docs(index): OtOpcUa master e857135e — Galaxy subscribe-reply race + reconnect fencing (#859/#861), #856–#862 batch

2026-11-11 11:29:07 -05:00
parent 2adcd75bdc
commit d8147bd0fd
2 changed files with 5 additions and 1 deletions
@@ -437,3 +437,7 @@ AddRoute of the dev Mac (`oee-tagdump-mac`, `10.181.206.245.1.1`) on the two tes
  - **ScadaBridge impact (#854):** ConditionRefresh/ConditionRefresh2 on Server-object subscriptions is no longer followed by older deferred condition events during event-queue overload; normal refresh is unchanged (off `Lock`). Folder subscriptions keep the residual (#855).
  - **MTConnect (#852):** `CommsLossConfirmMs` editable in AdminUI; native conditions also go quality-only BadNoCommunication on a refused connection. **AdminUI (#851):** driver Controls panel shows per-node health.
  - **Live gate 2026-11-11 (docker-dev, `03e51457`): PASS.** Both centrals boot clean (credential preflight on SQL Server); `drv`, another driver's `env:` variable, renaming a driver off its own reference and claiming another's reference name all refused in /raw (#848/#849); absolute OpcUaClient `UserCertificatePath` refused (#836); MTConnect refused connection ⇒ values + condition quality Bad in 2.0 s, recovery without re-raise (#852, closed); /hosts Controls per-node chips + per-node STALE (#851; follow-up #857: not stale for the first 30 s after opening); lockout locks one key id after 5 failures, other key ids still evaluated, anonymous unaffected (#844/#845); ConditionRefresh on `i=2253` frames correctly (#854). Not exercised live: #854 under overload (wire-tested), #850 split-config banner.
+- **2026-11-11 follow-up batch #856–#862 (master `e857135e`, not deployed).** Pushed; #856–#859, #861, #862, #821 closed; #818/#860 wait on a hosted CI run, #838 item 1 stays a watch item.
+  - **Galaxy (#859/#861) — ScadaBridge-visible fix:** the first value after a subscribe could arrive on the event stream before the SubscribeBulk reply and was dropped, leaving the node on its pre-subscribe value until the next change (subscribe-once reads timed out). Early events are now parked and replayed in order; the subscription registry is fenced per gateway session so a reconnect can never route a reused MXAccess handle number to the wrong tag. Four independent concurrency reviews. **#862:** with `ReplayOnSessionLost=false`, a reconnect now marks stranded tags `BadNoCommunication` and the driver Degraded instead of serving last Good values.
+  - **AdminUI (#857):** the Controls dialog marks a node stale immediately when its snapshot is already old at open. **CI (#860):** the v2-integration fixture script finds the docker host without `ip` (job image lacks it). **Tests:** Galaxy test waits hardened (#856/#858), >4096-event overflow under ACL enforcement wire-tested with no defect (#838).
+  - **Live check 2026-11-11 (docker-dev):** #857 PASS (central-2 stopped 1 m 50 s ⇒ dimmed + STALE on open, central-1 live; both live after restart). Galaxy changes are unit/fake-gateway tested only — no mxaccessgw reachable from docker-dev.