Update Phase 4 docs: the ClickHouse-side live verification actually ran

Docker access became available and §§1-10, 10a, 13, 14, and most of §8
of the runbook have now genuinely been run against a real docker-
compose stack, not just documented as a procedure to run. Rewrites the
runbook's "Verification status" section and the threat model's "Read
this first" finding to describe what was actually confirmed (including
the six real bugs this pass found and fixed) versus what's still gated
on a real external IdP (§3a/§3b/§12) or a real Kubernetes cluster
(§7/§11's live-cluster halves). Also fixes two stale runbook commands:
§5a/§6's docker run mounts needed the repo root, not just enterprise/,
for the same replace-directive reason the Dockerfile fix does; §14 cited
a chwriter test name that never existed.
This commit is contained in:
2026-08-15 17:17:35 -07:00
parent c5c68f22e9
commit 010f66ec70
2 changed files with 187 additions and 128 deletions
+98 -77
View File
@@ -6,75 +6,78 @@ logging, and a Kubernetes deployment path. Read those first.
## Verification status — read this before the rest of this doc
Every prior phase's runbook documents claims **checked against the live
stack**, not asserted. This one is different, and says so plainly rather
than papering over it: for the great majority of this phase's work,
**there was no working Docker daemon access and no reachable Kubernetes
cluster**, so most of what follows is a *procedure to run*, not a report
of what was already run and passed. Five genuine exceptions so far:
**Docker daemon access became available partway through this phase's
work, and the rest of this runbook (§§110, 10a, 13, 14, most of §8) has
now genuinely been run against a real docker-compose stack** — real
ClickHouse (`clickhouse/clickhouse-server:24.8`), real Postgres, real
`enterprise-auth`/`enterprise-api`/`enterprise-ingest` containers, two
real provisioned tenants (`acme`, `globex`). This closed every
previously-disclosed "written but never run" gap for the ClickHouse
side, and along the way found and fixed six real bugs no amount of
Docker-free testing could have caught:
- `enterprise/internal/audit`'s hash-chain, tamper-detection, and
concurrent-write guarantees (task 4) -- verified live against a real
Postgres earlier in this phase's work (see its own doc comments for
the exact `docker run` invocations), before the environment lost
Docker access.
- `enterprise/internal/loginhandler`'s full OIDC login flow (§3a) --
verified with real cryptography (a fake IdP that signs and verifies
genuine RS256 tokens) *without* needing Docker or a live database at
all, so this one was actually run in this runbook's own session, not
just an earlier one. What's still unverified is wiring it into a real
running `enterprise-auth` container against a real external IdP.
- `enterprise/internal/loginhandler`'s full SAML login flow (§3b) --
same bar as OIDC above, verified against a real fake SAML IdP
(`crewjam/saml/samlidp`: genuine XML signing and signature
verification, a real `AuthnRequest`/`Response` round trip), no Docker
needed. Writing this test caught two real bugs in
`enterprise/internal/saml`, now fixed: `ParseResponse` never called
`r.ParseForm()` before reading the POSTed `SAMLResponse` field (every
real ACS POST would have silently decoded to nothing), and the email-
attribute matching didn't recognize `urn:oid:0.9.2342.19200300.100.1.3`
(the standard LDAP "mail" OID), which is what an IdP sends by default
when the SP doesn't explicitly request an attribute literally named
"email" -- crewjam's own fake IdP hit this path. Same remaining gap as
OIDC: not yet tried against a real external IdP or a running
`enterprise-auth` container.
- Tantivy tenant isolation, both directions -- `search/src/registry.rs`'s
cross-tenant read isolation (§9), `search/src/consumer.rs`'s per-tenant
write-routing (§14), and `search/src/tenants.rs`'s active-tenant gate
(§14) -- verified live, no disclaimer needed, because Tantivy is an
embedded library with no Docker/broker dependency, and the active-
tenant gate's only external dependency (`enterprise-auth`'s HTTP API)
was exercised against a real hand-rolled test server, not a live
container: real indices, real documents, real commits, real HTTP
requests, all run in this environment. The one thing about it that's
still unverified is not Tantivy itself but the upstream credential/
header plumbing feeding it (ingest's `TenantResolver`, `enterprise-auth`'s
`/internal/authorize-ingest` and `/internal/active-tenants`) against a
real running stack.
- The tenant-picker frontend page (§12) -- the first frontend-only piece
in this phase exercised in a real browser rather than only
type-checked: `web/src/routes/select-tenant`'s cross-origin
credentialed fetch/CORS/cookie handling, driven end-to-end via
`mcp__claude-in-chrome` against a throwaway server standing in for
`enterprise-auth`'s exact wire contract. What's unverified is the same
shape as OIDC/SAML above: this round trip against a real running
`enterprise-auth` container, not a stand-in.
1. `enterprise/Dockerfile` used a `context: enterprise` build context
too narrow for `enterprise/go.mod`'s `replace ../api` directive once
`enterprise-auth` started importing `api/httpserver` and (transitively)
`api/dashboards` -- fixed to build from the repo root, matching
`enterprise-api`/`enterprise-ingest`'s Dockerfiles.
2. `api/dashboards.Store.AddPanel`/`UpdatePanel` didn't call
`validatePanel` the way `CreateDashboard`'s inline panel-creation path
already did, so a panel added via those two methods could hit a
`viz_config` NOT NULL constraint violation instead of getting the
same default-empty-JSON treatment every other panel-creation path
gets -- fixed by calling `validatePanel` in both.
3. `enterprise/internal/rbacstore.SetDataSourceClickHouseCredentials`
let a malformed (non-UUID) id leak Postgres's raw `22P02` error past
the store's `ErrNotFound` boundary instead of treating "can't
possibly match a row" the same as "no such row" -- fixed by
translating that specific Postgres error code to `ErrNotFound`.
4. `enterprise/cmd/enterprise-api/main.go` registered `GET /healthz`
twice -- once explicitly, once already covered by
`queryapi.Handler.RegisterRoutes` -- which panics `net/http`'s
`ServeMux` on startup. This meant `enterprise-api` could never
actually start; every previous "built" claim for this binary had only
ever been a successful `go build`, never a successful process start.
Fixed by deleting the redundant registration.
5. ClickHouse's `default` admin user genuinely lacked `CREATE USER`
privilege in `docker-compose.yml` -- the official image needs
`CLICKHOUSE_DEFAULT_ACCESS_MANAGEMENT=1` (not the more obvious-looking
`CLICKHOUSE_ACCESS_MANAGEMENT`, confirmed by reading the image's own
`/entrypoint.sh` after the wrong name silently did nothing), which
this compose file never set. Every tenant-provisioning code path was
correct Go that had simply never been able to authenticate its own
admin connection strongly enough to run.
6. `enterprise/internal/tenantprovision.ProvisionClickHouse` relied on
ClickHouse RBAC's assumed "default-deny for a freshly created user"
for `system.*` access -- verified live to be false on ClickHouse
24.8: a fresh tenant user could read `system.tables` (though not, it
turns out, `system.query_log`, which is genuinely access-checked).
Fixed with an explicit `REVOKE SELECT ON system.* FROM <user>`, and
`TestProvisionedUserCannotReadSystemTables` was corrected to check
what each system table actually does under a proper revoke (hard
deny for `query_log`, verified-empty for `tables`) instead of
assuming both hard-deny identically. See
`/docs/security/threat-model.md`'s "Read this first" section for the
full account.
Everything else — `internal/rbacstore`'s CRUD, the auth-enforcement
walkthrough, the dashboards tenant-scoping fix, the Helm chart, the
tenant-operator, and (newest) `internal/tenantprovision`/
`internal/chrunner`'s live-ClickHouse tests — has unit/fake-client/
`helm template` coverage (all passing, see each component's own `go
test`/`helm lint` output, including every `Skip*`-gated integration test
confirmed to skip cleanly offline) but has **not** been exercised
against a real running stack. Be specific when citing this runbook: "the
tests exist and pass structurally" is a true, verified claim; "isolation
was confirmed against real ClickHouse" is not, yet. If you're reading
this to decide whether Phase 4 is production-ready: it isn't yet,
independent of this gap — see `/docs/security/threat-model.md`'s
headline finding. This runbook exists so the first person with real
Docker/K8s access can actually close the loop, not to claim that already
happened.
**What's still not verified, and why**: §3a (OIDC) and §3b (SAML) both
need a real external IdP (Auth0/Okta developer tenant, or similar) that
this environment doesn't have credentials for -- the fake-IdP tests
(real RS256/XML crypto, no stand-in shortcuts) are the strongest
evidence available without one. §12's frontend-against-a-real-
`enterprise-auth`-container check is blocked by the same root cause:
there's no way to land on `/select-tenant` with a genuine pending-login
cookie from the real binary without a real IdP redirecting through it
first. §7 and §11's live-cluster steps need `kind`/`kubectl`, which
aren't installed in this environment -- their offline-only checks
(`go build`/`go vet`/`go test`, `helm lint`, `helm template` + parsing
the rendered YAML) all pass and are documented as such below.
If you're reading this to decide whether Phase 4 is production-ready:
closer than before, but not yet -- see
`/docs/security/threat-model.md`'s "Read this first" section for the
current, precise state of every control, including the ones still
gated on a real IdP or a real cluster.
## 1. Bring up the stack
@@ -283,12 +286,18 @@ PermissionStore`) have real integration tests, same skip-gated shape as
§6 below:
```sh
docker run --rm --network sentry_default -v $(pwd)/enterprise:/src -w /src \
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/... -run DashboardPermission -v
```
(Mount the repo root, not just `enterprise/`, with `-w /src/enterprise` as
the workdir -- same reason as `enterprise/Dockerfile`'s doc comment:
`dashboards_adapter.go` imports `api/dashboards`, and `enterprise/go.mod`'s
`replace ../api` directive needs that sibling directory actually present
in the build context, not just the package being tested.)
Expect all `TestDashboardPermission*`/`TestSetDashboardPermission*`/
`TestGetDashboardPermission*`/`TestRevokeDashboardPermission*`/
`TestListDashboardPermissions` tests to pass. This only takes effect
@@ -298,18 +307,24 @@ when `enterprise-api` (not plain `api`) is serving traffic -- see
## 6. `enterprise/internal/rbacstore` and `internal/audit` (already verified — reconfirm here)
```sh
docker run --rm --network sentry_default -v $(pwd)/enterprise:/src -w /src \
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
docker run --rm --network sentry_default -v $(pwd)/enterprise:/src -w /src \
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
```
(Same repo-root-mount reasoning as §5a above -- `internal/rbacstore`
imports `api/dashboards` unconditionally via `dashboards_adapter.go`, so
even running the whole package's tests, not just the `DashboardPermission`
subset, needs `api/` present. `internal/audit` has the same shape via
`queryapi_adapter.go`.)
## 7. `deploy`: Helm chart and Operator (offline-only so far — see `/deploy/README.md`)
No live cluster was available to `kubectl apply` any of this. What can
@@ -723,17 +738,23 @@ go test ./internal/tenantprovision/... -run TestProvisionedUserCanInsertIntoOwnD
# (tenantprovision.go). This test proves the fix, not just documents it,
# whenever a real ClickHouse is available to run it against.
go test ./internal/chwriter/... -run TestRegistryRoutesToCorrectTenant -v
go test ./internal/chwriter/... -run TestRegistryWritesEachTenantToItsOwnDatabase -v
# skip-gated (CHWRITER_TEST_CLICKHOUSE_ADDR) -- writes a mixed batch
# spanning two tenants in one WriteBatch call and confirms each row
# lands in its own tenant's database, none in the other's.
# lands in its own tenant's database, none in the other's. (Fixed from a
# stale `-run TestRegistryRoutesToCorrectTenant` -- that name never
# existed in this package; caught by actually running this command while
# closing out the live verification pass, same class of stale--run
# mistake §6's `TestRegistry` note already flagged once in this doc.)
```
**Not run in this environment**: no live ClickHouse was available while
this was built, so the skip-gated tests above are correct Go that has
never actually executed -- "the test exists" is not the same claim as
"write-routing is confirmed," same caveat §8 already states for the
read-side `chrunner` tests.
**Now genuinely run against a live ClickHouse** (previously only
skip-gated, correct Go that had never executed): all 7
`internal/chwriter` tests pass, including
`TestRegistryWritesEachTenantToItsOwnDatabase`,
`TestRegistryRefusesUnprovisionedTenant`, `TestRefreshAddsNewlyActiveTenant`,
and `TestRefreshRemovesNoLongerActiveTenant` -- write-routing is confirmed,
not just written, as of this pass.
**Tantivy's side is built too, and genuinely verified.**
`search/src/consumer.rs` reads the same `tenant_id` Kafka header and
+89 -51
View File
@@ -9,39 +9,68 @@ for the full design rationale behind the controls described here.
## Read this first: the single most important open finding
**Updated a sixth time.** This section originally read "log data queried
through `POST /query` is not tenant-isolated at all," then "ClickHouse
is isolated but Tantivy isn't," then "ingest tags records with a tenant
identity but nothing routes the write," then "ClickHouse write-routing
is built but Tantivy's isn't," then "both are write-routed but neither
rechecks tenant-active status live," then "both recheck, but ClickHouse's
snapshot never refreshes while Tantivy's does." Both storage engines are
now isolated on both the read and write paths, **both gate writes on an
active-tenant check, and both now refresh that check periodically** —
`chwriter.Registry.StartRefreshing` (new) closes the asymmetry the
previous version of this section named: ClickHouse's writer map now
re-lists active tenants every minute, the same interval
`tenants.ActiveTenantTracker` already used on the Tantivy side, opening
**Updated a seventh time — and this time the headline actually changes.**
This section originally read "log data queried through `POST /query` is
not tenant-isolated at all," then "ClickHouse is isolated but Tantivy
isn't," then "ingest tags records with a tenant identity but nothing
routes the write," then "ClickHouse write-routing is built but Tantivy's
isn't," then "both are write-routed but neither rechecks tenant-active
status live," then "both recheck, but ClickHouse's snapshot never
refreshes while Tantivy's does," then (sixth) "both engines are built
and code-complete, but the ClickHouse half has never actually run
against a real ClickHouse." That last gap is now closed: Docker access
became available, and every ClickHouse-dependent piece named below —
`tenantprovision`, `chrunner`, `chwriter`, the `system.*` metadata
isolation check, and `enterprise-api` itself actually starting and
serving traffic — has been run against a real
`clickhouse/clickhouse-server:24.8` and a real Postgres, not just
type-checked or run against fakes. Closing this loop caught and fixed
six real bugs that no amount of Docker-free testing could have found:
a Docker build-context bug that made `enterprise-auth` fail to build at
all, a validation gap in `dashboards.Store.AddPanel`/`UpdatePanel`, a
raw Postgres error leaking past `rbacstore`'s `ErrNotFound` boundary,
`enterprise-api` panicking on startup from a duplicate `GET /healthz`
route registration, ClickHouse's `default` user genuinely lacking
`CREATE USER` privilege until `CLICKHOUSE_DEFAULT_ACCESS_MANAGEMENT=1`
was added to `docker-compose.yml`, and — most load-bearing — a freshly
provisioned tenant user was **not** actually default-denied from
`system.*` on this ClickHouse version, exactly the risk task 2's
original design flagged as needing live verification rather than trust.
See `/docs/phase-4-runbook.md` §§114 for the full list of what was run
and what each finding was.
Both storage engines are isolated on both the read and write paths,
**both gate writes on an active-tenant check, and both refresh that
check periodically** — `chwriter.Registry.StartRefreshing` re-lists
active tenants every minute, the same interval
`tenants.ActiveTenantTracker` uses on the Tantivy side, opening
connections for newly-active tenants and closing/removing ones no
longer active — a deprovisioned tenant now loses ClickHouse write access
longer active — a deprovisioned tenant loses ClickHouse write access
within a minute, not "until the next `enterprise-ingest` restart."
Neither engine does a live per-write check (a database/HTTP round trip
per record would be a real throughput cost neither implementation
accepts), so a minute-wide staleness window remains on both sides by
design, not by oversight. What's left is narrower still: **whether a
given deployment actually runs the isolated binaries** (deployment-time,
not code-level). See below for both engines' write-routing, in full.
design, not by oversight. What's left: two tenants (`acme`, `globex`)
have been provisioned and exercised end-to-end in this environment, but
only via direct ClickHouse-user connections and Go integration tests —
not yet via a real logged-in human session walking through `POST
/query` in a browser, since that still needs OIDC/SAML wired to a real
external IdP (see §3a/§3b below). And **whether a given deployment
actually runs the isolated binaries** remains a deployment-time
decision, not a code-level guarantee — see below.
**ClickHouse (the SQL path) is built.** `enterprise/internal/
tenantprovision` (real `CREATE DATABASE`/`CREATE USER`/`GRANT`) and
`enterprise/internal/chrunner` (a per-tenant connection registry
implementing `api/querylang/executor.SQLRunner`, resolving the tenant
from request identity, never a parameter) are wired into
`enterprise/cmd/enterprise-api`. **Not yet confirmed against a real
ClickHouse** — this environment had no Docker/database access while
these were written; the tests exist and are correct Go, but "the test
exists" is not the same claim as "isolation is confirmed" (see
`/docs/phase-4-runbook.md`).
**ClickHouse (the SQL path) is built and now genuinely verified live.**
`enterprise/internal/tenantprovision` (real `CREATE DATABASE`/`CREATE
USER`/`GRANT`/`REVOKE`) and `enterprise/internal/chrunner` (a per-tenant
connection registry implementing `api/querylang/executor.SQLRunner`,
resolving the tenant from request identity, never a parameter) are wired
into `enterprise/cmd/enterprise-api`, which has itself been built,
started, and confirmed serving traffic on a real docker-compose stack.
Every ClickHouse-dependent test in both packages passes against a real
`clickhouse/clickhouse-server:24.8`, including the cross-tenant raw-SQL
probe and the corrected `system.*` metadata-isolation check (see "Known
residual risks" below for the real, version-specific wrinkle that check
found).
**Tantivy (the free-text path) is also built, and — unlike the
ClickHouse pieces — genuinely verified in this environment.**
@@ -75,8 +104,11 @@ sharing the same host port/network-alias trick to stay transparent to
`alerting`/`web` either way — verified via `docker compose config`
(renders and validates the merged YAML without a daemon; confirms
`api`/`enterprise-api` never both appear in `--services` output for the
same profile selection) — see `/docs/phase-4-runbook.md` §10a. This
still only constrains *deployment*, not *operation*: nothing stops an
same profile selection) *and* via an actual `docker compose up` of
`enterprise-api` in this environment, which is what caught the duplicate
`GET /healthz` route panic mentioned above — see
`/docs/phase-4-runbook.md` §10a. This still only constrains
*deployment*, not *operation*: nothing stops an
operator from manually running plain `api`'s image against a cluster
(or compose project) that has tenants
provisioned, pointing at the same ClickHouse/Postgres. The Helm chart
@@ -455,21 +487,27 @@ terms:
credentials. That's an operational control (credential custody,
infrastructure access review), out of scope for this system's own
code.
- **`system.query_log` metadata leakage — per-tenant users are now
real, but the check itself hasn't run yet.** Was an open verification
item because there were no per-tenant ClickHouse users to check
against; that blocker is gone (`enterprise/internal/tenantprovision`
exists), and `tenantprovision_test.go`'s
`TestProvisionedUserCannotReadSystemTables` asserts exactly what the
design calls for (`system.query_log`/`system.tables` inaccessible,
`SHOW DATABASES` not revealing other tenants) — but this environment
never had ClickHouse access to actually run it, so it remains
unconfirmed against the pinned version
(`clickhouse/clickhouse-server:24.8`) until someone with Docker access
runs it (`/docs/phase-4-runbook.md` §8). Also still contingent on the
deployment-shape caveat at the top of this document: even once
confirmed, this only holds when `enterprise-api` (not plain `api`) is
actually serving traffic.
- **`system.query_log` metadata leakage — now closed and verified live**
against the pinned version (`clickhouse/clickhouse-server:24.8`), with
a real, non-obvious wrinkle: a freshly created tenant user was **not**
default-denied from `system.*` the way the original design assumed
(`ProvisionClickHouse` now issues an explicit `REVOKE SELECT ON
system.* FROM <user>` — see that function's doc comment). Confirming
this live also surfaced a genuine ClickHouse behavioral split the
design doc didn't anticipate: `system.query_log` is a real
access-checked table (the REVOKE makes it hard-deny,
`ACCESS_DENIED`), but `system.tables` is a filtered *catalog* view
that ClickHouse 24.8 never denies outright regardless of grants — it
just silently returns zero rows for a properly-revoked user. Both
outcomes close the actual leak (no other tenant's query text or
database/table names are visible either way);
`tenantprovision_test.go`'s `TestProvisionedUserCannotReadSystemTables`
was corrected to assert what each table actually does (hard error for
`query_log`, verified-empty-and-no-foreign-database-names for
`tables`) rather than demanding a hard error from both. Still
contingent on the deployment-shape caveat at the top of this document:
this only holds when `enterprise-api` (not plain `api`) is actually
serving traffic.
- **No deny-override grants** — `dashboard_permissions` is additive-only
by design; a full allow/deny ACL system is unbuilt, future work.
- **No data retention/deletion policy** for a deprovisioned tenant —
@@ -506,20 +544,20 @@ terms:
| Role-based access control on `/query`, `/dashboards` | **Enforced** |
| `alerting``api` service-identity credential | **Enforced** |
| Tenant scoping on dashboards (control-plane data) | **Enforced** |
| ClickHouse per-tenant provisioning (`tenantprovision`) | **Built, not live-verified** — real integration test exists, not yet run against ClickHouse |
| ClickHouse query routing (`chrunner`) | **Built, not live-verified**and only applies when `enterprise-api` serves traffic, not plain `api` |
| `system.*` ClickHouse metadata isolation | **Built, not live-verified**same caveat as above |
| ClickHouse per-tenant provisioning (`tenantprovision`) | **Enforced, verified live** against `clickhouse/clickhouse-server:24.8` — real `CREATE DATABASE`/`CREATE USER`/`GRANT`/`REVOKE`, real two-tenant cross-read probe |
| ClickHouse query routing (`chrunner`) | **Enforced, verified live**real per-tenant connections, cross-tenant raw-SQL probe passes; only applies when `enterprise-api` serves traffic, not plain `api` |
| `system.*` ClickHouse metadata isolation | **Enforced, verified live**`system.query_log` hard-denies, `system.tables` returns zero foreign rows (see "Known residual risks" below for the ClickHouse-version-specific split between the two) |
| Tantivy per-tenant index routing (`search/src/registry.rs`) | **Enforced, verified live** — real Tantivy indices, real cross-tenant probe, all passing |
| Tantivy tenant_id resolution (`enterprise/internal/searchclient`) | **Enforced, verified live** — real gRPC wire-level test |
| Ingest tenant *identity* (credential validation, tagging) | **Built and tested** — fail-closed `TenantResolver`, `tenant_id` Kafka header attached per record |
| Ingest tenant *write-routing*, ClickHouse | **Built, not yet confirmed against a real ClickHouse**`enterprise-ingest`/`chwriter.Registry` route each tagged batch to its tenant's own database, fail-closed on an untagged/unprovisioned tenant; Docker-free tests pass, live-database tests are skip-gated. Active-tenant snapshot now refreshes every minute (`Registry.StartRefreshing`) — a deprovisioned tenant loses write access within a minute, not "until the next restart" |
| Ingest tenant *write-routing*, ClickHouse | **Enforced, verified live**`enterprise-ingest`/`chwriter.Registry` route each tagged batch to its tenant's own database, fail-closed on an untagged/unprovisioned tenant; both Docker-free and live-ClickHouse tests pass. Active-tenant snapshot refreshes every minute (`Registry.StartRefreshing`) — a deprovisioned tenant loses write access within a minute, not "until the next restart" |
| Ingest tenant *write-routing*, Tantivy | **Built and genuinely verified**`search/src/consumer.rs` routes each record into its own tenant's index via `IndexRegistry`, same registry the (already-verified) read side uses; no Docker needed, real tests pass. Active-tenant-gated too: `tenants::ActiveTenantTracker` polls `enterprise-auth` every 60s (off unless configured), refusing any tenant not in the polled allowlist — same one-minute staleness bound as ClickHouse's now-refreshing snapshot, no more asymmetry between the two |
| Deployment actually routing traffic to `enterprise-api` (Helm) | **Enforced**`api`/`enterprise-api` are mutually exclusive, same flag as RBAC/audit/SSO |
| Deployment actually routing traffic to `enterprise-api` (docker-compose) | **Enforced**`api`/`enterprise-api` are mutually exclusive via `COMPOSE_PROFILES`, same flag choice as Helm's `enterprise.enabled`; verified via `docker compose config`, not an actual `docker compose up` in this environment |
| Deployment actually routing traffic to `enterprise-api` (docker-compose) | **Enforced, verified live**`api`/`enterprise-api` are mutually exclusive via `COMPOSE_PROFILES`, same flag choice as Helm's `enterprise.enabled`; a real `docker compose up` of `enterprise-api` was run in this environment (and caught/fixed a startup-crashing duplicate `GET /healthz` route registration bug in the process), not just `docker compose config` |
| Human SSO login — OIDC | **Built, verified with a real fake IdP** (not yet tried against a real external IdP) |
| Human SSO login — SAML | **Built, verified with a real fake IdP** (not yet tried against a real external IdP) |
| Multi-tenant-membership login (tenant picker) | **Backend and frontend built and verified** (`GET /auth/memberships`, `POST /auth/select-tenant`, a pending-login token distinct from a real session; `web/src/routes/select-tenant` calls it via credentialed cross-origin fetch, genuinely exercised in a real browser) — not yet tried against a real running `enterprise-auth` container |
| Per-resource dashboard grants (`own/granted`) | **Built, unit-tested against a fake store; live-Postgres integration tests written, not run in this environment** (only when `enterprise-api` serves traffic — plain `api` falls back to own/Admin only) |
| Multi-tenant-membership login (tenant picker) | **Backend and frontend built and verified** (`GET /auth/memberships`, `POST /auth/select-tenant`, a pending-login token distinct from a real session; `web/src/routes/select-tenant` calls it via credentialed cross-origin fetch, genuinely exercised in a real browser) — not yet tried against a real running `enterprise-auth` container, still blocked on the same "no real external IdP" gap as OIDC/SAML above (there's no way to reach `/select-tenant` with a genuine pending-login cookie from the real binary without one) |
| Per-resource dashboard grants (`own/granted`) | **Enforced, verified live** — real Postgres integration tests for `dashboard_permissions` CRUD and the `PermissionStore` adapter all pass (only when `enterprise-api` serves traffic — plain `api` falls back to own/Admin only) |
| Query audit logging (routine queries) | **Enforced**, fail-open, and now wired to a real writer via `enterprise-api` (`audit.QueryAPILogger`) |
| Audit log tamper detection (hash chain) | **Enforced**, verified live |
| Audit log tamper prevention (external anchoring) | **Design only**`FileSink` is a dev stand-in |