feat(secrets): opt-in SQL-Server hub replication for the host secret store
Routes both host-container secret registrations (central role in Program.cs,
site role in SiteServiceRegistration) through a new SecretsRegistration
composition seam that optionally enables ZB.MOM.WW.Secrets.Replicator.SqlServer
hub-replication mode: each node keeps a LOCAL store that syncs bidirectionally
with a shared central SQL hub, so a site cluster keeps resolving secrets
straight through a WAN outage to central.
OPT-IN GATE (the load-bearing part). AddZbSecretsSqlServerReplication validates
its options EAGERLY at registration time, so wiring it unconditionally would
make ScadaBridge fail to START anywhere Secrets:SqlServer:ConnectionString is
unset -- every dev box, every docker node, every existing deployment. The
SQL-Server package is therefore only touched when BOTH Secrets:Replication:
Enabled is true AND a non-blank connection string is present; otherwise the
registration is byte-identical to the previous plain AddZbSecrets call.
Enabled-without-a-connection-string falls back to local-only and logs a warning
rather than failing the node or silently looking healthy.
BOOTSTRAP CYCLE. The hub connection string is itself a secret and can never come
from the hub -- a node cannot read the hub to learn how to reach the hub. It must
arrive from outside the replicated set: an environment variable, or a ${secret:}
reference seeded in that node's own LOCAL store. appsettings.json therefore ships
ConnectionString empty with a _comment saying so (leaf keys starting with '_' are
skipped by the reference expander, verified in SecretReferenceExpander). No real
connection string is committed. The pre-host ${secret:} expander in Program.cs is
deliberately left on a plain local SQLite store for the same reason.
Per-node docker appsettings are intentionally NOT modified -- replication stays
off there for now.
Tests written before the wiring and confirmed red first (2 failed / 4 passed),
green after (6/6). They BUILD a container and RESOLVE from it rather than
asserting over ServiceDescriptors: a decorator can look correct as a descriptor
list and still throw on first resolve because the undecorated concrete store it
depends on is missing -- that exact defect shipped once in this library with all
descriptor-level tests green, so the undecorated SqliteSecretStore gets its own
resolution test.
Verified: Host.Tests 285/285 pass; full build (all projects except the
pre-existing AngleSharp NU1902 CentralUI.Tests restore break) 0 warnings,
0 errors.
Claude-Session: https://claude.ai/code/session_01BL2Vu1ESDQ9SCN4gVKkdts
This commit is contained in:
@@ -50,7 +50,11 @@ public static class SiteServiceRegistration
|
||||
var siteDbPath = config["ScadaBridge:Database:SiteDbPath"] ?? "site.db";
|
||||
services.AddSiteRuntime($"Data Source={siteDbPath}");
|
||||
services.AddDataConnectionLayer();
|
||||
services.AddZbSecrets(config, "Secrets");
|
||||
// Local SQLite store by default; a local store replicating against a shared SQL-Server hub
|
||||
// only when Secrets:Replication:Enabled is true AND a hub connection string is present.
|
||||
// Hub mode matters most here: a site node keeps resolving secrets from its local store
|
||||
// straight through a WAN outage to central. See SecretsRegistration.
|
||||
services.AddScadaBridgeSecrets(config);
|
||||
// Adapter that surfaces the site id to
|
||||
// StoreAndForwardService through DI WITHOUT introducing a
|
||||
// StoreAndForward → HealthMonitoring project-reference cycle. Must be
|
||||
|
||||
Reference in New Issue
Block a user