Corrects the #459 finding. 2-node keep-oldest recovery works fine (the ScadaBridge
sister project proves it); OtOpcUa was missing the supervision pieces that make it
automatic, and docs/Redundancy.md wrongly claimed in-place oldest-crash failover.
Mechanism (confirmed on a 2-container rig + by decompiling Akka KeepOldest.OldestDecision):
on an OLDEST-node crash keep-oldest downs the LONE survivor (DownReachable including
myself) — down-if-alone can't rescue a lone survivor (its branch needs >=2 survivors).
Recovery is exit-and-rejoin: run-coordinated-shutdown-when-down terminates the node and
the service supervisor restarts it. My earlier 'total outage' was a docker-dev artifact
(no restart policy); production Install-Services.ps1 already has sc.exe failure restart.
Changes (ScadaBridge parity):
- ActorSystemTerminationWatchdog (Host, registered after AddAkka): watches
ActorSystem.WhenTerminated and on an unexpected self-down calls StopApplication so the
process exits (supervisor restarts it) instead of idling with a dead actor system.
Distinguishes graceful shutdown via _stopRequested + ApplicationStopping. 3 unit tests.
- docker-dev: restart: unless-stopped on the host anchor (models production supervision) +
both redundancy peers in SeedNodes so a restarted node re-forms via either peer.
- docs/Redundancy.md: rewrote the split-brain recovery section — younger-loss = in-place
fast failover; oldest-loss = exit-and-rejoin under supervision (not in-place); the three
requirements (supervisor + watchdog + both-node seeds); flagged HardKillFailoverTests as
non-representative (Transport.Shutdown, not a real crash). Instant in-place takeover on
ANY single loss needs 3+ members.
Cluster.Tests 29/29 (SBR guards), watchdog tests 3/3, full solution builds.
Live re-verify of the watchdog image pending (host docker disk full).
2-container docker-dev rig: hard-killing the oldest node makes the younger
survivor down ITSELF (SBR DownReachable including myself -> CoordinatedShutdown)
-> no failover, total outage. Reproduced twice. Confirms #459 is REAL and
disproves Option A (keep keep-oldest). Baseline ServiceLevel 250(leader)/240(follower)
captured before the kill. Documents why (textbook 2-node keep-oldest limitation;
down-if-alone doesn't rescue a crashed oldest) and why the in-process
HardKillFailoverTests gave false confidence (Transport.Shutdown keeps the
ActorSystem alive, not a real crash). Fix needs Option B (witness+keep-majority)
or Option C (DB-lease arbiter).
Captures the docker-dev live verification for follow-up #10 acceptance item 2:
a non-default per-instance ResilienceConfig (Subscribe.retryCount:999) authored on
MAIN rode the deploy artifact into BOTH central-1/central-2 runtimes, which parsed
and clamped it — the authored value appears verbatim in the spawn-time diagnostic.
Confirms the artifact bytes carry ResilienceConfig and the runtime pipeline applies
authored (not tier-default) policy. Item 1 (raw retry/breaker-on-fault log line)
remains, with the connect-then-idempotent-write recipe documented.
The arch-review #10 sub-gap worried that per-instance ResilienceConfig never
reaches the runtime pipeline. The threading is in fact wired end-to-end (task #13):
ConfigComposer serialises the whole DriverInstance entity, so ResilienceConfig
rides the artifact into DriverInstanceSpec, which DriverHostActor layers onto the
driver's Polly pipeline. The only leg with no test was the real composer->artifact
serialization — the existing DeploymentArtifactTests hand-build the JSON.
Adds ResilienceConfig_survives_ConfigComposer_to_ParseDriverInstances_round_trip
(mirrors the DeviceHost round-trip guard): seeds a DriverInstance WITH a non-default
override + one WITHOUT, runs the real SnapshotAndFlattenAsync, and asserts
ParseDriverInstances recovers the override byte-for-byte (and null stays null).
Guards against a future projection / [JsonIgnore] silently reverting authored
resilience policy to tier defaults while hand-built artifact tests stay green.
ControlPlane.Tests ConfigComposerTests 6/6 green.
The patched native SQLite bundle shipped: SQLitePCLRaw.bundle_e_sqlite3 2.1.12
is the first version outside the CVE-2025-6965 / GHSA-2m69-gcr7-jv3q vulnerable
range (<= 2.1.11), embedding the SQLite 3.50.2+ fix.
- Bump Microsoft.Data.Sqlite 9.0.0 -> 10.0.7 (aligns with the EF Core 10.0.7 family).
This alone still pulls the native bundle 2.1.11, so:
- Add a surgical direct pin of SQLitePCLRaw.bundle_e_sqlite3 2.1.12 in Core.AlarmHistorian
(the sole consumer), overriding the transitive 2.1.11 without enabling
CentralPackageTransitivePinningEnabled (which breaks the Roslyn 5.0/4.12 split).
- Remove the temporary NuGetAuditSuppress from Directory.Build.props.
Verified: dotnet restore/build clean with audit active (no GHSA-2m69), every
in-solution project resolves lib.e_sqlite3 2.1.12, AlarmHistorian tests 29/29 green
against the real native bundle.
Closes#458
The [reference_*] pointers target agent memory (not repo files) and rendered as
undefined markdown shortcut-reference links; converted to plain code spans. All
repo-relative links + paths in infra/README.md verified to resolve.
The 4 non-blocking residuals (#10 live-gate/ResilienceConfig sub-gap, R2-11 Phase C
strict parsers, CVE-2025-6965 suppression removal, SBR oldest-node failover gap) are
now tracked as Gitea issues, cross-referenced in the STATUS.md capstone.
Consolidates the scattered infra facts (docker host layout, driver fixtures, the
three env-gated live-gate recipes, the Host.IntegrationTests harness, the HistGW
local deploy, docker-dev rig, shared GLAuth, deploy) into infra/README.md as the
single entry point, cross-linked to dev-environment.md + the deep docs. CLAUDE.md
points at it from the Docker Workflow section.
Ran Category=LiveIntegration 6/6 against the running local 0.2.0 HistorianGateway
(localhost:5221 h2c) + the real AVEVA historian over VPN. Confirms R2-08 S3's
GatewayHistorianDataSource read path and re-confirms R2-06. With R2-01 + R2-08
GLAuth already self-serviced, every Round-2 live gate is now GREEN.
LdapGlauthOutageLiveTests drives the S2 directory-outage circuit through the
production auth path against a real GLAuth (docker pause/unpause) on 10.100.0.35.
Only R2-06 VPN + R2-08 S3 LiveIntegration gateway soak remain (both need VPN).
Closes the R2-08 S2 live leg. The offline LdapAuthResilienceTests prove the
directory-outage circuit with fakes + an unroutable-host blackhole; this drives
the SAME circuit through the production auth path (OtOpcUaLdapAuthService +
LdapOpcUaUserAuthenticator) against a REAL GLAuth that is paused mid-run and
unpaused — and doubles as the first end-to-end verification that the PR #451
GLAuth swap (retired bitnami/openldap) binds correctly.
Triple-gated (GLAUTH_LIVE_HOST + GLAUTH_OUTAGE_START_CMD/STOP_CMD), skips clean
offline (verified: 1 skipped, 8 ms). try/finally always unpauses GLAuth.
Verified GREEN against 10.100.0.35:3894 (12 s): healthy binds (alice+bob, mapped
roles) -> docker pause -> 3 bounded failures -> circuit opens -> sub-second
"unavailable" fast-deny -> docker unpause -> half-open probe closes -> binds
succeed again. A boundary AuthTimeout feeds the circuit (authenticator :123-130),
so the docker-pause timeout shape is a faithful outage.
Follow-ups from the fleet-wide read-timeout audit that the S7 R2-01 read-leg
fix (PR #453) prompted. The audit confirmed S7 was the ONLY driver with the
async-read-ignores-socket-timeout hang; these are the two adjacent (non-hang)
findings it surfaced.
1. FOCAS TimeoutMs:0 footgun (the risky one): a non-positive Timeout made
SynchronizedFocasClient DISABLE its per-call wall-clock ceiling, reverting to
the caller's long-lived poll token — reintroducing exactly the frozen-peer
wedge S7 just eliminated, under misconfig. Clamp non-positive TimeoutMs to the
2s default at the config boundary so the deadline can never be authored away.
2. TimeoutMs validation symmetry: apply the same clamp in the AbCip + AbLegacy
factories. libplctag's Tag.Timeout setter throws on <=0, faulting tag creation
on every read/write; clamping keeps a misconfigured TimeoutMs:0 running on the
default bound instead. (Shared PositiveTimeoutOrDefault helper per factory.)
3. AbLegacy reconnect parity with AbCip: AbLegacy evicted the cached libplctag
runtime on neither the non-zero-status nor transport-exception read/write path
(AbCip evicts on both), so a data-path fault recovered only via the probe loop
/ libplctag internals. Added EvictRuntime + wired it into both read and write
failure paths so a fresh handle is created on the next call.
Tests: FOCAS 265->269 (clamp theory + positive), AbCip 336->339 (clamp theory +
positive), AbLegacy 209->212 (read-nonzero / read-exception / write-nonzero evict).
No production regressions; all three driver suites green.
Running S7_1500ConnectTimeoutOutageTests (docker-pause blackhole, self-serviceable
over SSH) uncovered + fixed the read-leg wall-clock gap STAB-14 left. Updates the
R2-01 STATUS row, the needs-user/still-open lists, and the STAB-14 remediation note.
The R2-01 live gate (S7_1500ConnectTimeoutOutageTests, docker-pause blackhole)
surfaced a real gap the offline fakes couldn't: S7.Net's ReadTimeout/WriteTimeout
map to the TcpClient's ReceiveTimeout/SendTimeout, which govern only SYNCHRONOUS
socket calls — the async read/write paths S7.Net uses ignore them. On an
established-but-frozen peer (frozen PLC / firewall DROP / cable pulled mid-flow,
TCP session still open) a read blocked until the OS TCP stack gave up (minutes),
silently wedging the poll loop: no Bad tick, no reconnect. STAB-14 fixed only the
CONNECT leg (EnsureConnectedAsync CancelAfter); this is its READ-leg sibling.
- New S7OperationDeadline: bounds every data-plane wire op with a wall-clock
ceiling (= _options.Timeout), surfacing an overrun as TimeoutException. Applied
in S7PlcAdapter to Read/ReadBytes/Write/WriteBytes/ReadStatus. OpenAsync is left
to EnsureConnectedAsync's own CancelAfter (not double-bounded).
- IsS7ConnectionFatal now classifies TimeoutException fatal → handle marked dead →
next EnsureConnectedAsync reopens (connect-timeout fix takes over from there).
- Tests: 5 S7OperationDeadline unit tests (deadline / token-honouring / caller-
cancel-passthrough / resultless), 1 driver-reaction test (read TimeoutException
→ reopen). Driver.S7.Tests 260/260.
- Live gate S7_1500ConnectTimeoutOutageTests now GREEN against the real snap7 sim
(8s: baseline Good -> pause blackhole -> Bad tick -> unpause -> recovered Good).
Integration-sweep follow-up #6 (SQL leg). The sweep flagged 7 heavy 2-node
deploy/failover/reconnect E2E tests as 'time out under amd64-emulated SQL' and
suggested raising deadlines. That diagnosis was WRONG: run against real
(native-amd64) SQL on the Docker host, they still hang — even at 4x deadline.
Root cause is that these tests had never actually run against real SQL, only the
EF in-memory provider, which does NOT enforce foreign keys.
Two distinct defects, both hidden by in-memory:
1. Missing FK-parent seed (5 tests). The deploy records a NodeDeploymentState row
per node, FK'd to ClusterNode (FK_NodeDeploymentState_ClusterNode_NodeId). The
tests never seeded ServerCluster + ClusterNode, so on SQL each node-state INSERT
throws and the deploy never seals (waits the full deadline regardless of size).
Added TwoNodeClusterHarness.SeedDefaultClusterAsync (seeds a ServerCluster +
a ClusterNode per node; Warm/NodeCount=2 to satisfy the SQL CHECK constraint
CK_ServerCluster_RedundancyMode_NodeCount AND the ClusterEnabledNodeCountMismatch
validator). Called it in DeployHappyPath x2 / Failover / FleetDiagnostics /
EquipmentNamespace; fixed DriverReconnect to seed node B + NodeCount=2.
2. Stale EquipmentId (EquipmentNamespace test; pre-existing on BOTH providers).
Seeded EquipmentId='eq-1', but the later-added EquipmentIdNotDerived validator
requires EquipmentId == DeriveEquipmentId(EquipmentUuid) ('EQ-'+first 12 hex).
Fixed the seed to a canonical id + matching UUID. Only surfaced once the FK fix
let the deploy reach equipment validation.
Also added OTOPCUA_HARNESS_SQL_HOST / OTOPCUA_HARNESS_LDAP_HOST overrides so the
harness fixtures can run on the native-amd64 Docker host (avoids arm64 mssql
emulation entirely).
Verified: the 5 SQL-backed classes go 11/11 green vs native SQL; in-memory default
unregressed (12/12). RoslynVirtualTagEvaluatorTests racing test is a pre-existing
flaky race (not SQL-backed; passes in isolation) — left as-is.
The Host.IntegrationTests opt-in real-LDAP mode (OTOPCUA_HARNESS_USE_LDAP=1)
used bitnami/openldap:2.6, gone since Bitnami deprecated their free image
catalog in 2025. Replaced it with GLAuth — the same LDAP server the rest of
the project already uses (docker-dev + the live OPC UA data-plane auth, both
against the shared GLAuth on :3893) — removing the lone OpenLDAP outlier and
the gone-image breakage.
- Added tests/.../Host.IntegrationTests/glauth/config.toml: baseDN dc=zb,dc=local,
a search-capable serviceaccount, dev users alice (full access) / bob (read-only),
and role groups with the same gidnumbers as scadaproj/infra/glauth so GroupToRole
maps identically.
- Compose ldap service -> glauth/glauth:latest, mounting the config, host :3894 -> :3893.
- Repointed the harness real-LDAP override from the OpenLDAP cn=admin/ldapadmin to
GLAuth's cn=serviceaccount/serviceaccount123.
Live-verified against a deployed container on :3894: the serviceaccount bind
searches and returns alice + her 5 memberOf groups; alice/bob bind with the
correct password; a wrong password yields Invalid credentials (49) — exactly the
OtOpcUaLdapAuthService flow (proven identical to the data-plane's shared-GLAuth path).
NB: real-LDAP mode is opt-in; the default Host run uses StubLdapAuthService, so
this never blocked the default suite. Integration-sweep follow-up #6 (LDAP leg);
the amd64-emulated-SQL deadline leg remains open.
Two test/fixture bugs; NO OtOpcUa driver change.
1. Topology: FocasSimFixture reads an env-overridable endpoint
(OTOPCUA_FOCAS_SIM_ENDPOINT, default localhost:8193) and exposes Host/Port,
but WireBackendTests + WireBackendCoverageTests hardcoded
focas://127.0.0.1:8193. Against a remote fixture the skip-gate passed
(remote reachable) but the driver dialed localhost -> KeyNotFound. Added a
DeviceUri property to the fixture; each test now derives DeviceHost from it.
Recovered 8 of 9.
2. Mock timer endianness (surfaced once #1 let the tests reach the mock): the
last failure read 60397977600 for a 3600 s power-on timer (= 0x3C000000 x 60).
The focas-mock's _wire_timer encoded the minute/msec fields big-endian (_u32),
but cnc_rdtimer's timer fields are LITTLE-endian on real hardware — a
documented, live-31i-B-validated quirk that FocasWireClient.ParseTimer decodes
LE (docs/plans/2026-06-25-focas-pdu-v3-30i-b-support.md). The driver was
correct; the mock was wrong. _wire_timer now emits little-endian (_u32_le).
Suite 9F/1P -> 10/10 against the remote fixture at ~/otopcua-focas.
Integration-sweep follow-up #3.
The sweep's AbLegacy 4F was a fixture-deployment gap, not a driver/test defect.
The suite needs ab_server in PCCC mode (--plc=SLC500|Micrologix|PLC/5), which
reuses the AbCip image (already on the host) and binds :44818. No PCCC container
was deployed, and the 4F-not-skip means :44818 was answering CIP (a leftover
AbCip container) so the PCCC reads returned Bad — same shared-port contamination
class as the Modbus false alarm.
Deployed the PCCC compose to ~/otopcua-ablegacy and ran each family with
AB_LEGACY_COMPOSE_PROFILE set to match the running container (required — without
it the [Theory] parametrises over all 3 families and 2 fail against a single-
family container): slc500 2/2, micrologix 1/1 (+1 skip), plc5 1/1 (+1 skip).
Skips are family-inapplicable file types. No code change.
Corrected the sweep record + added the AbLegacy bring-up + run recipe.
Integration-sweep follow-up #4.
The 2026-07 integration sweep flagged the Modbus non-standard profiles as
8F/5F/3F/8F and guessed 'pymodbus 4.x drift'. Re-investigation shows that was
wrong on both counts:
- The image runs the PINNED pymodbus 3.13.0 (verified via import pymodbus;
__version__). The '...removed in v4' log is 3.13.0's forward-deprecation
warning, misread as running 4.x.
- The failures were a fixture-cycling artifact: each profile is a separate
profile-gated compose service sharing :5020, and a plain 'docker compose down'
(no --profile) does NOT stop the running profile container. It kept holding
:5020; the next '--profile up' container silently failed to bind (compose
still says 'Started'), so every later profile run hit the STALE sim. Proven by
a raw Modbus probe: the 'mitsubishi' sim served DL205's 0xCAFE at HR0.
After force-removing all modbus containers and bringing up each profile cleanly
(verifying it actually published :5020), all four pass: dl205 16/16,
mitsubishi 13/13, s7_1500 10/10, exception_injection 17/17. No OtOpcUa or test
code change needed.
Added a profile-cycling gotcha + reliable-cycle recipe to Docker/README.md and
corrected the sweep record. Integration-sweep follow-up #2.
OpcUaServer.IntegrationTests was 4F. Two distinct, previously-masked issues:
1. Cert-store gap: the in-test client SecurityConfiguration used a bare
new SecurityConfiguration()/CertificateIdentifier(), so ValidateAsync threw
'TrustedIssuerCertificates StorePath must be specified' before any connect.
Added TestClientSecurity helper building Directory own/issuer/trusted/rejected
stores under a throwaway temp PKI dir (mirrors DefaultApplicationConfigurationFactory);
wired into DualEndpointTests + SubscriptionSurvivalTests.
2. Fixing #1 unmasked an SDK-version drift: the 1.5.378 node-cache read path
throws ServiceResultException(BadNodeIdUnknown) for a removed node instead of
returning a bad DataValue. The two subscription-survival assertions now expect
the throw via Should.ThrowAsync. Behavior under test (removed node -> unknown)
is unchanged; only the delivery mechanism differs.
Suite 4F -> 4/4 green. Integration-sweep follow-up #5.
KnownProfiles.All used a { get; } = [...] initialiser declared textually
before the static readonly profile fields it references, so C# static-init
order captured an all-null list. ForFamily() then NRE'd on p.Family, faulting
AbServerFixture..ctor and blocking the whole AbCip.IntegrationTests suite.
Expression-body it so it evaluates on access, after the fields are set.
Live-verified against the controllogix fixture: suite goes 6F/1P/3skip ->
10 pass / 0 fail / 3 skip (skips = emulator-mode tests).
Integration-sweep follow-up #1 (archreview/plans/INTEGRATION-SWEEP-STATUS.md).
The 9cad9ed0 'strip tracking-ID comments' XML-doc sweep collaterally
removed the comment explaining why the S7 subscribe command routes
OnDataChange through the CliFx IConsole (not System.Console), leaving
SubscribeCommandConsoleHandlerCommentTests red. The Modbus copy still
carries the comment verbatim and the test guards S7-mirrors-Modbus
parity, so restore it (matching the Modbus wording) rather than weaken
the guard.
S7.Cli tests 49/49.
The historian's Int1 analog-tag creation path is server-degenerate:
EnsureTags(Int1) stores an unusable stub (live-probed on the deployed
2023 R2 historian; root-caused + documented gateway-side in
HistorianGateway #11 / PR #16). HistorianTypeMapper mapped
DriverDataType.Boolean -> HistorianDataType.Int1, so Boolean historized
tags failed auto-provisioning.
Map Boolean -> Int2 instead (store 0/1). Int2 is a proven-creatable
analog type, and the value-write path already sends every value as a
double on WriteLiveValues regardless of tag type, so Boolean 0.0/1.0
round-trips cleanly. Both Float and Int2 are supported, so a
pre-existing Float Boolean tag retypes cleanly on its next deploy.
- HistorianTypeMapper: Boolean => Int2 (+ remarks)
- Unit pins: HistorianTypeMapperTests, GatewayTagProvisionerTests
(Boolean_definition_carries_explicit_Int2_presence)
- Live gate: EnsureTags_boolean_sandbox_provisions_as_Int2 (was _Int1);
retype probe reframed Float->Int2 (drops the dead Int1 leg)
- Docs: docs/Historian.md 'Boolean historization maps to Int2', CLAUDE.md,
archreview STATUS.md (R2-06 live gate -> 6/6)
Resolves#441.
- StorageRateMs bug found+fixed (PR #439); live suite 3/6 -> 5/6
- U-7: EnsureTags is a real upsert; supported-type retype ACCEPTED (Float->Double);
Int1 not creatable on this histsdk (ProtocolEvidenceMissingException, Int1-only)
- record the observed retype policy in docs/Historian.md (fills the placeholder)
Fixes the historized-tag provisioning path that silently failed at the gateway
(StorageRateMs=0 -> ArgumentOutOfRangeException). Found via the R2-06 live gate.
GatewayTagProvisioner built HistorianTagDefinition without a StorageRateMs, so
the proto uint32 defaulted to 0. The gateway's EnsureTags maps StorageRateMs
straight to the AVEVA SDK's storageRateMs, which requires a positive value —
0 makes the SDK throw ArgumentOutOfRangeException ('Storage rate must be > 0 ms'),
so EVERY historized-tag auto-provision (the deploy-time IHistorianProvisioning
hook) silently failed at the gateway with failed=N.
Stamp DefaultStorageRateMs=1000 (matches the gateway's own live provisioner
convention; quantized 1000/5000/10000). Add a regression unit test asserting the
definition carries a positive rate, and set StorageRateMs on the live-suite's
inline definitions so the retype probe reaches a real EnsureTags.
Found via the R2-06 live gate against a real 0.2.0 gateway + AVEVA historian:
with this fixed, Float/Int2/UInt2/Int4/UInt4/Double all provision and the live
suite goes 3/6 -> 5/6. (Int1/Boolean remains blocked by the gateway histsdk's
ProtocolEvidenceMissingException for Int1 — a gateway-side type-support gap, not
this bug.)
Gateway driver unit 103/103.
dohertj2
Joseph Doherty