Audit per-node central caches for peer-node staleness (InvalidateMethod, ManagementActor cache) — same class as #83 #85

New Issue

2026-11-09T12:43:21-05:00

dohertj2 commented

2026-11-09 12:43:21 -05:00

Same class as #83: several central caches invalidate only on the local node. Inbound-method handler invalidation (InvalidateMethod, incl. via the bundle-import change bus) and the cache cleared at ManagementActor.cs:~2898 act locally; the peer central node may keep serving stale entries until TTL or restart. Audit every per-node central cache, record each one's staleness bound, and decide which need a cluster-wide invalidation message (could share one mechanism with #83's peer half).

Same class as #83: several central caches invalidate only on the local node. Inbound-method handler invalidation (`InvalidateMethod`, incl. via the bundle-import change bus) and the cache cleared at `ManagementActor.cs:~2898` act locally; the peer central node may keep serving stale entries until TTL or restart. Audit every per-node central cache, record each one's staleness bound, and decide which need a cluster-wide invalidation message (could share one mechanism with #83's peer half).

dohertj2 added the type:followup priority:p3 labels 2026-11-09 12:43:21 -05:00

dohertj2 commented

2026-11-09 14:02:08 -05:00

Done on main a16b7b4e: audit doc docs/plans/2026-11-09-central-cache-staleness-audit.md lists every central cache with its peer staleness bound — all peer-stale caches expire on 5–60 s timers, and Traefik routes only to the active node, so the window is hit only by direct-node calls or a failover inside it; none clearly a bug, so no cross-node invalidation added (the doc names the existing central DistributedPubSub topic if wanted). Fix (local node): management API create/update/delete of inbound API methods cleared the compiled script but not the cached method row — a deleted method stayed callable / a cached not-found hid a new one for up to 5 s; ManagementActor now evicts the row after commit (ApiMethodCacheInvalidationTests: 3 fail without the fix + a control). Owner decisions listed in the doc: role mappings (revoked role works on the peer ≤30 s), inbound PSKs (filed separately), external definitions (#83). Evidence: slnx 0 warnings/0 errors after rebase on #92/#93; DeploymentManager, ManagementService, CentralUI, InboundAPI, Host, Commons, Transport all passed.

Done on main a16b7b4e: audit doc docs/plans/2026-11-09-central-cache-staleness-audit.md lists every central cache with its peer staleness bound — all peer-stale caches expire on 5–60 s timers, and Traefik routes only to the active node, so the window is hit only by direct-node calls or a failover inside it; none clearly a bug, so no cross-node invalidation added (the doc names the existing central DistributedPubSub topic if wanted). Fix (local node): management API create/update/delete of inbound API methods cleared the compiled script but not the cached method row — a deleted method stayed callable / a cached not-found hid a new one for up to 5 s; ManagementActor now evicts the row after commit (ApiMethodCacheInvalidationTests: 3 fail without the fix + a control). Owner decisions listed in the doc: role mappings (revoked role works on the peer ≤30 s), inbound PSKs (filed separately), external definitions (#83). Evidence: slnx 0 warnings/0 errors after rebase on #92/#93; DeploymentManager, ManagementService, CentralUI, InboundAPI, Host, Commons, Transport all passed.

dohertj2 closed this issue

2026-11-09 14:02:08 -05:00

dohertj2 referenced this issue

2026-11-09 14:02:08 -05:00

SitePskProvider caches site PSKs forever: in-place PSK rotation breaks the central↔site link until restart #96

dohertj2 referenced this issue

2026-11-09 14:38:53 -05:00

Import: GetIdByNameAsync derived-template ambiguity; intra-record exact-name lookups; reuse the PSK pub-sub invalidation for D1/D3 #100

dohertj2 referenced this issue from a commit

2026-11-09 14:38:55 -05:00

fix(host): follow an in-place site PSK rotation without a central restart (ScadaBridge#96)

dohertj2 referenced this issue from a commit