Phase 1A live gate defect 4. Neither central node booted on this branch,
in ANY transport mode:
Akka.Hosting.MissingActorRegistryEntryException: No actor registered
for key ...ControlPlane.AuditWriterActorKey
at AdminUI.Audit.ActorAuditWriter..ctor(ActorRegistry registry)
at ControlPlane.ServiceCollectionExtensions.<WithOtOpcUaControlPlaneSingletons>
WithOtOpcUaControlPlaneSingletons' first WithActors starter resolved
IMeshCommandDialer and IMeshPskProvider eagerly for the Props. Their
graph reaches back into the registry — dialer -> IMeshPskProvider ->
ISecretResolver -> AddZbSecrets -> IAuditWriter -> ActorAuditWriter,
whose ctor calls registry.Get<AuditWriterActorKey>() — and
WithSingleton<AuditWriterActorKey> is registered ~60 lines LATER in the
same builder. Boot died before the entry existed. Mode-independent: the
resolve happened before anything consulted MeshTransport:Mode, so Dps
(the default, and the rollback target) died exactly like Grpc.
Both dependencies now arrive as Func<> and are invoked in PreStart, by
which point the host is fully built — the same shape, for the same
class of reason, as the coordinator proxy Func one argument earlier.
PreStart re-resolves on restart rather than caching across one; both
are container singletons, so a restart gets the same instances. A
resolution throw is caught and logged rather than propagated: a throw
out of PreStart is an actor failure, the supervisor answers it with a
restart, and a broken DI graph is not transient — that is an unbounded
restart loop that would take the transport-independent ack path down
with it.
Regression pin: ControlPlaneStarterOrderTests boots a REAL admin host
through WithOtOpcUaClusterBootstrap + WithOtOpcUaControlPlaneSingletons
in production starter order, with the production graph from
IMeshCommandDialer down to a registry-bound IAuditWriter, and asserts
StartAsync itself — the failure occurs while the builder's starters
execute, so building a ServiceProvider and resolving by hand cannot
reproduce it. Verified RED against the eager resolve: 3/3 modes fail
with MissingActorRegistryEntryException. A second fact pins that the
deferral did not become 'never resolved' — the actor must resolve the
provider from the container and drive it through the contact refresh,
signalled by a TaskCompletionSource (PreStart runs on the actor thread
and the contact load is a piped DB read, so no synchronous assertion
after StartAsync is correct).
Why nothing caught this: the dialer's container-activation test resolves
from a collection with no secrets registration, so its graph stops one
hop in; the actor facts inject fakes and never touch DI; and the
existing whole-host boots register no mesh provider, so the eager
lookups returned null and walked nothing.
Two aggravators recorded by the gate, no code here: the failed process
did not exit (docker Up while /health/ready 503s — Phase 2 readiness
concern, tracked), and this is the third instance of the same
eager-resolve-during-AddAkka class after scadaproj#1 and mesh Phase 6's
IClusterRoleInfo boot crash.
ControlPlane.Tests 236 -> 240, Host.Tests 92, Cluster.Tests 184,
0 warnings in every touched project.
The known-cluster gate published only under Mode=Grpc, but the
interceptor it feeds is registered on every admin node and 1B maps
CentralCommandService by LISTENER PORT — neither consults the mode. A
dark-switch rollback to ClusterClient or Dps would therefore have left
that endpoint serving with the gate permanently inactive, silently
restoring both the timing oracle and the pre-auth secret-store/audit
flood, as a side effect of a rollback nobody would connect to
authentication. The gate is central's cluster inventory, not transport
state, so it is now decoupled from the mode entirely:
- SetKnownClusters + InvalidateAll run on EVERY successful contact load
in EVERY mode, and the contact-refresh timer starts in all three.
Under Dps this is a ConfigDb query every ContactRefreshSeconds that
did not run before; central always has a ConfigDb, and one indexed
read a minute is the cheaper half of the trade.
- The GATE set is built from ALL ClusterNode rows. The eligibility
filter moved out of the query's WHERE and into the loop, so the FAN
set is byte-for-byte what it was (enabled && !maintenance) while the
gate widens. This also dissolves the maintenance coupling flagged in
1A.2: a cluster whose nodes are all disabled or in maintenance can
still ack, and refusing it would make maintenance mode an
authentication switch. Central still refuses clusters it has never
heard of.
- HandleContactsLoaded keeps a documented order — targets adopted, gate
published, cache dropped — now pinned by test.
Guards: PreStart mirrors the null-dialer error for a null PSK provider
(silent absence is the worst outcome there — nothing malfunctions, so
nothing is investigated, while the gate never publishes and keys never
rotate); a failed contact load now says out loud that the gate is stale
and, before any load has succeeded, inactive; MeshGrpcCommandDialer
clamps a non-positive DispatchTimeoutSeconds to 10 with a construction
Warning, because the validator for that key is driver-role-scoped and a
pure-admin central reads it unvalidated; the dialer's local PSK refusal
was shortened so its remediation survives the actor's 200-char
truncation of remote-authored status detail; the portless-node warn-once
set is rebuilt each refresh, so a node that regains then loses a port
warns again — the regression is the occurrence worth seeing.
Docs/pins: dialer disposal comment no longer claims one-way; the two
interceptor-test comments claiming Matches has a default implementation
corrected (it deliberately has none); new pins for per-refresh
invalidation, the SetKnownClusters-before-InvalidateAll order, and one
channel per endpoint via a new CachedChannelCount.
The_known_cluster_set_is_not_published_by_the_other_two_modes is
replaced — its premise is obsolete — by three facts: every mode
publishes, gate set is a strict superset of the fan set, and the handler
ordering.
ControlPlane.Tests 230 -> 236, Host.Tests 92, Cluster.Tests 184,
0 warnings.
MeshGrpcCommandDialer places one NodeCommandService.Execute call on one
node: a GrpcChannel cached per endpoint (the GrpcDeploymentArtifactFetcher
shape), the TARGET node's cluster's bearer plus x-otopcua-cluster naming
that cluster, and a per-call deadline linked to the caller's token. A
cluster with no resolvable key is refused locally with Unauthenticated —
there is no arm on which a command leaves central unauthenticated, even
one the far side would have rejected.
CentralCommunicationActor gains the third mode branch. The contact load
now also projects GrpcPort and, from the SAME rows in the SAME pass,
builds the dispatch set — which is what makes 'the gRPC fan reaches
exactly the SendToAll recipient set' structural rather than two queries
that have to keep agreeing. A row with a null GrpcPort is skipped
warn-once (the TelemetryNodeSource precedent). Per-node results are
piped back for LOGGING ONLY: named node, gRPC status, dropped. No retry,
no queue.
Every dial is scheduled on the thread pool rather than called from the
receive. Found by review and pinned red-first: an async method runs
synchronously to its first await, and the dialer's pre-await work is
TryGetPsk, which blocks on the secret store — bounded but seconds, and
repeated after every refresh since the refresh drops the cache. Dialing
inline parked the actor's mailbox, so deployment acks queued behind a
slow node; the regression test tells an ApplyAck while a dial is held
and fails without the hop.
The contact refresh also wires the Phase 1A register: it publishes the
loaded cluster ids as the PSK provider's known-cluster set and drops the
key cache, so a rotated key takes effect within one ContactRefreshSeconds.
Neither happens under Dps or ClusterClient.
Both existing mode branches are behaviourally untouched and every
existing fact passes unmodified. The one edit to shared text: the
Status.Failure warning drops the word 'ClusterClient', since that handler
now serves all three transports.
ControlPlane.Tests 205 -> 230, Host.Tests 92, Cluster.Tests 184, 0 warnings.
Phase 1A security register items 1 and 2, landed ahead of the command
fan-out because they are independently testable and the fan-out is the
first caller of TryGetPsk.
Item 1 — known-cluster gate. MeshPskProvider gains SetKnownClusters, an
atomic whole-set swap consulted by TryGetPsk BEFORE the config map and
before the secret store. A cluster id outside the set is refused in O(1)
with no store read and no audit write, which closes two things at once:
the pre-auth store/audit flood an unauthenticated caller could drive with
a forged x-otopcua-cluster header, and the timing oracle that defeated
CentralCommandAuthInterceptor's uniform denial text (unknown clusters are
never cached, so they paid a store read on every probe while known ones
answered from memory). Empty or never-published leaves the gate inactive
— central publishes the set from its DB contact refresh, and 'the DB told
me nothing' must not become a fleet-wide authentication outage during
admin boot.
Item 2 — bool Matches(clusterId, presented) on IMeshPskProvider, with
FixedTimeEquals inside the provider. CentralCommandAuthInterceptor now
calls it and never holds key material; its two provider-side denial logs
collapse into one (the provider answers both as false) and the operator
hint is restated there.
No production caller publishes a cluster set yet, so behaviour is
unchanged until the fan-out commit wires the contact refresh.
ControlPlane.Tests 190 -> 205, Host.Tests 91 -> 92.
Found on the docker-dev live gate, not in tests. Two drivers that faulted
during initialise (rtu-gate, opcua1) each rendered "0 tags" on /hosts against
two authored tags apiece, because a ledger that had never committed returned
SkippedTagReport.Empty.
A zero report is a positive claim -- "this driver skipped nothing" -- and
making it on behalf of a driver that never got as far as looking is exactly
the quiet inaccuracy #541 exists to remove. GetSkippedTags() and
SkippedTagLedger.Snapshot() now return null until the first committed build,
so the column renders the same em-dash it already used for a driver that
cannot report at all. Both are 'no answer'; neither is 'nothing skipped'.
Knock-on: a FOCAS driver whose only init threw now reports null rather than a
zero report, which is strictly more honest than the assertion the test
originally made.
A driver builds its RawPath -> definition table at init and tolerates a
per-tag miss: an unmappable TagConfig, an unparseable address, a
capability-matrix rejection. That tolerance is right -- one bad tag must not
take down the other few thousand. What was missing is any statement of the
fact afterwards. A skip is not an error and happens before any read, so the
driver stays Healthy, errors/5 min stays 0, Last error stays empty, and Last
read advances live off the tags that did map. On the wonder Makino in #540
that reading held while focas-z-34184 ignored 20 of its 22 tags.
New optional capability ISkippedTagReporter, backed by a shared
SkippedTagLedger so the reset/commit/cap rules are one implementation rather
than one per driver. The report -- authored count, exact skipped count, and a
capped sample of RawPath + reason -- rides DriverHealthChanged to a new Tags
column on /hosts, over the Phase 5 telemetry stream. No table: a driver-only
node has no ConfigDb to write to, the lesson from #521.
The issue named eight drivers. It is twelve -- every driver in
DriverTypeNames.All has an authored-tag skip path. Calculation (no scriptId),
MTConnect (no readable DataItem id), Galaxy (no attributeRef) and OpcUaClient
(no nodeId) were unlisted, and OpcUaClient's was the quietest of the twelve:
NamespaceMap drops such an entry with a bare `continue` and not even a log
line. SkippedTagReporterFleetTests pins the invariant reflectively over every
IDriver in src/Drivers/, so a thirteenth cannot re-introduce the silence.
Sql drops a tag at TWO points and its report is the union of both: a blob
that does not map, AND an identifier the 8.1 catalog gate rejects against the
live catalog. The second is what schema drift produces -- a renamed column, a
dropped view -- months after a deployment sealed green, and reporting only
the first would have that driver claim "0 skipped" while polling half its
tags, which is worse than reporting nothing because the column would then be
actively reassuring.
Deliberately additive to driver state, never a change of it. Latching
non-Healthy for a config problem would satisfy both halves of the #527
comms-loss conjunction -- 90 s without a successful read AND non-Healthy --
and a driver whose tags were ALL skipped has no successful reads by
construction, so it would reconnect-loop forever against a healthy device.
Three details that are load-bearing rather than incidental:
- The skip digest is part of the publish fingerprint. A skipped tag leaves
every other fingerprint component identical to a clean driver's, so without
it PublishHealthSnapshot's dedup swallows exactly the publish carrying the
signal -- the trap that already bit the rediscovery and host-status
channels. The digest covers the sample, not just the counts: a redeploy
that fixes one tag and breaks another leaves both counts unchanged.
- The count is exact; only the sample is capped, and the tooltip says so when
it is showing fewer than it counted. A truncated list that looked complete
would have an operator fix twenty tags and believe the job done.
- The ledger stages and commits, so an init that throws part-way leaves the
previous report standing. A partial snapshot would read as "we now skip
fewer tags", the opposite of the truth.
The proto field needs no companion presence flag, unlike host_statuses: a
singular message field carries real presence in proto3, and a has_ flag would
collide with protoc's generated HasSkippedTags.
Verified red-first three ways: dropping SkippedTagDigest from the fingerprint
fails the two dedup tests, removing the seam from one driver fails the fleet
guard, and dropping the catalog-gate leg fails the Sql union test. Full
solution builds with zero new warnings against 4e22f024.
Authoring FOCAS tags through /raw -> Browse device... produced tags the driver
could never read. Every committed tag read BadNodeIdUnknown forever, and the
failure was silent end to end: the AdminUI accepted it, the deploy gate sealed
green, and the driver reported Healthy with 0 errors while ignoring the tags.
Net effect: the whole fixed-tree surface could not be published through any v3
authoring path. Found cutting wonder-app-vd03 over to v3 against a real Makino
Pro 5 / Fanuc 30i, where 0 of 20 committed tags read while two hand-authored
MACRO: tags on the same device read Good.
A captured browse leaf's id IS its DriverAttributeInfo.FullName, and the fixed
tree emitted "{deviceHost}/{path}" there -- an internal cache key, not an
address. Both authoring paths wrote it verbatim into TagConfig.address, and
FocasAddress.TryParse accepts no such form, so the driver skipped the tags at
init.
Fixed at the emitter rather than by stripping the host at commit, which fixes
the browse-commit flow and the manual picker's tree-select together and leaves
one canonical address form:
- FocasAddress gains a fourth kind, FIXED:<path>, and moves to Contracts so the
AdminUI and ControlPlane can parse without referencing the runtime driver
(#526). A FIXED: read is served from the fixed-tree snapshot the poll loop
already maintains -- no area read, no connection. Writes refuse: nothing in
that tree is a writable location on the CNC.
- The address carries no device. A tag's device comes from its Device row, and
a second routing channel for one fact is what created this defect.
- An authored FIXED: tag arms the fixed-tree poll loop by itself. Enabled
defaults false and the browser force-enables it only for a browse, so the
operator would otherwise browse, commit, deploy, and land on a driver whose
loop never starts -- every tag Bad with a correct address.
- Fixed-tree values are now dated from the last answered poll and go
BadCommunicationError once it is stale (8 cadences, floored at 30s). The poll
loop swallows a failed tick by design, so a dead CNC would otherwise serve its
last snapshot as Good indefinitely -- harmless while nothing could bind to the
tree, not harmless now.
Guardrails, so this class cannot recur silently:
- FocasTagConfigModel.Validate() runs the real parser instead of a non-empty
check, and FocasTagDefinitionFactory.Inspect() warns at the deploy gate, which
covers CSV/API config that never touches the modal.
- The manual picker offered axis/spindle/program/status "parameter groups" --
none of which is a FOCAS addressing space -- and its own default value,
axis:5, was unusable. It now composes PMC / PARAM: / MACRO: through
FocasAddress, so what it emits is parseable by construction.
Tests assert addresses ROUND-TRIP through the parser rather than matching an
expected string: the previous FocasAddressBuilderTests pinned "axis:5" exactly,
which is how a green suite coexisted with a picker nobody could use. Verified
red-first against the old emitter. FOCAS 308, AdminUI 1095, ControlPlane 169,
Core 581 -- all pass, 0 warnings.
The remaining half of #540 -- a driver reporting Healthy/0 errors while
skipping tags at init -- is fleet-wide (eight drivers do this) and is filed
separately rather than patched per-driver.
SqlDriverProbe.DefaultDialectFor was left SqlServer-only when the provider
gates were widened, so AdminUI Test Connect failed for both new providers
with InvalidOperationException while Deploy and browse worked. Found by the
live chrome/AdminUI gate the plan deferred, not by any test: every case in
SqlDriverProbeTests builds its probe through ForTest(), which injects the
dialect and never reaches the production switch.
Widen the switch arm-for-arm with SqlDriverFactoryExtensions.CreateDialect,
and cover the gate directly via an internal seam — plus a reserved-provider
falsification control and an end-to-end production-constructor case.
Found by the new OracleReadTests against a real Oracle 23 Free: EVERY Oracle
driver instance failed its Initialize-time catalog load, and every AdminUI browse
of a table's columns threw, with
ORA-01745: invalid host/bind variable name
Oracle rejects a bind variable named after a reserved word, and the catalog
readers bind a parameter whose base name is `table` (SqlCatalogLoader /
SqlBrowseSession bind `schema` + `table`), so `ListColumnsSql` could not execute
at all. The provider was structurally non-functional.
OracleDialect now namespaces every marker/bind name through one `p_` constant
(`:p_k0` / `p_k0`), which a reserved word can never collide with, and its catalog
SQL is respelled to match. Prefixing everything rather than carrying a
reserved-word list is deliberate: the list is long and version-dependent, and a
future base name (`date`, `user`, `comment`, `level`…) would silently re-open the
same hole. Marker and bind name are derived from the same constant, so they
cannot drift apart.
No offline test could have caught this — only a server parses the SQL — so the
regression test asserts the invariant that made it possible (a bind name is never
the bare base name), with `table` called out explicitly.
Live evidence: 6/6 OracleReadTests red with ORA-01745 before, 6/6 green after,
against gvenzl/oracle-free:23-slim.
Widens the two provider gates so an authored `{"provider":"Postgres"}` or
`{"provider":"Oracle"}` binds that provider's own dialect instead of failing
"not available in this build":
- `SqlDriverFactoryExtensions.CreateDialect` gains a Postgres and an Oracle
arm. There is deliberately no fallback arm — a default resolving to
SqlServerDialect would construct happily and then emit T-SQL at a Postgres
or Oracle server. The rejection message now names the three supported and
the two still-reserved providers.
- `SqlDriverBrowser.DefaultDialectSelector` widens identically, so a provider
the runtime constructs is also browsable; MySql/Odbc keep resolving to null
and the caller keeps reporting them as unavailable.
Tests: the existing rejected-provider controls flip from Postgres to MySql
(factory) and Odbc (browser) so the reserved half stays falsifiable, and the
new legs assert the bound dialect TYPE rather than mere construction —
"no throw" alone would pass against a SqlServer fallback.
Doc-comment sweep (contract addition, one line each): the "v1 constructs
SqlServer only" / "reserved for P2" claims in SqlProvider, SqlDriverConfigDto,
ISqlDialect.Provider and SqlServerDialect are now false and were corrected in
the same change.
TryFormatValue's catch filter named only FormatException / InvalidCastException
/ OverflowException, and the call sits OUTSIDE WriteAsync's per-slot try. But
Convert.ToString falls back to the value's OWN ToString/IFormattable - arbitrary
caller code that may throw anything - so any other exception escaped WriteAsync
uncaught and failed the entire batch, breaking the documented never-throws
per-slot contract. Broadened to a catch-all; the slot gets BadTypeMismatch and
the exception is logged at Warning (TryFormatValue became an instance method to
reach _logger). One unformattable value now costs its own slot and nothing more.
Falsification checked, not assumed: restoring the narrow filter makes the new
test fail with a raw "System.InvalidOperationException : this formatter is
broken" propagating out of WriteAsync - i.e. it fails for the contract-breach
reason, not on an assertion.
Two tests added:
- a value whose ToString throws InvalidOperationException is refused
BadTypeMismatch while the *following* good slot still writes (the load-bearing
half - it proves the batch survived).
- mid-batch cancellation: cancellation injected from inside slot 1's own
formatting, so slot 0 has provably already reached the Agent. It keeps its
real Good outcome while slots 1-2 fill BadCommunicationError. Reporting a
landed write as failed is worse than reporting a failure, and a fill that
started at 0 instead of i would do exactly that.
Tests: 534 passed / 0 failed (was 532). Solution build: 0 errors, no warnings
in any MTConnect project.
MTConnectDriver now implements IWritable over the Task-1 agent-client seam
(WriteObservationAsync -> the Agent's AllowPut POST surface). A write is
OBSERVATION INJECTION, not machine command: it sets what the Agent reports for
a DataItem, is visible to every consumer of that Agent, and is overwritten by
the next adapter/SHDR update. Off at two independent gates, both off by default
- the Agent's own AllowPut and the driver's write.enabled (Task 0).
Per-slot refusal order (never throws, strict N-in/N-out batch shape):
write disabled -> BadNotWritable
no live client -> BadNotConnected (ReadAsync's existing convention)
reference unresolved -> BadNodeIdUnknown
no resolvable device -> BadNodeIdUnknown
CONDITION category -> BadNotWritable
null / non-scalar value -> BadTypeMismatch
agent refusal OR throw -> BadCommunicationError
OperationCanceledException fills the remaining slots and returns. The Agent's
refusal text is logged at Warning and never enters DriverHealth.LastError
(type-only discipline); a refused write leaves health untouched.
Device-name resolution: TagBinding gains WriteTargetsByDataItemId, built in
BuildBinding from a probe-model walk that remembers which device each DataItem
was declared under (name, or id when nameless) plus its category. Falls back to
options.DeviceName; unresolvable -> BadNodeIdUnknown rather than a guess.
Values render via Convert.ToString(..., InvariantCulture) - so a Boolean writes
the invariant string "True"/"False", NOT "1"/"0". Arrays are refused rather
than rendered: Convert.ToString would emit the literal "System.Double[]", which
the Agent stores happily and which is invisible until read-back.
FINDING - the CurrentWrite bit does NOT key on SecurityClassification, and the
discovery stamping below is therefore cosmetic. Traced end to end: a browsed
leaf's SecurityClass survives into the browse session's AttributeInfo
(CapturedTreeBrowseSession.cs:53) and stops there - RawBrowseModal never reads
it, RawBrowseCommitMapper.MapLeaf hardcodes TagAccessLevel.Read, and Tag has no
SecurityClass column. AddressSpaceComposer:746/:809 sets Writable purely from
Tag.AccessLevel == ReadWrite; AddressSpaceApplier ANDs !IsArray;
OtOpcUaNodeManager.ComposeAccessLevel:2002 composes CurrentWrite from that
alone. Driver capability is never consulted at compose time either - IWritable
is checked only at runtime in DriverInstanceActor:755. Consequences: (a) raw
tags already committed keep their authored AccessLevel, so nothing this commit
does makes an existing node writable; (b) an operator must still set the tag to
ReadWrite for an MTConnect write to be reachable over OPC UA; (c) conversely a
ReadWrite tag on a non-IWritable driver has always advertised CurrentWrite and
failed only at the actor - pre-existing, not touched here.
Discovery stamps CanWrite(writeEnabled, dataItem) = write.enabled && !CONDITION
-> Operate, else ViewOnly - kept honest so the browse never advertises a tier
the write path would refuse (MQTT's CanWrite principle), with the caveat above
recorded verbatim at the call site.
Status-code constants are spelled WITHOUT MqttDriver's "Status" prefix, to
match the two this file already declares (BadCommunicationError,
BadNotConnected). StatusCodeParityTests strips either spelling and passes.
TDD: WriteAsync landed first as a NotImplementedException scaffold - 19 of 20
new tests red on behaviour, 1 (the capability assertion) green on the scaffold
- then implemented. MTConnectFactoryTests' "not_writable" characterization test
is rewritten to pin the new truth, and is the guard for the concrete-return
contract the factory remark explains.
Tests: 532 passed / 0 failed (was ~509). Full solution build: 0 errors,
0 warnings; OTOPCUA0001 clean.
ISqlDialect gains ParameterMarker (the token in the SQL text), ParameterBindName
(the DbParameter.ParameterName) and ConfigureCommand (per-command provider setup,
where Oracle will set BindByName). SQL Server and the test SQLite dialect answer
"@"+name for both spellings and no-op the hook, so SQL Server behaviour is
byte-identical — every existing expected-SQL assertion passes untouched.
SqlGroupPlanner drops its hardcoded @k/@w consts and mints base names (k0..kN, w)
that the dialect spells; SqlQueryPlan carries the emitted markers AND the matching
bind names, so SqlPollReader keeps copying the plan's pairing rather than
re-deriving a provider convention (the two diverge on ODP.NET: ":k0" in the text,
bare "k0" on the parameter). SqlCatalogLoader and SqlBrowseSession bind
schema/table through the dialect, so their bind names cannot drift from the
markers in the dialect-owned catalog SQL. Every parameter-carrying command site
calls ConfigureCommand before its first parameter.
Controlled by a new test-only ColonMarkerDialect (":name" markers, bare bind
names) that really executes against the SQLite fixture. Each leg was falsified
against the pre-change behaviour: the planner emitted "@k0", the reader bound
":k0", and a hardcoded "@schema" in the loader failed catalog load outright.
Both defects were found by #491's live gate against a real HistorianGateway and filed rather than
fixed at the time, because each one's fix is a design decision rather than a patch.
#537 — both nodes of a redundant pair captured every value, so the historian received duplicate
live samples for every historized tag (measured: the gateway's write counter advanced twice per
value and halved when one central was stopped).
The recorder now reads a capture gate per value, fed from the new
PrimaryGatePolicy.ShouldRecordHistorizedValues through IHistorizationRoleView — a third verdict on
the same RedundancyRoleView object, registered unconditionally on driver nodes rather than
alongside the alarm section's IRedundancyRoleView (which only exists when that section is enabled;
piggybacking is the #511 registration accident).
Two things are deliberate and easy to get backwards:
* It gates the CAPTURE, not the drain — the opposite of the alarm sink. That queue is in LocalDb
and replicates, so the Secondary already holds the Primary's rows. The historization outbox is
a node-local FasterLog replicating to nobody, so a drain-only gate would leave the standby
appending forever to a queue no one empties: unbounded disk, then a flood of stale duplicates
the moment it is promoted. Entries a node already captured are still drained by that node.
* The peer it stands down for must RECORD THE SAME REFS, not merely hold the same rows — a
Primary in this node's own per-cluster mesh. A node carrying no cluster-scoped role cannot
establish that and never stands down: under a fleet-wide election the elected Primary is
usually running a different cluster's configuration, and deferring to it would end this
cluster's historization silently, with the drain reporting success on an empty outbox forever.
That is the scope mismatch that once suspended the alarm drain on every node of the rig.
An unknown role captures, matching the alarm drain — a duplicate sample is recoverable, a hole in
the trend is not.
#538 — the drain writes every per-tag group and then acks all-or-nothing (deliberate: RemoveAsync
truncates THROUGH an id, so partial acking could drop older unwritten entries). One permanently
failing tag therefore pinned the FIFO head while every healthy tag in the batch was re-submitted
to the historian on every pass, indefinitely — duplicate samples and an outbox that never
truncated, observed live at 8.3 MB from a single unprovisionable tag.
The drain now tracks consecutive failures per historian tag and abandons that tag's entries once
it exceeds MaxWriteAttempts (new, default 5), so the head advances. The counter only advances on a
pass where some OTHER tag wrote successfully: that is what separates "this tag is poison" from
"the backend is down". A count-only quarantine looks identical on the poison-tag case and silently
discards the whole backlog during an ordinary gateway restart, turning a duplicate-samples bug
into a lost-history one. Abandonment is also deferred until every failing tag is over budget, so
nothing is discarded on a pass that could not have advanced the head anyway. Each abandonment logs
a Warning naming the tag; the total rides RecorderStatus.AbandonedEntries.
Capacity also gains a non-zero default (500,000 entries) as the disk backstop for the one case the
quarantine deliberately cannot resolve — a historian refusing every write, where retaining is
correct. Set 0 to restore unbounded retention.
Every new test was falsified by reverting the production change it guards, and each one reddens
alone:
* drop the backend-reachable evidence -> only A_backend_that_fails_every_tag_abandons_nothing
* never abandon -> only A_permanently_failing_tag_is_abandoned_so_healthy_tags_stop_being_rewritten
* latch the capture gate at construction -> only The_capture_gate_is_re_read_per_value...
* treat any Primary as a scope peer -> only the two scope-mismatch cases
Full suite green. (Four OpcUaClient.Browser failures during the run were the opc-plc fixture being
down on the shared Docker host; brought it back up and they pass.)
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
Ran #491's live gate against a real ZB.MOM.WW.HistorianGateway talking to a real AVEVA
Historian, and it found a production-fatal defect on the first value it ever saw.
ContinuousHistorizationRecorder.OnValueChangedAsync touched `Self` after awaiting
IHistorizationOutbox.AppendAsync(...).ConfigureAwait(false). ConfigureAwait(false) discards
Akka's ActorTaskScheduler synchronization context, so `Self` — which reads ActorBase.Context —
threw NotSupportedException. The throw escaped into supervision, which RESTARTED the actor;
because production seeds an EMPTY ref set (the deployed set arrives later via
UpdateHistorizedRefs), the restart silently discarded every historized ref. Net effect: one
value captured, then nothing, with no further error to show for it. Fixed by capturing
`IActorRef self = Self` before the await — the idiom the GetStatus handler already applied to
Sender.
Thirteen unit tests missed it because InMemoryOutbox.AppendAsync returned
ValueTask.CompletedTask: a completed ValueTask never yields, so the continuation ran inline
with the context still installed. The production FasterLog outbox fsyncs per entry and does
yield. The first regression test was itself vacuous — `await Task.Yield()` posts back to the
CAPTURED context, so the append still completed on the actor thread and reverting the fix left
the test green. The fake must complete on a POOL thread (Task.Delay(1).ConfigureAwait(false)),
as a real fsync does; with that, reverting the fix reddens exactly one test.
The rig could not run this gate at all: docker-dev carried ServerHistorian__* but no
ContinuousHistorization__* keys, so the recorder could never spawn. Both centrals now carry
them (default OFF — the recorder needs ContinuousHistorization:Enabled AND
ServerHistorian:Enabled), with the outbox on the per-node named volume so it survives a
container recreate. ServerHistorian__UseTls was hard-coded true and is now parameterized; a
plaintext h2c gateway was unreachable without it.
Gate results: outbox append, drain, WriteLiveValues accepted by a real historian, and restart
convergence (capture resumes after a restart with no new deployment) all pass. The read-back
hop is NOT proven — the LAN historian accepts writes but serves no reads under a "License
violation: Database tag count exceeds license tag count", and the licensed corporate one needs
the VPN. EnsureTags provisioning is likewise unexercised there (RemoteGrpc-only, and that CAP
is not listening). #491 stays open for the read-back hop alone.
Also recorded, not fixed: the recorder has NO Primary gate, unlike the alarm store-and-forward
sink, so both nodes of a redundant pair send duplicate live values. Gating it would introduce a
failover window with no capture — that is a design decision, not a cleanup.
Evidence: docs/plans/2026-08-01-continuous-historization-live-gate.md
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
A browse-committed AB CIP / TwinCAT / FOCAS tag bound with its model's DEFAULT
data type, not the one the device reported. A browsed REAL committed as
`{"tagPath":"Pump_1_Speed","dataType":"DInt"}` — and those three drivers decode
the wire bytes by that field, not by the raw row's OPC UA DataType — so the tag
read four bytes of a float through an integer decoder and published a plausible
wrong number with no error raised anywhere. FOCAS was missed when #536 was filed.
The type is CARRIED, not derived. Reverse-mapping the leaf's driver-agnostic
DriverDataType looks like the smaller fix and is wrong: SInt/Int/DInt/USInt/UInt
all collapse into Int32 on the way out of the driver, so that version would bind
a 16-bit INT as a 32-bit DINT while appearing to fix the REAL. Drivers now state
their own type name on DriverAttributeInfo.NativeDataType, which reaches the
picker as AttributeInfo.AddressFields["dataType"] and the commit mapper from
there. A leaf that states nothing keeps the model default — the pre-#536
behaviour, kept deliberately as the fallback rather than a guess.
Numeric strings are refused before parsing: Enum.TryParse accepts "9" for any
enum and Enum.IsDefined then confirms it, so an unguarded parse would bind a tag
to a type nobody named. A test caught this in the first version of the guard.
--- and the other half, which the fixture was hiding ---
Test Connect on a device with NO authored tags — every AB CIP device at the
moment an operator first presses the button — reported "CIP handshake failed"
against a healthy controller. That was written down as a fixture limitation
("ab_server doesn't implement @raw_cpu_type"). Both halves of that were wrong:
* @raw_cpu_type is not a libplctag special tag at all (only @raw, @tags and
@udt/ are), so it goes out as an ordinary symbolic name that no controller
defines, and the controller's "tag does not exist" IS the successful answer.
* ab_server answered an unresolvable tag path with CIP 0x03 where a Logix
controller answers 0x04 — ErrorBadParam vs ErrorNotFound, which the probe
classifies oppositely. Fixed in vendored patch 0002.
* With the fixture then telling the truth, the real defect surfaced:
AbCipDriverProbe.IsReachableException matched the native C status spellings
("STATUS_NOT_FOUND"), but libplctag.NET throws with the MANAGED enum name
("ErrorNotFound"). Those substrings can never match, so the reachable
classification could not fire on the exception path — which is the path every
no-authored-tags probe takes. This was failing on real hardware too, not just
against the sim. It now recovers the Status and defers to IsReachableStatus,
so there is one classification rather than two that can drift.
Rejected: switching the fallback to @tags. libplctag builds a tag-listing tag
with NO network I/O, so it reports OK against any TCP listener — measured green
against a Mosquitto broker.
Verification: full solution suite green. Falsification runs confirm exactly the
intended tests go red — 9 for the type carry, 3 (the reachable statuses only) for
the probe classification. Live on docker-dev: browse-committed TestString/
Motor_Run/TestSINT now store String/Bool/SInt, alongside last session's
Pump_1_Speed still holding the wrong DInt. TestSINT is the falsifier — its row
DataType is Int32, identical to TestDINT's, so no reverse map could have produced
"SInt".
Known consequence: pre-existing browse-committed tags keep their wrong dataType;
the fix applies to new commits only.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
The AbCip controller browse could never be verified end to end: libplctag's
ab_server answers CIP service 0x55 (Get Instance Attribute List) on the Symbol
Object class with "unsupported", so the `@tags` walk had no backend. #468's own
definition of done offered a sim that serves `@tags` as an alternative to real
hardware; this is that.
The fixture now applies a vendored patch teaching ab_server to serve the Symbol
Object listing, with partial-transfer paging. Correctness is checked against
libplctag's own `list_tags_logix` rather than against our decoder — our decoder
is the thing under test, so its agreement would prove nothing. The clone is also
pinned to a concrete tag: the Dockerfile called `--branch release` "pinned" while
tracking a branch that moves.
That immediately exposed a real defect. CipSymbolObjectDecoder advanced to an
even byte boundary after each symbol name, but entries are packed with no
padding — libplctag walks them by sizeof(entry) + string_len, and a captured
reply confirms it. Any odd-length name therefore desynchronised every entry
behind it, and odd-length names are ordinary. In the live browse below,
`Motor_Run` is 9 bytes and first on the wire, so the old decoder would have
shown that one tag and silently dropped the other eight.
Every existing multi-entry test passed throughout, because the test builder
padded exactly as the decoder did: the two agreed with each other and neither
agreed with a controller. The builder no longer pads, and the layout is pinned
by a captured reply (CipSymbolObjectWireCaptureTests) instead of by a second
statement of the assumption. Restoring the alignment step reddens exactly the
three new tests and none of the 29 others.
Live gate on docker-dev, all three of #468's criteria:
- populated tree — all nine seeded tags render as leaves, in wire order
- refresh re-captures
- leaf commit — Pump_1_Speed + TestDINT saved with their CIP tag paths
The gate also surfaced two things it was not asked about: browse-commit drops
the discovered data type, so a browsed REAL commits as a DINT and the driver
decodes it as one (filed as #536), and Test Connect fails against this fixture
regardless of the device because it probes `@raw_cpu_type` (documented).
TwinCAT and FOCAS still have no browsable backend; #468's AbCip half is closed,
theirs is not.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
An area or line could be named "Packaging" or "Line 1", save cleanly, and then
fail the NEXT deployment with UnsSegmentInvalid. Because a deployment snapshots
the whole database, one badly-named area blocked EVERY cluster's deploy - and
whoever hit that was usually not whoever typed the name. Only the equipment form
carried the rule; the deploy gate held its own private copy.
The rule now lives once, on Configuration.Validation.UnsSegment, and both ends
call it. DraftValidator no longer restates it.
Enforced in UnsTreeService, not just on the forms:
- Create/Update Area, Line and Equipment reject an invalid name before touching
the row.
- The CSV import reports a bad name as an ERROR, not a skip - a skip means "you
already have this row", which would tell the operator the equipment exists
when it does not. Good rows in the same batch still insert.
A form attribute could never have covered the import, and with no bUnit in this
repo a rule only a form enforces is also a rule no test can reach. The forms
keep an attribute for the immediate message, but it now references the shared
pattern rather than restating it - a second copy is how these drifted apart.
One behaviour change beyond the fix: UnsSegment.AttributePattern admits
"_default", which the equipment form's hand-written literal rejected. The form
was stricter than the gate it was mirroring, so a _default row was uneditable
through the AdminUI.
Tests: 41, falsified with three production mutations - dropping the CreateArea
guard reds exactly the CreateArea rejections, turning the import error back into
a skip reds only the import test, and narrowing AttributePattern reds only the
attribute/IsValid agreement. Every rejection is paired with a name that must
still be accepted, or a guard that refused everything would pass.
Live-verified on docker-dev: the area modal now refuses "Packaging" in place
with the rule spelled out, "packaging-b" saves, and a CSV import of two rows
reports Inserted 1 / Skipped 0 / Errors 1 naming the offending row - the path no
form attribute reaches. Rig cleaned up and a further deployment sealed.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
The last stage of #520. Enforcement landed in stages 5-6; this closes the
authoring end, where a grant could be saved, deployed and sealed green while
resolving to nothing.
Scope kind and scope target are now chosen TOGETHER from a picker over the
cluster's raw and UNS trees (AclScopePickerModal -> AclScopeTree, backed by
IAclScopeService). They were previously an unrelated dropdown and a free-text
box, so "RawDevice + a TagId" was a well-formed row. Because the trees are
rooted at the grant's own cluster and each node supplies its own kind,
NodeAclScopeCrossCluster, NodeAclScopeIdUnexpected, NodeAclScopeIdMissing and
NodeAclScopeKindRetired all become unauthorable by construction.
CheckScopeAsync mirrors DraftValidator.ValidateNodeAclScopes rule for rule and
blocks the save when a scope resolves to nothing - re-checked on submit, not
trusted from form load, because a row whose entity was deleted since is exactly
where the operator never touches the picker. The list page marks such rows
unresolved with the reason. The principal field suggests known principals and
reports how many live keys reach the one typed; reaching nobody warns rather
than blocks, since authoring ahead of minting is legitimate.
Deliberate departures from the plan line, all recorded in the design doc:
- The banners are reshaped, not removed. "Not enforced" stopped being true at
stage 6, but enforcement is still a per-node flag the AdminUI cannot see, so
a narrow grant really does confer nothing on a default-configured node.
- A Cluster row produces a NULL ScopeId. The deploy gate rejects one that
carries an id, so filling in the cluster id would author undeployable rows.
- AclScopeTree is a new component rather than a third mode on RawTree, which
already carries the /raw editing tree and the equipment reference picker.
- UNS tags are not offered: a UNS signal is a UnsTagReference, and a Tag grant
already covers the raw node and every UNS projection of it.
Also fixed: the ACL editor had a DataAnnotationsValidator and no
ValidationSummary, so an invalid NodeAclId made Create a silent no-op.
Tests: 30 in AdminUI.Tests/Acls, falsified with six production mutations -
each turned exactly the intended tests red and no others.
Live gate PASSED on docker-dev, 11 legs; both central containers asserted on
the same image by container id. The pair that matters is legs 8 and 11: the
page and the deploy gate reject the same row with the same reason. Legs 5 and
9 are the positive controls - without them the gate passes against a page that
refuses everything. Not covered: UNS-realm scope selection (SITE-A has no UNS
areas). Rig restored to zero ACL rows.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
It told the operator "Read, Browse and Subscribe remain ungated" — which stage 6
is precisely the change to. A warning that misstates what a security control
does is worse than no warning, because it is read as authoritative.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
SubscribeToEvents' sourceId is the manager HANDLE, not a NodeState — the same
object-typed trap as ContinuationPoint.NodeToBrowse, found the same way. The
`sourceId as NodeState` cast compiled, always yielded null, and silently
disabled the gate: an ungranted session could subscribe to an equipment's event
notifier and receive every alarm it raises, including the condition fields,
while every value on it was refused.
Now uses the SDK's own IsHandleInNamespace, which cannot be wrong about the type.
The new test passes a null monitored item deliberately: the gate must deny
without touching the item, which also makes the test a discriminator rather than
a restatement — against the broken cast the call falls through to the base
implementation, which does dereference it, so the test fails loudly.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
On a BrowseNext the references list already carries the earlier pages, so the
summary's denominator counted them again. Log accuracy only — the filtering
itself was already scoped to this call's appended range.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
Completes data-plane enforcement behind Security:Acl:Enabled. Read (every
attribute, not just Value), Browse, TranslateBrowsePath, CreateMonitoredItems,
SubscribeToEvents and SetMonitoringMode now consult scoped NodeAcl grants.
With the flag off nothing changes.
Four things went beyond the stage as written, each because leaving it out
would have made the feature look like it worked:
- Browse needed ancestor visibility. The trie inherits downward, and a client
reaches a deep node by browsing the nodes above it, so without this every
grant narrower than the cluster was authored, shipped and unreachable. It is
a separate non-inheriting marker, NOT a synthesized Browse grant on each
ancestor: the additive downward walk would carry such a grant from Plant onto
Plant/Siemens/dev9 and every other sibling subtree. Pinned by a test that
goes through the evaluator — probing the trie helper directly passes against
the broken design, because a sibling is absent from the trie either way.
- Revocation had to reach LIVE subscriptions. Every other operation re-asks per
request; a subscription is authorized once and then streams, and streaming is
how machine consumers actually read this server — so stopping only the
re-asking operations would have made revocation close to cosmetic.
IDataPlaneAuthoritySource gained a Changed event.
A denied item is queued BadUserAccessDenied and then DETACHED from its
MonitoredNode2 fan-out, not set MonitoringMode.Disabled. Disabling discards
the item's queued notifications, so the client is silenced without ever being
told why. The over-the-wire test caught this; the unit-level path reports the
item as stood down either way. Detaching is also reversible, so a restored
grant resumes the feed with no client action.
- The decision cache holds the grant WALK, not the decision. Grants and the
trie are fixed for a generation; a key's ExpiresUtc passes with the clock, so
caching the whole decision would let an expired key keep working until the
next deploy — the stage-5 open-session hole in a subtler form.
- TranslateBrowsePath is gated, without which browse filtering is decorative: a
client refused a browse asks TranslateBrowsePathsToNodeIds for the same path.
Found en route: UserAuthorizationState's staleness ceiling couples two clocks.
MembershipResolvedUtc is stamped from DataPlaneAuthorizer's TimeProvider while
IsStale is evaluated against TriePermissionEvaluator's; they are the same
instance in production and diverge the moment anything injects a test clock, at
which point the evaluator denies EVERYTHING with a verdict indistinguishable
from "no grant matched". Stage 5's claim that these predicates cannot fire was
accidentally true; they are now disabled explicitly.
Also fixed: ContinuationPoint.NodeToBrowse is the manager HANDLE, not a
NodeState. The declared type is object, so the wrong cast compiled cleanly and
silently yielded null — skipping the browse source gate entirely.
Falsified: disabling the transit marking reddens exactly the three tests that
assert transit works; implementing transit as an ancestor grant reddens the
sibling-leak test; skipping the fan-out detach reddens the revocation test.
Full solution suite green.
The stage 8 live gate has NOT run yet.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
`Security:Acl:Enabled` (default false) makes scoped `NodeAcl` grants decide
data-plane authorization at the three seams that have one, and gives HistoryRead
a check it never had. Read, Browse and Subscribe stay ungated — stage 6.
The ACL decision REPLACES the server-wide role check rather than chaining with
it. Chained, a grant would have to confer the role as well as pass the scope
check, and only Cluster-scoped grants confer a role — so every narrow grant
would have stayed inert and the flag would have looked inert too. The
load-bearing test is an ALLOW: a session carrying no roles at all writing
successfully on the strength of its grant. A deny-only test passes against a
build where the ACL is never consulted, because the role gate denies too.
Found en route: the principal never reached the session. The authenticator
passed it as OpcUaUserAuthResult.DisplayName, but RoleCarryingUserIdentity
derives DisplayName from the presented token — for a key-bearing client that is
the static username `apikey`, identical fleet-wide. So the trie's matching key
was dropped at the identity boundary, and the audit ClientUserId on a rejected
write recorded `apikey` for every consuming system. Principal and ApiKeyId are
now carried explicitly.
Beyond the stage as sketched:
- Revocation reaches an OPEN session. Every decision re-checks the
authenticating key against the current snapshot; without it a key revoked by a
deployment kept its authority until the client chose to reconnect, which for a
machine-to-machine consumer can be indefinitely.
- UserAuthorizationState is built per CALL, not per session. The per-session
shape exists so an LDAP session can re-resolve membership that changed out of
band; nothing can change out of band under API keys, so the generation stamp is
the whole invalidation signal and a per-call record picks it up for free.
IsStale/NeedsRefresh therefore cannot fire here and are documented as vestigial.
- The trie is built inside DataPlaneAuthorityState.Apply, under the same
generation as the snapshot, rather than by a second wiring at the publish
actor. An uninstalled trie denies everything — indistinguishable from working
enforcement. Deleting the install turns 13 of the 22 new tests red.
- The deploy log reports what the BUILD dropped instead of re-deriving it. The
old count agreed with the builder on one of its four drop reasons; a rule with
a blank ScopeId, a retired scope kind, or a path not starting at a realm
vanished with the log reading zero.
Both IApiKeyStore implementations widened to the full grant set. The narrowing
was redundant — ApiKeyInterimRoleMapper applies the same filter itself — and
became a trap once scope was enforced. Pinned by asserting the roles are
unchanged, with the two grants carrying different flags so it cannot pass by
accident.
Falsified by mutation: enforcement never engaging, the write seam ignoring the
ACL, the realm dropped from the scope, the revocation check removed, and the
trie never installed each turn the intended tests red and no others.
Full solution suite green.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
Both libplctag drivers carried Port on their tag-create params and consumed it
nowhere. The runtimes set Tag.Gateway to the bare host, and libplctag does not
fail on a portless gateway — it substitutes 44818. So an authored port was not
ignored in the harmless sense; the connection was silently REDIRECTED to whatever
answers on the default port. On a plant network that is another PLC, and the
driver would read its tags and report them Good.
This is the root cause of #534's four AbLegacy integration failures, and the
explanation for every observation in that issue: with each PCCC family moved to
its own port (44819/44820/44821), the driver still dialled 44818 and kept talking
to the AB CIP fixture's ControlLogix container. Hence PCCC-against-ControlLogix
errors under every server-path / client-cip-path permutation — the permutations
were never reaching a PCCC server at all.
Established by a direct libplctag 1.5.2 probe outside the repo, which answered
#534's open question: PCCC works fine against ab_server for all three families,
with path="1,0" AND with path=null (only path="" throws ErrorBadData). Since the
library could reach the simulators and the driver could not, the difference had
to be in our Tag construction, and Gateway was the only field that differed.
Same probe confirmed libplctag accepts host:port in the gateway attribute, which
is what the fix emits.
AbCip has the identical defect and is fixed alongside. It has never been
observable there: every AbCip fixture profile binds 44818, so no test that driver
has could distinguish "uses the authored port" from "always uses 44818". A
fixture that only ever exercises the default value cannot falsify a rule about
non-default values, however many tests point at it.
The port is now emitted unconditionally rather than only when non-default, so the
value does not depend on libplctag's own default; a port <= 0 still falls back to
the bare host rather than producing "host:0".
The load-bearing guard is the AbLegacy integration suite, which addresses each
family on a non-default port and cannot pass unless the port is honoured — 4/4
green, having never once passed before. Unit tests in both drivers pin the
composition for machines with no fixtures. Falsified both ways: reverting the
runtime line reddens all 4 integration tests; dropping the port > 0 guard reddens
exactly the two fallback cases.
AbCip integration 10/13 (3 pre-existing skips), AbCip unit 345, AbLegacy unit 219.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
Two independent defects surfaced in review of #520 stage 3.
#532 — Cluster:Roles had no OTOPCUA_ROLES fallback, despite two doc sites
promising one. A node configured only through the env var — which drives ALL of
Program.cs's conditional wiring, so it reads as the authoritative knob — bound an
empty Roles array and joined the cluster carrying no Akka roles at all. It ran,
served, and logged nothing wrong, while central reported "no driver-role nodes
are Up" and every deployment to it timed out. Not hypothetical: the field fix on
wonder-app-vd03 was to hand-add Cluster__Roles__0/1 to a Windows service's
registry environment. Anyone reading the code to diagnose that box would have
concluded from the XML doc that the fallback was already applied.
AddOtOpcUaCluster now post-configures AkkaClusterOptions through the shared
RoleParser.ResolveEffectiveRoles, so BuildClusterOptions (Akka's own member
roles), ClusterRoleInfo, the ControlPlane singleton scope and
ConfigDbApiKeyStore's ClusterId all see one answer. Two sibling validators
(ConfigSource, Telemetry) read Cluster:Roles directly and were exempting the
exact node they exist for — an env-only driver-only node has no ConfigDb and
could sit on ConfigSource:Mode=Direct unchallenged. They resolve the same way now.
Also collapsed the four open-coded `Roles.FirstOrDefault(IsClusterRole)` +
ClusterIdFromRole copies into RoleParser.ClusterIdFromRoles. They had drifted:
three classified the cluster role ordinally, so a verbatim-bound `Cluster-SITE-A`
yielded a null ClusterId — and a null ClusterId applies NO cluster filter, which
made #528's cross-cluster over-grant guard fail OPEN. IsClusterRole is now
case-insensitive; the ClusterId suffix keeps its authored spelling, because
normalizing it would silently rename the Akka role of a running deployment.
#533 — three string-identity rules in one authorization path. AclScopeRef used
default (ordinal) equality while the trie builder filtered ClusterId and hashed
principals OrdinalIgnoreCase; the deploy gate's cross-cluster check was ordinal
too, which was the worse half — it could reject a whole deploy over a ClusterId
case difference the runtime would have accepted. It only worked because resolver
and builder were handed the same NodeAcl instances.
The rule is now stated once, on AclScopeRef, and applied throughout:
identifiers (ScopeId, ClusterId, Principal) compare OrdinalIgnoreCase; path
segments compare Ordinal. AclScopePathResolver re-keys its own maps rather than
documenting a comparer contract for callers, so the two independent producers —
EF at the gate, artifact JSON on the node — cannot diverge from the outside.
PermissionTrieNode.Children stays Ordinal: those are browse names, and Speed /
speed are two nodes that can coexist under one device.
Split PermissionTrieBuildDiagnosticReason.RetiredScopeKind out of
UnresolvedScope. A legacy FolderSegment row and a typo'd id had the same message
and different remedies.
Falsified: five mutations, each reddening only its own guard — remove the
PostConfigure (3 tests), revert IsClusterRole to ordinal (3), revert AclScopeRef
equality (1), revert the resolver's map comparer (6), revert the gate's
comparisons (2), remove the FolderSegment short-circuit (1).
Test isolation: OTOPCUA_ROLES is process-global and xUnit runs classes in
parallel, so the three validator suites that now resolve effective roles share a
non-parallel collection. Without it, cases passing no roles would intermittently
observe another class's env var.
Full solution build clean; full suite green apart from two pre-existing failures
confirmed identical on 3a6b2a1d (PrimaryGateFailoverTests, and
DriverProbeRegistrationTests expecting 10 probe types where 12 now register)
plus the AbLegacy integration suite, which needs a docker fixture.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
Makes the /raw device tree addressable by an ACL rule, and builds the
scope-path producer PermissionTrieBuilder has accepted since it was written
but which had no caller anywhere.
The plan for this stage said "add NodeHierarchyKind.Raw, whose path is the
RawPath segments Folder/Driver/Device/TagGroup/Tag". That shape is wrong.
RawFolder.ParentRawFolderId and TagGroup.ParentTagGroupId are both
self-referencing, so folders and tag groups each nest without bound and a real
RawPath is folder…/Driver/Device/group…/Tag. NodeScope was fixed-arity and
PermissionTrie.CollectMatches was five hardcoded steps, so the scope model
could not address a real raw tree at all. NodeScope is now {ClusterId, Kind,
Segments} and the walk is a segment loop. The trie STRUCTURE was already
variable-depth (NodeAclPath.Segments is a list, Children is a recursive
dictionary), so only the walker and the scope record had to change.
Four decisions worth keeping:
- The namespace level became the realm. Level 1 gated on NodeScope.NamespaceId,
but v3 has no Namespace entity — the snapshot was retired when the address
space went dual-namespace, and the identifier appears nowhere in production
outside NodeScope. So the level could only ever fail to match, which silently
made every sub-cluster grant unreachable while cluster grants kept working.
It is now a realm token derived from the scope's Kind, not carried on it, so
the walk and the producer cannot disagree about its spelling.
- Provenance comes from the granting row's own ScopeKind, recorded on TrieGrant
at build time. Depth-derived provenance could not name a raw level at all,
since the depth of "Device" varies with folder nesting.
- The trie comparer moved from OrdinalIgnoreCase to Ordinal. Raw sibling-name
uniqueness is ordinal, so Speed and speed can both exist under one device and
a grant on either would have matched both.
- The null-scopePaths "deterministic test mode" is deleted. It planted
unresolvable rows at a direct child of the trie root, a position the walker
cannot reach, so the grant looked placed and could never match — and every
test using that mode proved nothing about production placement. scopePaths is
now required, and a row that does not resolve is dropped with a diagnostic.
A scope resolves to a LIST of paths: a raw tag projected into N equipments is
reachable at N+1 NodeIds and a write through any of them lands on the same
driver reference, so a Tag-scoped grant covering only the raw node would be
bypassable through a UNS one. The lookup is keyed on (ScopeKind, ScopeId), not
ScopeId alone — ScopeId has no FK and no type discriminator, so an id collision
across entity types would hand a grant the wrong node's paths.
Also closed here, all found by review rather than asked for:
- NodeAcl rows had no deploy-time validation of any kind and were not reachable
from DraftValidator at all (DraftSnapshot carried no NodeAcls). They are now
gated for dangling ids, cross-cluster scopes, blank/contradictory ScopeId, the
retired FolderSegment kind, and undefined ordinals.
- ConfigDbApiKeyStore conferred roles from Cluster-scoped grants authored in ANY
cluster, because it filtered on scope kind with no ClusterId predicate while
ArtifactApiKeyStore read an already-cluster-filtered snapshot. On a fused
admin+driver node that was a cross-cluster over-grant. Its doc comment claimed
the two stores matched.
- The two AdminUI ACL banners still said enforcement was LDAP-group -> role
mapping, which stages 1/2 made false in both directions.
- The /apikeys page was never classified in PageAuthorizationGuardTests, so that
test has been red since the key-minting commit merged — I did not run
AdminUI.Tests before merging stage 4.
Falsified rather than assumed: 8 mutations against the trie tests, 5 against the
deploy gate (including one proving a cross-cluster row is not silently
mis-diagnosed as unresolved), a staged red->green for the over-grant regression,
and the realm gate, the comparer and the malformed-path guard each broken
individually to confirm only their own tests go red.
Node ACLs are still not ENFORCED — nothing constructs a NodeScope or calls the
evaluator yet. That is stages 5 and 6; this stage makes the model able to
express what those will enforce.
Claude-Session: https://claude.ai/code/session_015p7wGqy3YpZNCpDzTpGMKo
Joseph Doherty