fix(central): audit per-node central caches; management API drops ApiMethodCache entries (ScadaBridge#85)

Audit (docs/plans/2026-11-09-central-cache-staleness-audit.md) of every central in-process
cache holding configuration or secret state, with its staleness bound on the peer central
node. Every config cache with peer staleness is TTL-bounded (5-60 s) and Traefik routes to the
active node only, so no peer case was clearly a bug and no cross-node broadcast is added; the
doc names the existing DistributedPubSub mediator (health replication topic) as the mechanism
if one is wanted. Owner decisions listed: RoleMappingCache 30 s revocation lag on the peer,
SitePskProvider caching a PSK forever across an in-place rotation (both nodes), and the #83
ExternalDefinitionLookupCache 10 s bound.

Fixed: ManagementActor API-method create/update/delete invalidated only the compiled handler,
not ApiMethodCache, so the node that made the change kept serving the pre-change row for up to
the 5 s TTL (a deleted method stayed callable; a cached not-found hid a new one). The handlers
now drop the entry post-commit, like the role-mapping cache.

Tests: ApiMethodCacheInvalidationTests (create/update/delete fail without the fix; a
negative control checks other methods stay cached).

(ScadaBridge#85)

Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV

docs/plans/2026-11-09-central-cache-staleness-audit.md

@@ -0,0 +1,75 @@
+# Central per-node cache staleness audit (Gitea #85)
+
+**Date:** 2026-11-09. **Scope:** every in-process cache on a **central** node that holds
+configuration-database (or secret-store) state, and how stale it can be on the **peer** central
+node after a change is made through the other one. Same defect class as #83
+(`ExternalDefinitionLookupCache`).
+
+## Context that sets the risk
+
+- Traefik routes UI, `/management` and `/api/*` to the **active** central node only
+  (`/health/active`). An edit and a use on different nodes therefore only happen when someone
+  talks to a node directly (CLI pointed at a node's own port) or across a **failover**.
+- Every TTL below is measured from when the entry was loaded, so an entry a node cached while it
+  was previously active is still bounded by its TTL after it becomes active again.
+- No cache below replicates its invalidation to the peer. Central already has one cross-node
+  channel: the `DistributedPubSub` topic `CentralCommunicationActor.HealthReportTopic` (peer
+  health replication). A config-invalidation broadcast, if an owner wants one, should be a second
+  topic on the same mediator, published from `AfterConfigurationCommit`.
+
+## Inventory
+
+| Cache | Holds | Local invalidation | Peer staleness bound | Verdict |
+|---|---|---|---|---|
+| `InboundScriptExecutor` compiled handlers (InboundAPI) | Compiled method delegates | `InvalidateMethod` on delete; change bus on import | **None** — every request revision-checks the cached handler against the resolved `ApiMethod.Script` and recompiles on mismatch | OK |
+| `ApiMethodCache` (InboundAPI) | `ApiMethod` rows, incl. not-found | Change bus (bundle import); **management create/update/delete — added by #85** | `InboundApi:ApiMethodCacheTtl`, default **5 s** | OK (local gap fixed, see below) |
+| `ExternalDefinitionLookupCache` (ESG, central only) | External systems / methods / DB connections (found only) | 9 ManagementActor handlers (#65); bundle import (#83) | `ExternalSystemGateway:DefinitionCacheTtl`, **10 s** on central | Owner decision — #83 |
+| `RoleMappingCache` (Security) — the cache at `ManagementActor` `InvalidateRoleMappingCacheAfterCommit` | LDAP group → role map + Deployer site scopes | Mapping / scope-rule commands, post-commit | `Security:RoleMappingCacheSeconds`, default **30 s** (`0` disables) | Owner decision — see D1 |
+| `PendingUpdatesCache` (DeploymentManager, #91) | Fleet pending-updates snapshot | Deployment finished on this node | `DeploymentManager:PendingUpdatesCacheTtl`, default **60 s** | OK — display only |
+| `CentralCommunicationActor` site-address cache (+ `GrpcSiteTransport` channel pairs) | Site gRPC endpoints | Site create/update/delete → `RefreshSiteAddresses` | Periodic DB reload, **60 s** (hard-coded) | OK — rare edit, self-heals |
+| `SiteStreamGrpcClientFactory` clients | Per (site, endpoint) stream clients | Site delete → `RemoveSiteAsync` (local only) | Not stale data: a removed site's client on the peer is reclaimed by idle eviction, `Communication:GrpcClientIdleTtl` (**30 min**) × 1.5 after last use | OK — resource, not correctness |
+| `SitePskProvider` (Host) | Per-site gRPC PSK from config / secret store | Site removal (via either gRPC client path, on every node within the 60 s reload) | **Unbounded** for an in-place key rotation — on **both** nodes, not only the peer | Owner decision — see D2 |
+| `DefaultSecretResolver` (shared `ZB.MOM.WW.Secrets`) | Decrypted secret values | Lib's own invalidator (UI writes, sync service) | `Secrets:ResolveCacheTtl`, **30 s** | OK — owned by the shared lib |
+| `KpiSnapshotCache` (CentralUI) | KPI query results (operational data) | — | **8 s** | OK — not config |
+| `SharedAlarmSummaryService` (CentralUI) | Per-site alarm summary fan-out | — | **12 s** cold-cache window | OK — not config |
+| `IDeploymentStatusNotifier` (DeploymentManager) | Push of deployment status to UI circuits | In-process event | Peer UI never sees the push, but the UI is only served by the active node and the pages also poll | OK |
+| `ITemplateGraphWatermark` / `FlattenSession` | Template-chain memo keyed by in-process version | Repository bumps | Session lives for one deploy/validate operation; the watermark is process-local by design (peer starts from 0 → extra work, never stale work) | OK |
+| `ScriptCompileVerdictCache`, `TrustVerdictCache`, `AuditRegexCache`, `ScriptAnalysisService` `IMemoryCache`, `OAuth2TokenService` | Pure functions of their key (code hash + surface, pattern, credential hash) | — | Not applicable — a changed input is a different key | OK |
+| `ZB.MOM.WW.Auth.ApiKeys.SqlServer` (inbound API keys) | — | — | No in-process cache in the shared store (checked in the 0.3.0 source) — revocation is immediate on both nodes | OK |
+
+## Fixed in #85
+
+**`ApiMethodCache` was not invalidated by the management API even locally.** Only the compiled
+handler was dropped, so for up to the 5 s TTL on the node that made the change:
+
+- a just-**deleted** method stayed callable — the executor recompiled it from the cached row;
+- a just-**created** method answered 404 if a not-found had been cached.
+
+`ManagementActor` now drops the method's cache entry post-commit on create, update and delete
+(`InvalidateApiMethodCacheAfterCommit`), the same post-commit pattern as the role-mapping cache.
+The peer is still bounded by the 5 s TTL. Tests: `ApiMethodCacheInvalidationTests` (3 of 4 fail
+without the fix; the fourth is the negative control).
+
+No peer-staleness case was **clearly** a bug: every config cache with peer staleness is
+TTL-bounded (5–60 s), its bound is documented at the cache, and the active-only routing means the
+window is only hit via a direct-to-node call or a failover inside the TTL. So no cross-node
+broadcast was added.
+
+## Needs an owner decision
+
+- **D1 — `RoleMappingCache` revocation lag on the peer (30 s).** Removing a mapping or narrowing a
+  Deployer's site scope takes effect at once on the node that made the change and within 30 s on
+  the other. It is a security-relevant window (revoked access keeps working briefly), but only on
+  the node the change did *not* go through. Options: accept and document (current state); lower
+  `RoleMappingCacheSeconds`; or add the `DistributedPubSub` invalidation topic described above.
+  Recommendation: accept unless the operators' revocation SLA is tighter than 30 s.
+- **D2 — `SitePskProvider` caches a successful PSK resolve forever.** Its own remarks say
+  `Invalidate` covers "a key rotated in place", but nothing calls it on rotation — only on site
+  removal. Rotating `SB-GRPC-PSK-{site}` in the secret store (with the site pair restarted on the
+  new key) leaves **both** central nodes dialling with the old key until they restart. Options: make
+  central restart part of the rotation runbook; give the cache a TTL; or invalidate on a
+  `PermissionDenied` from the site. Not peer-specific, so it is out of #85's fix scope — file it
+  separately.
+- **D3 — `ExternalDefinitionLookupCache` peer bound (10 s).** Already tracked by #83; the
+  recommendation there (accept and document) still stands. If D1 gets a broadcast topic, this cache
+  and `ApiMethodCache` should join it rather than add a second mechanism.

docs/requirements/Component-InboundAPI.md

@@ -175,7 +175,7 @@ The `sbk_<keyId>_<secret>` design already supports zero-downtime rotation — a
 
 ## Script Compilation & Hot-Reload
 
-API method scripts are pre-compiled in the **background after startup** by `InboundMethodWarmupService` (a hosted service registered after `AkkaHostedService`, so it never delays the cluster join or `/health/*`): it waits for `ApplicationStarted`, then — only once this node is the active central node (`IActiveNodeGate`; a standby pays nothing until a failover promotes it) — loads all method definitions and compiles them into in-memory delegates. Lazy compile + the per-request revision check guarantee correctness regardless. Compiles are **single-flight per method name + script revision**: N concurrent first callers share one compile (and one non-collectible loaded assembly). `InboundScriptExecutor.CompileAndRegister(ApiMethod, out IReadOnlyList<string> errors)` compiles once and returns the diagnostics, so a save needs no separate validate pass; the management Create/Update path uses its sibling `CompileForPublish(method, out errors, out publish)`, which compiles once for the `compileWarning` and hands back a `publish` action queued as the post-commit registration — one compile (one pinned assembly) per save even inside a configuration transaction (review 2026-10-21 05/M1). `ApiMethodCache` prunes expired entries on a miss once large (scopes are exact method names — pinned by test — so growth is bounded).
+API method scripts are pre-compiled in the **background after startup** by `InboundMethodWarmupService` (a hosted service registered after `AkkaHostedService`, so it never delays the cluster join or `/health/*`): it waits for `ApplicationStarted`, then — only once this node is the active central node (`IActiveNodeGate`; a standby pays nothing until a failover promotes it) — loads all method definitions and compiles them into in-memory delegates. Lazy compile + the per-request revision check guarantee correctness regardless. Compiles are **single-flight per method name + script revision**: N concurrent first callers share one compile (and one non-collectible loaded assembly). `InboundScriptExecutor.CompileAndRegister(ApiMethod, out IReadOnlyList<string> errors)` compiles once and returns the diagnostics, so a save needs no separate validate pass; the management Create/Update path uses its sibling `CompileForPublish(method, out errors, out publish)`, which compiles once for the `compileWarning` and hands back a `publish` action queued as the post-commit registration — one compile (one pinned assembly) per save even inside a configuration transaction (review 2026-10-21 05/M1). `ApiMethodCache` prunes expired entries on a miss once large (scopes are exact method names — pinned by test — so growth is bounded). A Management API/UI create, update or delete drops that method's `ApiMethodCache` entry post-commit on the node that made the change; the peer central node converges within `ApiMethodCacheTtl` (default 5 s) — Gitea #85, audit `docs/plans/2026-11-09-central-cache-staleness-audit.md`.
 
 ### Update Workflow
 

src/ZB.MOM.WW.ScadaBridge.InboundAPI/ApiMethodCache.cs

@@ -14,13 +14,15 @@ namespace ZB.MOM.WW.ScadaBridge.InboundAPI;
 /// <list type="bullet">
 /// <item><description><b>Fast path.</b> <see cref="ScriptArtifactChangeSubscriber"/> calls
 /// <see cref="Invalidate"/> for every name in an <c>ApiMethod</c>
-/// <c>ScriptArtifactsChanged</c> notification (bundle-import changes today).</description></item>
+/// <c>ScriptArtifactsChanged</c> notification (bundle-import changes), and the
+/// <c>ManagementActor</c> API-method create/update/delete handlers drop the changed name
+/// post-commit (Gitea #85). Both are node-local.</description></item>
 /// <item><description><b>Fallback.</b> Every entry — including a negative (not-found) result —
 /// expires after <see cref="InboundApiOptions.ApiMethodCacheTtl"/> regardless of whether an
-/// invalidation notification ever arrives. This is what keeps an operator edit via the
+/// invalidation notification ever arrives. This is what bounds staleness on the PEER central
-/// Management API/UI (which does not publish to the change bus) correct within a bounded
+/// node (no invalidation crosses nodes), and is also the correctness backstop for the bus's
-/// window, and is also the correctness backstop for the bus's at-least-once/in-process-only
+/// at-least-once/in-process-only contract — a missed or duplicate notification is
-/// contract — a missed or duplicate notification is harmless.</description></item>
+/// harmless.</description></item>
 /// </list>
 /// <para>
 /// A negative (method not found) result is cached too — bounded by the same short TTL — so a

src/ZB.MOM.WW.ScadaBridge.InboundAPI/InboundApiOptions.cs

@@ -41,9 +41,9 @@ public class InboundApiOptions
     /// before the next request re-fetches it from the repository. The
     /// <c>ScriptArtifactChangeSubscriber</c>/<c>IScriptArtifactChangeBus</c> pipeline
     /// invalidates a specific method's entry immediately on a bundle-import change (the fast
-    /// path); this TTL is the self-healing fallback for changes the bus does not cover (e.g.
+    /// path), and a Management API/UI create/update/delete drops it post-commit on the node that
-    /// an operator edit via the Management API/UI, which does not publish a bus notification),
+    /// made the change (Gitea #85). Both are node-local, so this TTL is the bound on the PEER
-    /// bounding staleness without requiring every mutation path to remember to invalidate.
+    /// central node (and the self-healing fallback for any path that forgets to invalidate).
     /// Short by design — the whole point is to absorb repeated-request-storm SQL, not to
     /// serve minutes-old method definitions.
     /// </summary>

src/ZB.MOM.WW.ScadaBridge.ManagementService/ManagementActor.cs

@@ -3811,6 +3811,7 @@ public class ManagementActor : ReceiveActor
         };
         await repo.AddApiMethodAsync(method);
         await repo.SaveChangesAsync();
+        await InvalidateApiMethodCacheAfterCommit(sp, method.Name);
         // The script is persisted regardless; surface a non-fatal warning when it does
         // not compile so the caller is not misled into thinking a broken method is live
         // (a never-compiled method returns "Script compilation failed" on first call).
@@ -3832,6 +3833,7 @@ public class ManagementActor : ReceiveActor
             method.TraceAllCalls = traceAllCalls;
         await repo.UpdateApiMethodAsync(method);
         await repo.SaveChangesAsync();
+        await InvalidateApiMethodCacheAfterCommit(sp, method.Name);
         // The persisted DB row is authoritative: on a compile failure the method returns
         // a compilation error on every request (ExecuteAsync revision-checks the cached
         // handler against the fresh row), so the saved-but-broken script does NOT keep a
@@ -3849,11 +3851,32 @@ public class ManagementActor : ReceiveActor
         await repo.DeleteApiMethodAsync(cmd.ApiMethodId);
         await repo.SaveChangesAsync();
         if (method != null)
+        {
             await AfterConfigurationCommit(sp, () => { sp.GetService<InboundAPI.InboundScriptExecutor>()?.InvalidateMethod(method.Name); return Task.CompletedTask; });
+            await InvalidateApiMethodCacheAfterCommit(sp, method.Name);
+        }
         await AuditAsync(sp, user, "Delete", "ApiMethod", cmd.ApiMethodId.ToString(), method?.Name ?? cmd.ApiMethodId.ToString(), null);
         return true;
     }
 
+    /// <summary>
+    /// Drops this node's <see cref="InboundAPI.ApiMethodCache"/> entry for <paramref name="name"/>
+    /// once an API-method create/update/delete has committed (Gitea #85). Without it the
+    /// node that made the change kept serving the pre-change row for up to
+    /// <c>ApiMethodCacheTtl</c>: a just-deleted method stayed callable (the executor recompiled
+    /// it from the cached row), and a cached not-found answered 404 for a just-created one.
+    /// Queued post-commit, like the role-mapping invalidation, so a concurrent request cannot
+    /// re-cache the pre-change row. The peer central node still converges within the TTL
+    /// (default 5 s) — see <c>docs/plans/2026-11-09-central-cache-staleness-audit.md</c>.
+    /// </summary>
+    private static Task InvalidateApiMethodCacheAfterCommit(IServiceProvider sp, string name)
+    {
+        var cache = sp.GetService<InboundAPI.ApiMethodCache>();
+        return cache is null
+            ? Task.CompletedTask
+            : AfterConfigurationCommit(sp, () => { cache.Invalidate(name); return Task.CompletedTask; });
+    }
+
     /// <summary>
     /// Recompiles and re-registers an inbound API method's script after a Create/Update,
     /// returning a non-null human-readable warning when the script did NOT compile. The

tests/ZB.MOM.WW.ScadaBridge.ManagementService.Tests/ApiMethodCacheInvalidationTests.cs

@@ -0,0 +1,114 @@
+using Akka.Actor;
+using Akka.TestKit.Xunit2;
+using Microsoft.Extensions.DependencyInjection;
+using Microsoft.Extensions.Logging.Abstractions;
+using NSubstitute;
+using ZB.MOM.WW.ScadaBridge.Commons.Entities.InboundApi;
+using ZB.MOM.WW.ScadaBridge.Commons.Interfaces.Repositories;
+using ZB.MOM.WW.ScadaBridge.Commons.Interfaces.Services;
+using ZB.MOM.WW.ScadaBridge.Commons.Messages.Management;
+using ZB.MOM.WW.ScadaBridge.InboundAPI;
+
+namespace ZB.MOM.WW.ScadaBridge.ManagementService.Tests;
+
+/// <summary>
+/// Gitea #85: an API-method create / update / delete through the management API must drop this
+/// node's <see cref="ApiMethodCache"/> entry once the change commits. Before, only the compiled
+/// handler was invalidated, so the node kept resolving the pre-change row for up to the cache
+/// TTL — a deleted method stayed callable and a cached not-found hid a newly created one.
+/// </summary>
+public class ApiMethodCacheInvalidationTests : TestKit
+{
+    private readonly ServiceCollection _services = new ServiceCollection().AddTestConfigurationTransaction();
+    private readonly IInboundApiRepository _apiRepo = Substitute.For<IInboundApiRepository>();
+
+    // A TTL far longer than the test, so only an explicit invalidation can clear an entry.
+    private readonly ApiMethodCache _cache = new(TimeSpan.FromHours(1));
+
+    /// <summary>Initializes the harness with a long-TTL method cache.</summary>
+    public ApiMethodCacheInvalidationTests()
+    {
+        _services.AddScoped(_ => _apiRepo);
+        _services.AddScoped(_ => Substitute.For<IAuditService>());
+        _services.AddSingleton(_cache);
+        _apiRepo.SaveChangesAsync(Arg.Any<CancellationToken>()).Returns(1);
+    }
+
+    private IActorRef CreateActor() =>
+        Sys.ActorOf(Props.Create(() => new ManagementActor(
+            _services.BuildServiceProvider(), NullLogger<ManagementActor>.Instance)));
+
+    private static ManagementEnvelope Envelope(object command) =>
+        new(new AuthenticatedUser("designer", "Designer", new[] { "Designer" }, Array.Empty<string>()),
+            command, Guid.NewGuid().ToString("N"));
+
+    private Task Seed(string name, ApiMethod? row) =>
+        _cache.GetOrFetchAsync(name, _ => Task.FromResult(row), CancellationToken.None);
+
+    /// <summary>Returns true when the cache re-fetches <paramref name="name"/> (the entry was dropped).</summary>
+    private async Task<bool> Refetches(string name)
+    {
+        var fetched = false;
+        await _cache.GetOrFetchAsync(name, _ => { fetched = true; return Task.FromResult<ApiMethod?>(null); }, CancellationToken.None);
+        return fetched;
+    }
+
+    /// <summary>A cached not-found must not hide a method created through the management API.</summary>
+    /// <returns>A task that represents the asynchronous operation.</returns>
+    [Fact]
+    public async Task Create_DropsCachedNotFound()
+    {
+        await Seed("Fresh", null);
+        Assert.False(await Refetches("Fresh")); // positive control: the negative entry is served
+
+        CreateActor().Tell(Envelope(new CreateApiMethodCommand("Fresh", "return 1;", 30, null, null)));
+        ExpectMsg<ManagementSuccess>(TimeSpan.FromSeconds(30));
+
+        Assert.True(await Refetches("Fresh"));
+    }
+
+    /// <summary>An update drops the cached pre-change row.</summary>
+    /// <returns>A task that represents the asynchronous operation.</returns>
+    [Fact]
+    public async Task Update_DropsCachedRow()
+    {
+        var existing = new ApiMethod("Edited", "return 1;") { Id = 7, TimeoutSeconds = 10 };
+        _apiRepo.GetApiMethodByIdAsync(7, Arg.Any<CancellationToken>()).Returns(existing);
+        await Seed("Edited", existing);
+
+        CreateActor().Tell(Envelope(new UpdateApiMethodCommand(7, "return 2;", 10, null, null)));
+        ExpectMsg<ManagementSuccess>(TimeSpan.FromSeconds(30));
+
+        Assert.True(await Refetches("Edited"));
+    }
+
+    /// <summary>A delete drops the cached row so the deleted method is no longer resolved.</summary>
+    /// <returns>A task that represents the asynchronous operation.</returns>
+    [Fact]
+    public async Task Delete_DropsCachedRow()
+    {
+        var existing = new ApiMethod("Gone", "return 1;") { Id = 9, TimeoutSeconds = 10 };
+        _apiRepo.GetApiMethodByIdAsync(9, Arg.Any<CancellationToken>()).Returns(existing);
+        await Seed("Gone", existing);
+
+        CreateActor().Tell(Envelope(new DeleteApiMethodCommand(9)));
+        ExpectMsg<ManagementSuccess>(TimeSpan.FromSeconds(30));
+
+        Assert.True(await Refetches("Gone"));
+    }
+
+    /// <summary>A change to one method leaves other methods' entries cached.</summary>
+    /// <returns>A task that represents the asynchronous operation.</returns>
+    [Fact]
+    public async Task Delete_LeavesOtherMethodsCached()
+    {
+        var existing = new ApiMethod("Gone2", "return 1;") { Id = 10, TimeoutSeconds = 10 };
+        _apiRepo.GetApiMethodByIdAsync(10, Arg.Any<CancellationToken>()).Returns(existing);
+        await Seed("Other", new ApiMethod("Other", "return 3;"));
+
+        CreateActor().Tell(Envelope(new DeleteApiMethodCommand(10)));
+        ExpectMsg<ManagementSuccess>(TimeSpan.FromSeconds(30));
+
+        Assert.False(await Refetches("Other"));
+    }
+}