -- the human login flow
internal/rbacstore/ users/tenants/tenant_memberships/data_sources/dashboard_permissions CRUD (pgx against sentry_metadata)
internal/tenantprovision/ real ClickHouse CREATE DATABASE/USER/GRANT
internal/tenantcrd/ syncs -provision-tenant's real result into deploy/operator's Tenant CRD (K8s dynamic client, no cluster needed to test)
internal/chrunner/ tenant-scoped api/querylang/executor.SQLRunner
internal/searchclient/ tenant-scoped api/querylang/executor.SearchClient
internal/audit/ append-only, hash-chained query audit log, plus the
api/queryapi.AuditLogger adapter (queryapi_adapter.go)
internal/apiconfig/ enterprise-api's own env-var config
internal/config/ enterprise-auth's env-var config
Future additions: ingest tenant-awareness (undesigned), and real
deployment-topology wiring for enterprise-api -- see "Status" above.
Why OIDC and SAML aren't hand-rolled
coreos/go-oidc (built on golang.org/x/oauth2) and crewjam/saml
handle token/assertion signature verification, XML signing, and the
protocol-level trust establishment — exactly the parts of an SSO
integration where a from-scratch implementation is the highest-risk
code in the whole feature. Both are well-established libraries, matching
this project's existing "boring, well-understood dependency" pattern
(clickhouse-go/v2, jackc/pgx/v5).
Building & testing
go build ./...
go vet ./...
go test ./...
internal/audit's real guarantees (the audit_writer grant
restriction, the immutability trigger, hash-chain correctness under
concurrency) can only be proven against a real Postgres — those
integration tests are skipped by default and only run with
AUDIT_TEST_POSTGRES_ADDR set:
docker run --rm --network sentry_default -v $(pwd)/..:/src -w /src/enterprise \
-e AUDIT_TEST_POSTGRES_ADDR=metadata-postgres:5432 \
-e AUDIT_TEST_POSTGRES_PASSWORD=audit-writer-dev-only \
-e AUDIT_TEST_ADMIN_PASSWORD=sentry-dev-only \
golang:1.25-alpine go test ./internal/audit/... -v
internal/rbacstore's tests are the same shape (real SQL, real
constraints), skipped unless RBACSTORE_TEST_POSTGRES_ADDR is set:
docker run --rm --network sentry_default -v $(pwd)/..:/src -w /src/enterprise \
-e RBACSTORE_TEST_POSTGRES_ADDR=metadata-postgres:5432 \
-e RBACSTORE_TEST_POSTGRES_PASSWORD=sentry-dev-only \
golang:1.25-alpine go test ./internal/rbacstore/... -v
internal/tenantprovision and internal/chrunner need a real
ClickHouse instead (they mount the repo root, not just enterprise/,
since internal/chrunner imports api/authz/api/querylang/executor
via go.mod's replace directives to ../api):
docker run --rm --network sentry_default -v $(pwd)/..:/src -w /src/enterprise \
-e TENANTPROVISION_TEST_CLICKHOUSE_ADDR=clickhouse:9000 \
-e TENANTPROVISION_TEST_CLICKHOUSE_PASSWORD=sentry-dev-only \
golang:1.25-alpine go test ./internal/tenantprovision/... -v
docker run --rm --network sentry_default -v $(pwd)/..:/src -w /src/enterprise \
-e CHRUNNER_TEST_CLICKHOUSE_ADDR=clickhouse:9000 \
-e CHRUNNER_TEST_CLICKHOUSE_PASSWORD=sentry-dev-only \
golang:1.25-alpine go test ./internal/chrunner/... -v
Turning on auth enforcement for manual testing
Off by default (see "Status" above). To exercise the RoleService path
end to end:
docker compose up -d enterprise-auth
TOKEN=$(docker compose run --rm enterprise-auth -mint-service-token=alerting)
# api: set ENTERPRISE_AUTH_URL=http://enterprise-auth:8082 and restart
# alerting: set API_SERVICE_TOKEN=$TOKEN and restart
docker build -f Dockerfile -t sentry-enterprise-auth . # context is enterprise/, not the repo root
Bootstrapping a tenant and its first human user
-create-tenant/-grant-membership-* are enterprise-auth operator
flags, same "offline action gated by access to enterprise-auth's own
environment, not a network-reachable endpoint" shape as
-mint-service-token -- deliberately not an authenticated HTTP admin
API, which would have to solve "who's allowed to create the very first
tenant/membership" itself. Replaces what used to be a manual psql
dance (see /docs/phase-4-runbook.md §3a's history if you're wondering
why old references to it still show up in git blame):
docker compose run --rm enterprise-auth -create-tenant=acme -display-name="Acme Corp"
# Log in once via /auth/oidc/login or /auth/saml/login -- it fails with
# "no tenant membership" (403), but UpsertUserBySSO already created the
# users row by then, which -grant-membership-user-email needs.
docker compose run --rm enterprise-auth \
-grant-membership-tenant=acme -grant-membership-user-email=[email protected] -grant-membership-role=owner
# See who's actually in a tenant, and take access away again:
docker compose run --rm enterprise-auth -list-memberships-tenant=acme
docker compose run --rm enterprise-auth \
-revoke-membership-tenant=acme -revoke-membership-user-email=[email protected]
-create-tenant only touches rbacstore -- pair with enterprise-api -provision-tenant (below) for a tenant to actually be able to run
queries, not just log in. role=owner also calls SetOwner, since a
tenant's Owner is a dedicated tenants.owner_user_id column, not just
the highest tenant_memberships role -- -revoke-membership-* refuses
to revoke a tenant's current Owner for the same reason (transferring
ownership first has no flag yet, only rbacstore.SetOwner at the
storage layer). Changing a role is just re-running
-grant-membership-* with a different -grant-membership-role
(SetMembership's upsert already supports it). Not yet built: a flag
for dashboard_permissions grants (those go through the HTTP endpoints
api/dashboards' handler now exposes -- PUT/DELETE /dashboards/{id}/permissions/{userId}, GET .../permissions).
Provisioning a tenant and running enterprise-api
api/enterprise-api are mutually exclusive in docker-compose.yml,
gated behind COMPOSE_PROFILES (.env checks in single-tenant, i.e.
plain api, as the zero-config default -- mirrors Helm's
enterprise.enabled flag). -provision-tenant itself doesn't bind a
port, so it runs fine regardless of the active profile; actually
serving traffic on enterprise-api needs the enterprise profile
active, since it now binds the same host port (8080) plain api does
(it gets a default.aliases: [api] network alias too, so alerting's
API_QUERY_URL/web's VITE_API_BASE_URL need zero changes either
way):
COMPOSE_PROFILES=enterprise docker compose build enterprise-api # context is the repo root, not enterprise/ -- see cmd/enterprise-api/Dockerfile
COMPOSE_PROFILES=enterprise docker compose run --rm enterprise-api -provision-tenant=acme -display-name="Acme Corp"
COMPOSE_PROFILES=enterprise docker compose up -d enterprise-api
curl -s http://localhost:8080/healthz
docker-compose.yml has no Kubernetes cluster to sync into, so
TENANT_CRD_NAMESPACE is never set here -- -provision-tenant's
internal/tenantcrd sync step is a documented no-op in this deployment
shape, same as everywhere else this codebase has an "off unless
configured" optional dependency. It only does anything in a real
cluster with deploy/helm/sentry's tenantOperator.enabled=true -- see
/deploy/helm/sentry/README.md's "Trying the two-tenant example."
-provision-tenant creates the tenant/data_source rows in rbacstore if
they don't exist, provisions ClickHouse, persists the credentials, and
marks the tenant active -- refuses to run twice for the same tenant
(re-provisioning would either rotate a live credential or silently fail
to, see tenantprovision.ProvisionClickHouse's doc comment).
Environment variables (enterprise-auth)
| Var | Default |
|---|---|
HTTP_LISTEN_ADDR |
:8082 |
POSTGRES_ADDR |
localhost:5432 |
POSTGRES_DATABASE |
sentry_metadata |
POSTGRES_USERNAME |
sentry |
POSTGRES_PASSWORD |
(empty) |
OIDC_ISSUER_URL |
(empty — OIDC discovery skipped if unset) |
OIDC_CLIENT_ID |
(empty) |
OIDC_CLIENT_SECRET |
(empty) |
OIDC_REDIRECT_URL |
(empty — must be <enterprise-auth base URL>/auth/oidc/callback, registered with the IdP) |
SAML_ENTITY_ID |
(empty) |
SAML_ACS_URL |
(empty) |
SAML_IDP_METADATA_URL |
(empty — SAML disabled if unset; if set, fetched and parsed at startup via samlsp.FetchMetadata, same trust level as OIDC_ISSUER_URL's discovery fetch) |
ENTERPRISE_SESSION_SIGNING_KEY |
required, min 32 bytes |
POST_LOGIN_REDIRECT_URL |
http://localhost:3000 — where the browser lands after internal/loginhandler sets a session cookie |
SELECT_TENANT_REDIRECT_URL |
{POST_LOGIN_REDIRECT_URL}/select-tenant — where the browser lands for a multi-membership identity instead; nothing serves this route yet, see "Tenant selection" above |
Environment variables (enterprise-api)
| Var | Default |
|---|---|
HTTP_LISTEN_ADDR |
:8083 |
CLICKHOUSE_ADDR |
localhost:9000 |
CLICKHOUSE_ADMIN_USERNAME |
default |
CLICKHOUSE_ADMIN_PASSWORD |
(empty) |
SEARCH_GRPC_ADDR |
localhost:50052 |
POSTGRES_ADDR |
localhost:5432 |
POSTGRES_DATABASE |
sentry_metadata |
POSTGRES_USERNAME |
sentry |
POSTGRES_PASSWORD |
(empty) |
AUDIT_WRITER_USERNAME |
audit_writer |
AUDIT_WRITER_PASSWORD |
(empty) |
ENTERPRISE_AUTH_URL |
(empty — RBAC becomes a no-op, but chrunner.Registry.RunSQL still refuses every query with no resolved tenant identity, so leaving this unset does not mean "open access," it means "every query fails") |
CORS_ALLOWED_ORIGIN |
* |
QUERY_TIMEOUT_SECONDS |
30 |