docs(overview): Phase 4 check 1 now passes for both products (lmxopcua#494 closed)

2026-07-24 12:46:05 -04:00
parent 280e57896b
commit c81a6615ba
1 changed files with 4 additions and 2 deletions
@@ -222,10 +222,12 @@
      "OTOPCUA CONTRACT GAP (not a dashboard bug, worth filing): /health/active does not answer 'am I the active node'. On the standby admin (central-2) the admin-leader check returns Degraded ('Role admin member but not leader'), and ASP.NET maps Degraded to 200 \u2014 so a standby looks Active. On driver-only nodes it returns Healthy with the description 'Active for role admin (or not a role member)', i.e. it passes vacuously for every node that is not an admin at all. Meanwhile CLAUDE.md documents the real election as the oldest Up driver member per Cluster (IsDriverPrimary, ServiceLevel 250 vs 240), which this tier never consults. For the tier to mean what MapZbHealth's status mapping implies, a non-active node must return Unhealthy (503) \u2014 which is exactly what ScadaBridge's SitePairActiveNodeHealthCheck does, and why ScadaBridge reads correctly. FILED as lmxopcua#494 (https://gitea.dohertylan.com/dohertj2/lmxopcua/issues/494), with the traefik-dynamic.yml LB-healthcheck impact flagged: docker-dev uses /health/active as its load-balancer probe, so making a standby return 503 would drop it from the UI pool.",
      "OtOpcUa rig rebuilt from a new branch feat/mesh-phase6-health-0.2.0 = feat/mesh-phase6 + a cherry-pick of the 3-line Health 0.2.0 bump (2e515c34). Checking out feat/health-0.2.0-bump directly would have regressed the rig past four commits of in-progress Phase 6/7 work. All three meshes formed cleanly (2 members, 0 unreachable, both members agreeing on one leader) \u2014 OtOpcUa's per-pair self-first seeding did NOT hit the cold-boot self-form race that ScadaBridge did.",
      "TRAEFIK LEADER-PINNING FIXED (ZB.MOM.WW.Health 0.2.1, published + restore-verified; OtOpcUa bumped at 8dd9da7d). Root cause was in the shared library: ActiveNodeDecision.Evaluate returned Degraded for role-member-but-not-leader, and MapZbHealth maps Degraded to 200 \u2014 so /health/active answered 200 on every node and OtOpcUa's Traefik kept the standby in the admin-UI pool. The shared spec had always said the tier 'Fails (503) on a standby or role-member-but-not-leader node', and OtOpcUa's own docs said '200 only on the Admin-role leader; 503 elsewhere'; the docs were right and the code was wrong. Live-verified: central-1 200/traefik UP, central-2 503/traefik DOWN; the README failover drill (which had never exercised anything) flipped Traefik to central-2 within 20s of stopping central-1 with the UI answering 200 continuously across the swap, and central-1 reclaimed the role-leader on restart. Blast radius one consumer \u2014 ScadaBridge uses its own OldestNodeActiveHealthCheck, mxgw/HG have no Akka.",
-      "RIG OPERATIONS LANDMINE: the docker-dev migrator (dotnet ef database update) HANGS under x86 emulation \u2014 22 minutes at 0.00% CPU with no output, and all six node containers sit in Created because they depend_on service_completed_successfully. Worked around by stopping the migrator and docker-start-ing the nodes directly, which is safe when the change is a package-version bump with no EF model change. Worth knowing before assuming a rebuild is merely slow."
+      "RIG OPERATIONS LANDMINE: the docker-dev migrator (dotnet ef database update) HANGS under x86 emulation \u2014 22 minutes at 0.00% CPU with no output, and all six node containers sit in Created because they depend_on service_completed_successfully. Worked around by stopping the migrator and docker-start-ing the nodes directly, which is safe when the change is a package-version bump with no EF model change. Worth knowing before assuming a rebuild is merely slow.",
      "lmxopcua#494 CLOSED. The site-node half needed a local ClusterPrimaryHealthCheck: this node is active iff it is the OLDEST Up member carrying its own active role (admin when present, else driver). IMPORTANT SELF-CORRECTION \u2014 the Health 0.2.1 fix alone was NOT sufficient. It made the tier a real 503 but still selected by RoleLeader (lowest address), and on the rebuilt rig the orderings diverge live: akka leader = central-1, oldest admin member = central-2 (which hosts the singletons). 0.2.1 alone would have pinned Traefik to central-1, the node NOT hosting the work \u2014 the exact failure RedundancyStateActor.SelectDriverPrimary was written to avoid, reappearing in the health tier. Only the live rig exposed it. Follow-up not filed: two of four family apps now need 'oldest Up member of role X' and both had to avoid the shared ActiveNodeHealthCheck to get it (ScadaBridge's OldestNodeActiveHealthCheck, OtOpcUa's ClusterPrimaryHealthCheck) \u2014 a candidate for promotion into ZB.MOM.WW.Health.Akka later.",
      "RIG BUILD LANDMINE: all six docker-dev node services share ONE image (otopcua-host:dev), so `docker compose build` with six service names builds the same image six times in parallel. Under x86 emulation that thrashed for 46+ minutes on `dotnet restore` with no output and looked wedged. Building a single service took 29 SECONDS for the same restore. Build one service, then recreate all six containers."
    ],
    "phase4_results": {
-      "1_all_render_single_active_per_pair": "PARTIAL, and now partial for a smaller reason. ScadaBridge PASS (all four pairs: one Active + one Standby). OtOpcUa central pair PASS after ZB.MOM.WW.Health 0.2.1 \u2014 central-1 Active / central-2 Standby, live-verified. OtOpcUa's four DRIVER-ONLY site nodes still all read Active: the admin-leader check is scoped to the admin role and returns Healthy for any node lacking it, so it never consults the real per-Cluster Primary (IsDriverPrimary). Tracked in lmxopcua#494, Traefik half closed.",
+      "1_all_render_single_active_per_pair": "PASS for both products. ScadaBridge: all four pairs one Active + one Standby. OtOpcUa: after lmxopcua#494 was fully closed (ClusterPrimaryHealthCheck, commit 45504861), exactly one 200 per mesh \u2014 MAIN central-2, SITE-A site-a-1, SITE-B site-b-1 \u2014 with the partner 503. Dashboard renders one Active + one Standby for all seven cluster groups.",
      "2_leader_chip_and_split_brain": "PASS for BOTH products after the rig redeploy and the LeaderResolver fix. All seven cluster groups (ScadaBridge central/site-a/b/c, OtOpcUa MAIN/SITE-A/SITE-B) render a Leader chip with zero split-brain flags. Verified earlier that the chip MOVES when the leader is stopped and that the split-brain flag appears while a pair is self-formed and clears once it genuinely joins.",
      "3_kill_and_restart": "PASS. Stopped scadabridge-site-c-a 12:01:27; Unreachable by 12:02:33 (two-strike damping = 2 failing sweeps). Restarted 12:04:01; Up again by 12:04:16, i.e. on the first good poll (recovery is deliberately asymmetric)",
      "4_degraded_with_failing_entry": "PASS. Canned Degraded and Unhealthy endpoints render as Degraded/Down with the failing entry and its description visible, and check-dot ok/warn/bad classes per check",