`OpcUaClientTagDefinitionFactory.TryReadNodeId` still honours a legacy PascalCase
`NodeId` when the v3 camelCase `nodeId` is wholly absent, but
`OpcUaClientTagConfigModel` read only `nodeId` — so a legacy blob was SERVED by
the driver yet opened with an empty reference field and could not be saved
untouched. Authoring stricter than the reader, which #554 names as never
legitimate: the editor was blocking config the runtime honours.
Fixed with the #550 pattern (read both spellings, normalise on save), NOT by
retiring the spelling from the reader — that would break every legacy blob the
driver currently serves at the next deploy, a data-plane regression traded for
an editor asymmetry.
The precedence is PRESENCE-shadows, not first-non-blank. The reader's
`TryGetProperty("nodeId", out n) || TryGetProperty("NodeId", out n)`
short-circuits on the mere PRESENCE of the camelCase key and then checks
string-and-non-blank ONCE, so a blank / non-string / JSON-null `nodeId` beside a
valid legacy `NodeId` is a REJECTED blob, not a fallback. Copying MTConnect's
`FirstNonBlank` walk here would have made authoring WEAKER than the reader —
the same defect in mirror image.
- `NodeIdKeys` drives both halves, so the read and the drop cannot drift.
- `ToJson` writes `nodeId` and drops `NodeId`; the TagModal history-merge keys
(`isHistorized` / `historianTagname`) are disjoint and untouched.
- `Validate`'s required-reference fast path is now precedence-aware by
construction, so it agrees with the #554 parse consult across the whole
rejection surface.
- The `TagConfigNonEnumParityTests` exemption for OpcUaClient is REMOVED — its
anti-vacuity guard `Every_deliberately_stricter_entry_is_still_exactly_that`
went red on the fix, which is this change's falsification control. The legacy
shapes moved into the corpus, where both fleet theories now cover them.
Tests: a nine-row precedence theory driving the authoring verdict against the
READER's verdict per shape, on the original blob AND on the `ToJson()` output;
no-retype (legacy loads showing its reference, saves canonical, same node id
before and after); no-drift (output never carries `NodeId`). Each half was
reverted in isolation to confirm it reddens tests of its own: read-only-revert
=> 6 red incl. the fleet parity theory; drop-only-revert => 4 red.
Suites: AdminUI.Tests 1459/1459, ControlPlane.Tests 261/261,
Driver.OpcUaClient.Tests 143/143, Driver.OpcUaClient.Browser.Tests 10/10.
#549/#551 made every typed editor judge its driver's enum KEYS against the runtime's
own reader and stopped there. Every other reason a reader refuses a tag met no gate
until deploy, or until the driver silently skipped the tag: an absent Modbus `region`,
an `"address":"notanumber"`, an S7 `stringLength` of 300, a TwinCAT `dataType:
"Structure"` (a VALID enum name the reader refuses), a Sql blob the parser cannot
recognise as Sql at all, an MQTT `"qos":null`.
Rather than hand-mirroring N more rules into N models — the drift #551 closed on the
deploy side — each `<Driver>TagConfigModel.Validate()` now asks its driver's own parse
whether it would read the AS-LOADED blob, via the new `TagConfigParseGate`. One
authority for the verdict AND the wording.
Two cautions are load-bearing and are pinned by tests:
- The gate is the PARSE VERDICT, never "the inspector said something". FOCAS, MQTT and
Calculation inspectors emit advisories for blobs their reader ACCEPTS, so a
non-empty-inspection gate would make the modal stricter than the runtime and leave
rows nobody can save (#520/#535). Inspect is consulted only to word a refusal that
already happened.
- It judges the blob the model was LOADED FROM, not the typed fields — the lenient load
("address":"notanumber" ⇒ 0) is what hides the defect. Same capture-the-source shape
as S7's `_dataTypeMissingInSource`.
Deliberately-stricter editor rules were surveyed and KEPT, not aligned: AbLegacy's
arrayLength > 256 (reader clamps, but the un-clamped value still sets ArrayDimensions),
FOCAS's unparseable address (#540), MQTT wildcard topics + Sparkplug segment chars, and
Calculation's trigger/50 ms floors. They are now recorded fleet-wide in a table the
tests assert is still accurate.
Behaviour change: Modbus `{}` is now refused. Modbus was the one mapped editor with no
required field, and `{}` is exactly what the modal persists for a config nobody touched
— the driver skips it and it reads BadNodeIdUnknown forever with the deploy green. Its
pinned test asserted the old, weaker behaviour and is rewritten with the rationale.
Falsification: neutering only the Modbus consult turns 14 tests red, all Modbus-scoped;
every #549 enum test stays green.
RawBrowseCommitMapper.BuildTagConfig's unknown-driver fallback claimed to be
guarded by RawBrowseCommitMapperParityTests, which did not exist anywhere under
tests/. The claim it made — "every BROWSABLE driver is handled above" — is the
exact claim that stopped being true when SqlDriverBrowser registered (deferment.md
G-6), so the comment was asserting the invariant instead of pinning it.
The guard now exists and enumerates DriverTypeNames.All:
- Browsability's bespoke arm is DERIVED from production, not hand-listed: the test
runs AddAdminUI() and reads back every registered IDriverBrowser's DriverType —
the same set BrowserSessionService.CanBrowse indexes. A driver that gains or
loses a bespoke browser moves the guard with it.
- The universal arm (DiscoveryDriverBrowser -> ITagDiscovery) is NOT derivable from
the AdminUI assembly: it references driver Contracts/Browser projects, never the
runtime drivers, and IDriverFactory is composed in the Host. Covered instead by
requiring every declared driver type to be typed or explicitly exempted with a
reason (Modbus, Calculation), with a third fact mechanically refuting any
exemption the derived arm contradicts — the Sql failure mode.
Falsified three ways, each red naming the offender: disabling the Sql branch trips
the derived-arm fact; disabling the MTConnect branch trips the enumeration fact
(the universal-arm half, which no bespoke registration would have caught); adding
Sql to the exemption list trips the contradiction fact.
Comment corrected to state what the test actually asserts, including the residual
trust the universal arm leaves.
`RawBrowseCommitMapper.BuildSqlTagConfig` emitted a WideRow blob carrying only
table + columnName. A wide-row tag must also say WHICH row it reads, so both
`SqlEquipmentTagParser` (the runtime authority) and `SqlTagConfigModel.Validate`
(the authoring gate) refused it — a browse-committed Sql tag was never readable,
first silently as BadNodeIdUnknown and, since #551 put the gate on the import,
loudly at commit.
`SqlBrowseSession` now states a fourth address field: the table's preferred
date/time-typed column, computed from the columns its attributes call ALREADY
read (no extra catalog round-trip). Candidacy is decided by
`ISqlDialect.MapColumnType` reporting `DriverDataType.DateTime` — never by column
name, which would order a wide-row read by a string column that merely looks like
a timestamp. Multiple candidates tie-break on FIRST IN CATALOG ORDER (every
dialect's column SQL orders by ordinal position), so the pick is stable across
calls and servers; it is a visible, editable default the typed editor round-trips.
No candidate (a table with no date/time column) is an operator-actionable refusal,
not an invented selector: the only remaining selector is a whereColumn/whereValue
pair whose value is data, not schema, so `DescribeUncommittableLeaf` refuses that
leaf by name, names the table, and points at the tag editor.
The session and the mapper now agree via `SqlBrowseAddressFields` constants rather
than two independently spelled string literals — a disagreement there is silent,
and is exactly the shape this defect had.
`RawTagConfigGateTests`' Sql leg flips from "records the gap" to "browse-commit
output passes the gate", with the no-candidate case pinned as its own test.
ModelBackedCsvDriverTagMap.ApplyTypedColumns round-tripped the row's free-form
TagConfigJson cell through a driver model's FromJson/ToJson to fold the typed
columns in. That load is deliberately lenient — it has to be, so a poisoned blob
can be opened and repaired — so a typo'd value under a key the model owns
("dataType":"Intt16") came back out as the model's DEFAULT, before the #551
import gate or any later gate could judge it. The row imported clean and the tag
the operator described was gone. Only keys the model did not own survived to be
judged, which is why RawTagConfigGateTests looked like it covered this.
MTConnect carried the same defect from the export end. It is the one driver whose
blob accepts several spellings per field, and the runtime PREFERS driverDataType
over the dataType key the MtDataType column writes. The export residual carried
that alias (the map owned only the canonical key), so a re-imported row let it
out-rank the very column the operator had just edited — accepted, then silently
ignored at the driver — while the exported cell printed the spelling the driver
was ignoring, or nothing at all.
The fix preserves evidence; it does not move the verdict:
- ApplyTypedColumns writes the base cell's own tokens back over the model's
serialisation for every key no non-blank typed column overrode. The model still
stamps defaults for keys the row declares nowhere, so fragment cells compose
exactly as before.
- CsvTypedColumn gains ReadKeys / OwnedJsonKeys: the full ordered set of spellings
the RUNTIME reads for that column, mirroring TryFromTagConfig's FirstNonBlank
pairs. A firing column clears the out-ranking aliases; the export residual drops
them; the export cell shows the EFFECTIVE value.
- The review grid's override provenance reads the whole owned set, so an
overridden driverDataType is announced as an overridden dataType would be.
Nothing here decides which values are legal. The verdict stays with
TagConfigValidator on the assembled blob and with IRawTreeService.ImportTagsAsync
behind it — the deliberate reason NOT to follow the issue's own "run the validator
over the pre-merge cell" direction, which would have imposed completeness on a
cell that is legitimately a fragment and false-rejected every typed-column row.
Semantics chosen for the override contradiction: a cell token that a non-blank
column overrides is ACCEPTED with the column winning, and reported by name in
TypedOverrodeFallback. Typed-column-wins is the map's documented contract, the
override is announced rather than silent, and erroring would remove the only way
to repair a poisoned tag through the CSV surface. Every token that reaches the
stored blob is judged; a token the operator overrode is discarded openly.
Falsification: neutering the restore loop + alias handling turns exactly 7 of the
15 new tests red, while RawTagConfigGateTests (#551), RawTagCsvRoundTripTests and
CsvColumnMapReflectionTests all stay green — layering, not replacement.
AdminUI 1248, MTConnect 534, Modbus 373, S7 285, TwinCAT 205 — all green.
Verified against the live mapper: every browsable driver's browse-committed
TagConfig passes the gate the commit now runs, EXCEPT Sql — its branch emits a
WideRow blob with no rowSelector, which SqlTagConfigModel and
SqlEquipmentTagParser both refuse. That tag was never readable; the gate only
moves the failure from the deploy gate to the commit. Reported for its own fix
rather than special-cased here.
Gitea #551.
1. CSV import ran no driver-typed config gate. Its typed columns were always
strictly parsed, but the free-form TagConfigJson column reached the database
unread, so a CSV could author a tag the driver rejects — one that deploys
green and reads BadNodeIdUnknown forever. RawTagCsvMapper.Parse now names the
offending row in the staged review grid, and RawTreeService enforces the same
gate on CreateTag/UpdateTag/ImportTags: a form cannot be the guarantee (the
API and any future caller reach none of them, and this repo has no bUnit).
A rejected import row is an ERROR — the batch inserts nothing and the row is
named — never a skip.
Note the brief's premise that manual entry rides ImportTagsAsync is wrong:
RawManualTagEntryModal commits row-by-row via CreateTagAsync, which is why
the gate had to land on all three write paths.
2/3. The seven pre-#549 Inspect methods ran a hand-list of DescribeInvalidEnum
checks, so every rejection cause not on the list was invisible. They are now
ACCEPT-GATE-FIRST (the MTConnect/Sql/OpcUaClient/Calculation shape): ask the
driver's own reader for the verdict, then explain the refusal, with a
catch-all so "the parse rejects but Inspect said nothing" is unreachable.
That closes S7's absent dataType, AbCip members[], Modbus stringByteOrder and
the type-dependent guards, plus causes the issue did not name (absent
region/address/tagPath/symbolPath/topic, TwinCAT Structure, MQTT's Sparkplug
tuple). FOCAS and MQTT keep advisories about ACCEPTED blobs, so they
diagnose rather than return empty.
Authoring half: ModbusTagConfigModel gains stringByteOrder;
AbCipTagConfigModel asks the new AbCipTagDefinitionFactory.DescribeInvalidMember;
S7TagConfigModel captures whether the SOURCE blob declared dataType and asks
S7TagDefinitionFactory.DataTypeKey/DeclaresDataType. One authority per rule.
4. Doc sweep: the six factory Inspect summaries, EquipmentTagConfigInspector's
class remarks and AdminOperationsActor's comment all claimed invalid enums are
"silently defaulted by the lenient runtime" — false since R2-11 Phase C made
the runtime reads strict. They describe rejections now.
MTConnectTagDefinitionFactory.TryFromTagConfig resolves the coercion type as
FirstNonBlank(blob.DriverDataType, blob.DataType) — driverDataType OUT-RANKS
dataType — while MTConnectTagConfigModel read and wrote dataType alone. That
broke in both directions at once:
* a blob carrying "driverDataType": "Strng" was accepted at authoring while
the runtime rejected the tag (the residual slice of #549), and
* a VALID driverDataType silently overrode whatever type the editor displayed
and saved, so the editor showed a lie and a save RETYPED the tag.
Both halves land together, mirroring the existing IdKeys pattern:
* FromJson reads DataTypeKeys = [driverDataType, dataType] in the runtime's
precedence, with the runtime's own per-key semantics — absent/JSON-null and
present-but-blank fall through, a non-string token stops the walk (the
runtime deserialise throws JsonException on it and rejects the whole tag),
and the raw TOKEN is captured so Validate() judges what the driver judges.
* ToJson drops every non-canonical DataTypeKeys entry and writes dataType as
the enum NAME, so the alias can never out-rank the key the editor writes on
the next load.
* Validate() now names the offending spelling rather than always saying
"dataType".
Doing only one half silently retypes existing tags on their next save; the
falsification runs recorded in the PR confirm each half-fix reddens a distinct
test.
MTConnectTagConfigModel read `dataType` with TagConfigJson.GetEnum, which falls
back to String on ANY unrecognised value. A misspelled name ("Strng") from a
hand-authored blob, a CSV import or the API was therefore accepted at authoring
and silently rewritten, while the driver's strict parse rejected it and the tag
died with BadNodeIdUnknown. Found by the G-7 parity tests' falsifiability
control, which had to use the digits-only shape to have a case that failed.
The rule is not restated in the AdminUI. MTConnectTagDefinitionFactory exposes
its parse as `IsAcceptedDataTypeName`, documented as the one authority the
driver, the deploy gate and the authoring model all consult; the model's
Validate() asks it about the raw value the blob carries. The old digits-only
message is NOT retained as a special case: the runtime draws no distinction
between a bare number and a typo — one predicate, one Inspect warning — so a
second message here would be a second statement of a rule this model does not
own. The raw value is captured as the JSON TOKEN, because the runtime
deserialises the key into a `string?` DTO field and an unquoted number throws
there too.
The survey found the same asymmetry at six more editors, contrary to the
assumption that those runtimes silently default: R2-11 Phase C moved every
driver's TagConfig enum read to TagConfigJson.TryReadEnumStrict, which REJECTS
the tag on a present-but-invalid string, while the editors kept loading them
with a silent fallback. Modbus (region/dataType/byteOrder), S7, AbCip, AbLegacy,
TwinCAT and FOCAS now check their enums in Validate() through the runtime's own
TagConfigJson.DescribeInvalidEnumText — added alongside a public
ClassifyEnumText that TryReadEnum now delegates to, so the JSON-token door and
the string door reach one implementation. Mqtt and Sql already carried
equivalent strict checks and are untouched.
Care taken NOT to become stricter than the runtime: a present-but-non-string
token stays accepted for those six (TryReadEnum calls it absent), as does an
absent key, each pinned by a test.
Also adds a direct ProjectReference to Core.Abstractions from the AdminUI — the
save gate now leans on a type it was consuming only transitively.
G-7 wired OpcUaClientTagDefinitionFactory.Inspect into the deploy-time TagConfig
gate, which gave ControlPlane a ProjectReference to
Driver.OpcUaClient.Contracts — and with it, transitively, the whole OPC UA client
stack (5 OPCFoundation packages) into every host that loads the control plane.
NamespaceMap was the only reason Contracts carried
OPCFoundation.NetStandard.Opc.Ua.Client: its entire public surface is
ISession/NamespaceTable/NodeId-typed. It moves to a new leaf project,
ZB.MOM.WW.OtOpcUa.Driver.OpcUaClient.Session, referenced by the runtime .Driver
and by .Browser. Contracts is left with Core.Abstractions and NOTHING else —
the same shape as Calculation.Contracts.
The namespace stays ZB.MOM.WW.OtOpcUa.Driver.OpcUaClient, so no consumer's
`using` changed. The .Session -> .Contracts direction is deliberate: the blob
read NamespaceMap needs is OpcUaClientTagDefinitionFactory.TryReadNodeId, the
one authority it shares with the deploy gate, and that must stay in the
dependency-light half.
Why a new project rather than folding NamespaceMap into the runtime .Driver and
having .Browser reference it (the Driver.Mqtt.Browser #512-style deliberate
exception): the AdminUI project-references .Browser and NOT .Driver, so that
shape would pull the OpcUaClient driver runtime — plus Core,
Opc.Ua.Configuration and Secrets.Abstractions — into the AdminUI's build/deploy
graph. That is exactly the violation #512 spent two extractions removing from
Mqtt.Browser; re-introducing it to save one csproj is a bad trade.
Guard: ContractsIsDependencyLightTests (Driver.OpcUaClient.Tests) asserts the
Contracts manifest names no Opc.Ua* assembly, plus a counterpart asserting
.Session does. An ABSENCE assertion is sound on the manifest where a presence
one is not — the compiler elides an unused reference, but it MUST emit one the
moment Opc.Ua-typed code is added. Falsified by temporarily re-adding the
package plus an Opc.Ua.NodeId-typed member: red on Opc.Ua.Types, green again on
removal.
Evidence — `dotnet list ControlPlane package --include-transitive | grep -i opc`:
before: OPCFoundation.NetStandard.Opc.Ua.{Client,Configuration,Core,
Security.Certificates,Types} 1.5.378.106
after: none (and 0 Opc.Ua*.dll in ControlPlane's output directory)
Adds the Sql / MTConnect / Calculation / OpcUaClient Contracts ProjectReferences to
ControlPlane and their entries to EquipmentTagConfigInspector, closing the G-7 asymmetry: the
deploy gate now checks every driver the AdminUI checks. Galaxy is the sole remaining unmapped
type and is so by design — its dotted tag_name.AttributeName reference has no typed shape, and
the AdminUI has no typed editor for it either, so the two gates agree.
Adding the ProjectReference is what makes EquipmentTagConfigInspectorCoverageTests demand a map
entry (its sweep scans the output directory, so an unreferenced Contracts assembly is invisible
to it). Falsified: deleting the Sql entry fails
Every_driver_shipping_an_Inspect_is_wired_into_the_dispatch_map with
["SqlTagDefinitionFactory"]; restored.
Tests: three cases per driver in EquipmentTagConfigInspectorTests (canonical valid config warns
nothing / present-but-invalid field draws a warning naming it / broken JSON warns without
throwing), plus a Calculation control pinning that the two shapes the runtime accepts and the
editor refuses draw nothing. OpcUaClient moved out of the unmapped-driver theory.
New TagConfigAuthoringDeployParityTests (in AdminUI.Tests, the only suite that sees both gates)
pins the direction that matters: a config the AdminUI model's own ToJson produces and Validate
accepts draws no deploy warning. Its falsifiability control — the same driver types warning
about a blob the editor refuses — caught a pre-existing, unrelated authoring-side gap:
MTConnectTagConfigModel reads dataType leniently, so a misspelled name loads as String and
saves clean while the strict runtime rejects it. Left alone (rewriting the editors is out of
scope); the control uses the digits-only shape both gates refuse, and the case is documented
inline.
Blazor does not serialize OnParametersSetAsync against its own outstanding
task, and the modal's `_open` bool cannot tell two opens apart. ResetAndResolveAsync
awaits LoadMergedProbeConfigAsync and ResolveGroupContextAsync and then wrote
_driverType/_mergedConfig/_effectiveDeviceId/_groupPrefix/_canBrowse/_disabledReason
with no staleness re-check, so a close-reopen onto another device let the first
open's late continuation install its device's browse context under the displayed
modal — "Connect & browse" would silently dial the wrong controller and the commit
would land under the wrong tag-group.
Ports the per-open `_openGeneration` counter proven in the two sibling modals
(RawManualTagEntryModal, RawTagModal, 2026-08-05) to every post-await write here:
after both awaits, in the catch's error banner, and in the finally. The finally is
the one departure from the sibling shape and it is load-bearing — this modal resolves
inside a try/finally, so a stale `return` still runs the tail and would drop the LIVE
open's spinner mid-resolve.
Pinned by four RawBrowseModalStaleOpenTests facts, each observed red before the guard
and each re-falsified afterwards by reverting only its own guard. The modal calls
StateHasChanged on the open path, which throws without a render handle, so the test
attaches an inert Renderer subclass (file-local BL0006 pragma); [Parameter]s are set
by reflection per the BL0005 convention.
AdminUI suite 1117/1117 green.
Follow-up to 20a816db: mxaccessgw published the full MxStatusDetail enum
(gateway.md @ 75c71ad, "MxStatus Detail Vocabulary", lifted from
Interop.aaMxDataConsumer.MxStatusDetail), confirming the live-captured codes
(1007 = MX_E_NotWriteable, 1008 = MX_E_WriteAccessDenied). FromCategory now
maps the write-relevant band per code before falling back to the category
bucket: 1004 -> BadOutOfRange, 1005 -> BadTypeMismatch, 1006 -> BadNotReadable,
1007 -> BadNotWritable, 1008/1012/1013/1016/1017 -> BadUserAccessDenied,
1010/1011 (initializing, transient) -> BadOutOfService. 8000
(MX_E_AutomationObjectSpecificError) deliberately stays category-level — the
object supplies its own semantics. Codes <= 0xFF are value-ambiguous with the
OPC DA quality-byte space and keep the quality path (still Bad-family).
Four new named constants, all validated by StatusCodeParityTests against the
pinned SDK. Live refusal smoke re-run green (read-only refusal now surfaces
BadNotWritable instead of generic Bad).
The mxaccessgw OnWriteComplete correlation is live (server b948e69, worker
53f69cd — all four unary write kinds), so every write reply's statuses[0] now
carries the real COM commit outcome. Live-gating the cutover on windev found
and fixed three defects:
- StatusCodeMap.FromMxStatus truncated the correlated failure row's MxStatus
detail code to a byte (1007 & 0xFF = 239 -> OPC DA Good band), mapping a
REFUSED write to Good. New invariant: a row that is not IsSuccess() never
maps Good — details > 0xFF (and contradictory Good-band substatuses) route
by category (SECURITY_ERROR -> BadUserAccessDenied, named const covered by
StatusCodeParityTests; COMMUNICATION/CONFIGURATION keep their codes).
- The borrow-from-subscription-registry write optimization could never commit:
a subscription handle is plain-advised, and MXAccess refuses both the
no-login write through it (SECURITY_ERROR detail 1008, live-captured) and
the supervisory advise upgrade (E_INVALIDARG) — one advise per HANDLE, not
per attribute. Removed: writer subscribedHandleSource ctor param + borrow
branch, GalaxyDriver wiring, SubscriptionRegistry.TryResolveItemHandle +
its forward index and test file. The writer always AddItems its own
supervisory-advised handle, which commits fine while the same attribute is
subscribed (live-proven; pinned by the new write-while-subscribed smoke).
- (gateway-side, fixed there) the correlation initially covered WriteSecured
only, while the dominant no-login path emits plain Write.
Live gate, all legs PASSED against windev: happy write Good off a confirmed
status row with galaxy.writes.unconfirmed + advise_failed both FLAT; read-only
refusal surfaces Bad-family (the #5 node-revert input Galaxy could never
produce before); WriteCompletionWaitMilliseconds=0 control -> provisional Good
+ metered unconfirmed (override removed after, restoration re-verified). New
env-gated smokes: Live_correlated_write_statuses_confirm_commit_and_surface_refusal,
Live_write_to_subscribed_tag_commits_via_own_handle (replaces the borrow
smoke), GatewayGalaxyWriteMeterLiveTests (run filtered — meters are
process-wide).
The connect path was the one capability call the resilience pipeline never
saw, so a driver that could not connect contributed zero breaker input and
the /hosts resilience panel read 'All targets healthy' against a dead
endpoint. InitializeAsync now routes through the invoker as
DriverCapability.Initialize, keyed on the driver-instance id.
- No pipeline retry: the actor's RetryConnect timer is the retry loop
(parser forces retryCount overrides to 0, same funnel as write-shaped).
- Breaker threshold 2, not the Read-sized 5: the breaker samples 30 s and
the reconnect timer fires every 10 s, so at most three attempts fit one
window — threshold 5 could never open. Pinned by its own test.
- An open breaker fast-fails into the same InitializeFailed path as a real
connect failure; the 15 s break window costs at most one extra tick.
The best-effort teardown ShutdownAsync stays unwrapped.
- Companion fix: DriverCapabilityInvokerFactory.Create now invalidates the
instance's tracker snapshots with its pipelines — a respawn during an
outage discarded the only breaker whose OnClosed could clear the OPEN
flag, stranding a false 'Breaker OPEN' chip forever.
Live gate PASSED on docker-dev (image symbol-verified per container):
rtu-gate against dead 10.100.0.35:5021 opened its Initialize breaker on
both centrals with the red 'BREAKER OPEN' chip on /hosts; a socat
5021→5020 forwarder recovered it (both breakers closed within 1 s, panel
back to 'All targets healthy'); healthy MAIN-modbus recorded zero breaker
events. Full suite 50/50 green. Evidence:
docs/plans/2026-08-07-adminui-followups-and-live-gates.md §7.
The two items deferment.md §0.1 named as the remaining untracked engineering.
adminui-followups (19 pending / 0 completed) — CLOSED by the overlap audit the plan
itself asks for first, with no code written. All 19 tasks are shipped, superseded by
v3, or moot. WS1's premise had also expired: it assumed driver-embedded tag lists in
DriverConfig are the runtime source of truth, which v3 ended (the artifact populates
RawTags). Its capability shipped in the v3 shape — TagConfigEditorMap + DeviceFormMap,
both verified reachable from RawTagModal/DeviceModal rather than merely present.
Live gates on docker-dev (image rebuilt: the rig was running a build 3 h older than the
last code commit; parity asserted per container plus a positive check that the symbols
under test are in the shipped DLLs):
- R2-03 T11 PASSED — build 0 errors, Runtime.Tests 555/555, RoslynVirtualTagEvaluator 20/20.
- R2-03 T12 PASSED — VT 42 -> 0x80020000 -> 42, staged WITHOUT a deployment (a Modbus
register flipped on the fixture) so the VT child survived; a deploy-driven trip creates
a fresh child with no last value, which is correct-by-design and masked the result twice.
P7 confirmed: two byte-identical redeploys, zero disturbance.
- driver-reconfigure Task 3 PASSED — a faulted driver adopted a corrected config and
connected with the node's startedAt unchanged and RestartCount=0.
- R2-10 T10 still OPEN, but the recorded cause was wrong. Not ConnectionBackoff
throttling (the breaker opens on 2 failures / 30 s): the dead-endpoint failure mode
never enters the resilience pipeline. IDriver.InitializeAsync is not invoker-wrapped
(DriverInstanceActor.cs:707) and ModbusDriver swallows per-tag failures in BOTH
directions, so even the plan's fallback trigger cannot work. Live-proven: a driver
timing out every read against a blackhole reported ConsecutiveFailures=0 while the
panel read "All targets healthy."
Finding from T12: the Bad publish's documented "carrying the last-known value" is
unobservable by any client — subscription and a raw Read service call both return NULL.
The actor is correct; every existing failure assertion checked StatusCode alone, so the
value half was asserted nowhere. Adds
VirtualTagActorTests.The_Bad_result_carries_the_last_known_value, which proves the actor
emits it. The documentation is what overstates.
A CNC can report duplicate sibling names — the wonder Makino returns two
axes both called AA — which made the Axes/ subtree untrustworthy: the
browser offered two identical folders the commit gate had to refuse, and
the poll loop's name-keyed snapshot cache let the second axis silently
overwrite the first. Spindles and servo loads had the same collision in
their own keying, plus a resolve-to-first-match on the spindle read path.
Scheme: every member of a collision set gets a 1-based occurrence-ordinal
suffix in slot order (AA_1, AA_2); unique names stay bare. Chosen over the
alternatives deliberately:
- Names, not slot numbers, stay the primary identity — the segment map is
re-derived from the live device each session, so a name-keyed path
follows the physical axis across config changes, where a slot-qualified
path would silently rebind on any axis-list shift and rename every axis
on every machine for one pathological CNC.
- The WHOLE collision set is suffixed: bare AA ceases to exist when
ambiguous, so a tag authored against it reads BadNodeIdUnknown rather
than one arbitrary axis's data — the #534/#540 silent-wrongness class.
- Occurrence ordinal, not absolute slot: inserting/removing an unrelated
axis never renames an existing duplicate pair. If the collision-set
membership itself changes, segments change on the next browse and stale
authored segments read Bad, never another axis's data.
One pure helper (FixedTreeNames.Disambiguate, collision-proof against
genuine names by suffix escalation) feeds FocasFixedTreeCache.AxisSegments
/ SpindleSegments, and the emitter, axis snapshot keys, servo-load keys
(cnc_rdsvmeter answers in slot order, so the k-th AA load belongs to the
k-th AA axis) and the spindle read resolution all consume those — they can
no longer disagree. The nameless-spindle S{n} fallback applies before
disambiguation so a fallback name colliding with a real one is handled
like any other duplicate.
Tests are red-first on the pre-fix driver (all five driver-level tests
fail on master's code): browse emits AA_1/AA_2 and no bare AA, each
duplicate reads its own position and servo load, duplicate spindles
resolve per slot, bare AA reads Bad. The wire half runs live against
focas-mock (duplicate names survive the managed wire parse and publish as
suffixed segments); the mock's dynamic store is name-keyed so the
distinct-values half stays with the offline tests. FOCAS 327/327,
integration 11/11, FOCAS.Cli 52, AdminUI 1113, ControlPlane 231.
Closes#548.
Claude-Session: https://claude.ai/code/session_014WNM4vjoVksyyBraTXSZE1
The two discovery-path libplctag tags (@tags symbol-list enumerator, @udt
template reader) still passed the bare Gateway host, so a browse against a
device authored on a non-default port silently browsed whatever answers on
44818 — #534's defect surviving in its discovery half. Both sites now consume
AbCipTagCreateParams.LibplctagGateway, with the Tag construction extracted
into internal static builders so the seam is unit-assertable (the #534 tests
pinned the property's composition but nothing pinned each site's consumption,
which is exactly how these two survived).
New guard: LibplctagGatewayConsumptionGuardTests scans both AB drivers'
sources and fails on any libplctag Tag built from the bare Gateway, so a
future call site cannot reintroduce the silent 44818 redirect.
Live-gated against the fixture host: browse of ab://10.100.0.35:44818
returns the nine-tag symbol table; browse of ab://10.100.0.35:44999 (nothing
listening — the issue's exact repro) now fails ErrorTimeout where it
previously returned the 44818 fixture's table.
Closes#539.
Claude-Session: https://claude.ai/code/session_014WNM4vjoVksyyBraTXSZE1
#542 filed FleetStatusBroadcaster as blind to site nodes since the Phase 6 per-cluster mesh
split: it derives "the fleet" from central's own Akka gossip ring, which after the split is
just central's own pair. Scoping the cross-mesh rewrite found the deeper truth — the whole
chain is dead end to end, so this deletes it rather than reviving it.
- Zero consumers. FleetStatusBroadcaster published FleetStatusChanged on the DPS
"fleet-status" topic; FleetStatusSignalRBridge forwarded it to FleetStatusHub
("/hubs/fleet-status"); nothing anywhere connected. The AdminUI bans self-connecting
HubConnections, no browser JS subscribes, and /fleet (Fleet.razor) reads
NodeDeploymentState from the ConfigDb plus IClusterRoleInfo on a 5 s poll instead.
- Its second input never existed. DriverHostStatusHeartbeat had no production sender —
only the broadcaster's own test told it — so CurrentRevision was always null and the
30 s staleness -> Degraded flip could never fire.
- Post-Phase-6 DPS is pair-local, so even the central-only view it did produce was an
accident of topology rather than a design.
The fleet visibility that actually works is unchanged and already cross-mesh-correct:
/fleet over NodeDeploymentState, and the four Phase 5 per-node telemetry dials.
Precedent for deleting inert machinery rather than activating it: #522 (driver stability
tiers), #507 (discovered-node injection), #525 (GetMemoryFootprint).
Deleted: FleetStatusBroadcaster (+ its nested DriverHostStatusHeartbeat), its admin
singleton registration and FleetStatusBroadcasterKey, FleetStatusSignalRBridge and its
key, FleetStatusHub and its route mapping, and the FleetStatusChanged / FleetNodeStatus /
FleetNodeHealth messages. Both Fleet/ directories survive — ControlPlane keeps
ClusterNodeAddressReconciler, Commons keeps GetDiagnostics. Docs swept where they claimed
a fleet-status channel exists (Telemetry.md's Phase 5 non-migrated channels,
Configuration.md, CLAUDE.md); completed-program plan docs under docs/plans and docs/v2 are
left as the historical record they are.
Build 0 errors, 868 warnings — byte-identical to master's rebuild baseline.
ControlPlane.Tests 231/231, AdminUI.Tests 1109/1109.
Claude-Session: https://claude.ai/code/session_014WNM4vjoVksyyBraTXSZE1
DriverStatusPanel was orphaned — zero references repo-wide — so the Restart and
Reconnect driver commands had NO reachable UI at all, despite working end to end
over the new gRPC mesh transport. Nothing failed to build, test or deploy while
that was true, which is why it went unnoticed.
The panel is now reachable per driver row on /hosts, behind a per-row "Controls"
expander that is collapsed by default: the table stays scannable, and only the
opened panels hold snapshot-store subscriptions. Expansion state is a HashSet
keyed by DriverInstanceId held on the page, so it survives the 5 s refresh that
replaces every row record. Both rows are @key'd by instance id so the panel —
which binds its DriverInstanceId once in OnInitializedAsync — can never be diffed
onto a different driver.
The panel's Enabled parameter is fed from real config rather than defaulted:
HostsDriverInstanceInfo/HostsDriverRow now carry DriverInstance.Enabled, which
LoadConfigAsync already had in hand. A snapshot with no matching configured
instance is treated as enabled — the row exists only because live health arrived,
so "Disabled — not deployed" would contradict the same row.
The expander is visible to every AuthenticatedRead viewer; the panel self-gates
only its Reconnect/Restart buttons on DriverOperator.
A source-scan pin (HostsPanelWiringTests) asserts /hosts still renders the panel
and still binds it to the row's own driver, cluster and enabled state — removing
the tag would silently restore the orphaned state.
Claude-Session: https://claude.ai/code/session_014WNM4vjoVksyyBraTXSZE1
DeadlineElapsed_for_current_deployment_marks_TimedOut carried two races,
both test-side. ConfigPublishCoordinator itself is untouched.
(a) The NodeDeploymentState row was inserted AFTER Sys.ActorOf. The
coordinator's PreStart reads NodeDeploymentStates to recover the
expected-ack set, and it runs as soon as the actor is scheduled — so if
PreStart won, the expected-ack set was empty and the coordinator could
short-circuit to Sealed instead of ever reaching TimedOut. The row is now
seeded before the spawn.
(b) The applyDeadline was 150 ms of real wall clock, and it arms a real
Akka timer. Under load that timer could fire on its own and race the
test's explicit DeadlineElapsed tell — so the test was intermittently
exercising the timer rather than the message handler it names. The
deadline is now TimeSpan.FromMinutes(5), matching the other two tests in
this class, which makes it un-hittable inside the test's own 3 s
AwaitAssert bound and leaves the explicit tell as the only deadline
signal the coordinator ever sees. That tell, and the AwaitAssert bound,
are unchanged.
Verified: the single test 5/5 green in a row, then the whole class 3/3.
Claude-Session: https://claude.ai/code/session_014WNM4vjoVksyyBraTXSZE1
TelemetryStreamAuthInterceptor, ConfigServeAuthInterceptor and
LocalDbSyncAuthInterceptor all decided "is a key configured?" with
string.IsNullOrEmpty. A whitespace-only configured value therefore passed
that guard and fell through to the constant-time comparison — where it is
satisfied by a whitespace-only bearer token. The gate would be open to
anyone who guessed a blank credential.
This backports the check the two NEW mesh interceptors already use:
MeshCommandAuthInterceptor (and its central-side clone) validate with
string.IsNullOrWhiteSpace, so the four/five planes now answer the same way
to the same misconfiguration. Nothing else changes — the denial status
codes (PermissionDenied / Unauthenticated / PermissionDenied respectively)
and every denial message are untouched, and no key material is logged.
No deployment configures a whitespace key: the docker-dev rig and
production both set real keys, so this is a fail-closed hardening of an
unreachable-today misconfiguration rather than a live exposure.
Tests: each of the three test files gains a ..._WithBlankKeyConfigured_IsDenied
fact mirroring MeshCommandAuthInterceptorTests.ExecuteMethod_WithBlankPskConfigured_IsDenied
(configured key " ", presented "Bearer "), verified red before the fix.
ConfigServeAuthInterceptor and LocalDbSyncAuthInterceptor also gain the
HasExactlyOnePublicConstructor pin their two siblings already had —
Grpc.AspNetCore silently stops invoking an interceptor with more than one
public constructor, which would disable auth without failing anything.
Claude-Session: https://claude.ai/code/session_014WNM4vjoVksyyBraTXSZE1
MeshPaths.CentralCommunication / .NodeCommunication (the "/user/..." full-path constants) had
ZERO production consumers after the ClusterClient deletion: both gRPC services resolve their
comm actor through ActorRegistry by typed key, never by ActorSelection over a path. Only the
path-probe test still read them, while the class doc went on calling them "the wire contract
of the central↔node boundary" — a claim that stopped being true the moment nothing addressed
an actor across a process boundary by name.
- The two full-path constants are deleted; the two Name constants remain.
- MeshPaths' doc now says what it is (the local actor-naming convention the registry-keyed
services are spawned under) and records what it stopped being, so the next reader does not
re-derive a contract from the file's history.
- MeshCommActorPathTests composes "/user/" + Name itself — the absolute path is that test's
own addressing detail now, not something the product declares. Renamed
Both_mesh_comm_actors_exist_at_their_contract_paths ->
..._are_spawned_under_their_declared_names, because what it still pins is that the two
registration sites actually ran on a real booted node; a rename is now a compile error, so
"contract" was the wrong word for what would break.
- NodeCommunicationActor's summary no longer cross-references the deleted constant or calls
the name an address. CentralCommunicationActor was checked and makes no such claim.
MeshGrpcBoundaryTests replaces the MeshClusterClientBoundaryTests deleted in step 1. Three
self-seeded single-member clusters sharing the ActorSystem name, joined ONLY by gRPC, driving
the REAL CentralCommunicationActor -> MeshGrpcCommandDialer -> Kestrel h2c ->
MeshCommandAuthInterceptor -> NodeCommandGrpcService -> NodeCommunicationActor -> EventStream:
1. No_cross_mesh_Akka_association_forms_between_central_and_either_site — the program's own
point, kept as a test rather than a live-gate note: each cluster stays single-member and
none ever holds another's address as a member.
2. One_dispatch_through_central_reaches_a_node_in_every_cluster_mesh — the fan, with each
site holding its OWN per-cluster PSK, so a dial that selected the wrong cluster's key
would be refused at the far side.
3. Falsifiability_control_a_dial_carrying_the_wrong_cluster_key_delivers_nothing — without
it, (2) cannot distinguish "the gRPC boundary carried it" from "something else in the
process did". The same node then accepts the right key, so the silence is the auth gate
and not a dead endpoint.
All three run fully offline — no Docker fixture — and pass.
Comment sweep across every remaining clusterclient|receptionist hit in src/ and tests/:
the program label becomes "mesh gRPC command plane", and every comment that described the
deleted transport as live now describes what actually runs (Dps = pair-local default,
Grpc = cross-mesh). mesh_transport.proto's header/service comments are reworded; no message,
field or enum is touched — the contract is locked. MeshCommActorPathTests is reshaped rather
than retired: the two paths still exist and still matter, but what a mismatch costs is now a
misleading Unavailable from the gRPC ingress rather than a silently dropped receptionist send.
The only surviving mentions of the old names are the ones that must survive: the fail-closed
rules that name "ClusterClient" as a retired value refused at boot, their tests, and this
test class's own account of what it replaces.
MeshTransportOptions.ModeClusterClient is gone. MeshTransportOptionsValidator accepts Dps or
Grpc and nothing else, so a node left carrying the string "ClusterClient" is now REFUSED at
boot naming MeshTransport:Mode and both remaining values — deliberately fail-closed rather
than falling back to Dps, which on a split-topology node would deliver nothing at all. A new
fact pins exactly that.
SplitTopologyTransportValidator tightens from an allow-list of two to Grpc only, as its own
comment said it would once ClusterClient was deleted. Dps on a cluster-role node stays
refused; the rule is still an explicit allow-list, so a typo'd mode fails too.
Tests: the two ClusterClient-mode facts in MeshTransportOptionsValidatorTests are deleted
(their Dps siblings already assert the same rule for the one remaining non-Grpc mode), the
SplitTopology fixtures move to Grpc, and the three-mode Theories in
ControlPlaneStarterOrderTests / CentralCommunicationActorTests become two-mode.
The property, the four validator rules that graded it (empty / non-akka.tcp / actor-path
suffix / one-bad-among-good) and all twelve docker-compose entries (6 nodes x 2) are gone.
The compose comment blocks that explained "/system/receptionist is appended at construction
time" and the client's contact rotation go with them, and the surrounding prose now
describes the Grpc command plane it actually runs.
CentralGrpcEndpoints' doc no longer says it "replaces" the Akka contact points — there is
nothing left to replace — and ContactRefreshSeconds' failure message names the node
dispatch set rather than the ClusterClient contact points.
`rg CentralContactPoints src tests docker-dev` is now empty.
akka.extensions loses ClusterClientReceptionistExtensionProvider and the whole
akka.cluster.client block (buffer-size / reconnect-timeout / receptionist.role) goes with
it. The DistributedPubSub extension and the maximum-frame-size / log-frame-size-exceeding
canary both stay — DPS is the pair-local transport and the frame canary guards every
remoting message, neither of which the cross-mesh deletion touches.
MeshTransportHoconTests drops its three ClusterClientSettings/ClusterReceptionistSettings
asserts and keeps the frame-size pin and the DPS-extension pin, renamed since it no longer
has a receptionist to survive alongside.
WithOtOpcUaRuntimeActors no longer builds a ClusterClient from
MeshTransport:CentralContactPoints, and NodeCommunicationActor.Props takes only the gRPC
IApplyAckSender. The ackRouter gate narrows from `useClusterClient || useGrpc` to `useGrpc`,
and the CentralClusterClientName constant is gone.
NodeCommunicationActor's no-transport warning now names MeshTransport:CentralGrpcEndpoints
and MeshTransport:Mode only — the old text pointed an operator at a key that no longer
exists.
Ripples in this step: the two ClusterClient-shaped facts in NodeCommunicationActorTests
(Send-not-SendToAll, and sender-wins-over-a-client) are deleted — the first has no subject
left, the second guarded a both-transports-present state that is now unconstructible. The
Runtime ServiceCollectionExtensions fact keeps its ackRouter pin and drops its
"no ClusterClient actor exists" half for the same reason.
IMeshClusterClientFactory/DefaultMeshClusterClientFactory are gone, along with
CentralCommunicationActor's _clients map, RebuildClient, the SendToAll fan and the
receptionist-address half of the contact load. LoadContactsFromDb SURVIVES — it feeds the
gRPC dialer's endpoint resolution and the known-cluster gate — but now projects only the
columns those two need (the AkkaPort/ActorPath.TryParse parse and its Malformed list went
with the receptionist addresses).
Ripples fixed in this step because the build must be green at every commit:
- ControlPlane/ServiceCollectionExtensions drops the factory argument.
- CentralCommunicationActorTests: the six ClusterClient-only facts are deleted (each has a
Grpc counterpart that already asserted the same property independently), the three
transport-agnostic facts move to Grpc mode, and the Dps fact keeps its own coverage.
- MeshClusterClientFactoryTests deleted (its subject is gone).
- MeshClusterClientBoundaryTests deleted here rather than in step 7: it drives
DefaultMeshClusterClientFactory directly, so it cannot compile past this commit. Its
replacement, MeshGrpcBoundaryTests, lands in step 7 and keeps the load-bearing assertion
(two separate meshes, no cross-mesh Akka association).
Follow-up to a94eb8a3, addressing CHANGES REQUESTED on code review. Production
logic was confirmed correct on all spec points; this commit fixes the review's
test/comment findings only.
IMPORTANT-1: A_permanently_unresolvable_pskProvider_is_reported_once_per_incarnation_not_once_per_refresh
did not prove the second refresh ran — its Expect(0) window gated on a dialer
dispatch landing, but LoadContactsFromDb is fire-and-forget and the first
block's AwaitCondition already re-Tells MeshCommand per poll, so the dialer
call count could reach the threshold before PublishKnownClusters was ever
re-invoked by the second load. Fixed by having the PSK factory count its own
invocations and gating both AwaitConditions on that count directly (pskCalls
>= 2 proves the first retry; pskCalls > pskCallsAfterFirstLoad, awaited INSIDE
the Expect(0) window, proves the second one). Bite-checked: with
_pskProviderAbsenceReported's guard temporarily removed, the test now
correctly fails ("Received 1 message too many... received 1 that matched
filter"); guard restored and reverified green.
MINOR-2: reworded four comments (CentralCommunicationActor.cs ~109, ~259,
~305-309, ~853) from narrating the change ("this fix removes", "already
existed, unchanged", "added by this fix", "the old two-call-site shape") to
stating the timeless invariant.
MINOR-3: added A_supplied_dialer_factory_that_always_throws_still_drops_each_command_loudly_while_PreStart_stays_quiet
— the symmetric case to the existing never-supplied-dialer test, exercising
Resolve's catch + firstUse:true path (a factory that WAS supplied and always
throws) rather than the null-factory path. Asserts PreStart plus the first
contact load stay silent and DispatchOverGrpc's existing per-dispatch Error
still fires on every dispatch, unguarded.
MINOR-4: "Check the ControlPlane registration wires IMeshPskProvider" ->
"Check that the ControlPlane registration wires up IMeshPskProvider"; no test
pinned the old substring, so no test update was needed for this one.
Skipped item 5 (restart-re-arms pin) per instruction — out of scope.
Claude-Session: https://claude.ai/code/session_014WNM4vjoVksyyBraTXSZE1
PreStart's eager resolve of IMeshCommandDialer/IMeshPskProvider loses a
millisecond-wide race against the Akka builder's remaining starters on
essentially every admin boot (live gates: 1A 2/3 boots, 1B 4/4 boots) and
self-heals at the memoized Dialer/PskProvider properties' first use. Before
this change that produced two FALSE Errors per healthy boot, which made a
soak error-grep on a healthy rig impossible.
Resolve() now takes a `firstUse` flag and logs at Debug on BOTH call sites
(PreStart and the memoized properties' retry) — from inside Resolve a lost
race and a genuinely broken factory are indistinguishable, so severity is
decided by the CONSUMER of a still-null result after the retry, not by
Resolve itself:
- Dialer: DispatchOverGrpc's existing per-dispatch Error ("no command
dialer is wired") is unchanged and is now the ONLY loud signal for that
dependency.
- PskProvider: PublishKnownClusters gained a new Error carrying the old
PreStart guard's diagnostic content (gate never published, cached keys
never dropped, rotated key needs restart), guarded to fire ONCE per
actor incarnation via `_pskProviderAbsenceReported` — that handler runs
on the contact-refresh timer for the actor's whole life, so an
unguarded Error there would repeat every ContactRefreshSeconds forever.
Design choice for the "factory never supplied" case (Dps/ClusterClient test
constructions that pass no factories at all): kept silent at BOTH call
sites for both dependencies, same as before. For the dialer this was
already true (DispatchOverGrpc's check doesn't care why the dialer is
null). For the PSK provider the new once-guarded Error is gated on
`_pskProviderFactory is not null` specifically so the many existing
ClusterClient/Dps facts in CentralCommunicationActorTests that never wire a
PSK provider (using the bare Spawn helper) keep passing unmodified — a
never-supplied factory is a deliberate wiring/test choice, not a resolve
failure, and Resolve() already stayed silent about it for the same reason.
Two new tests pin the net behavior: a healthy boot (factories that throw
once then succeed) produces zero Warning/Error end-to-end through PreStart,
the first contact load and the first dispatch; a permanently-broken PSK
provider logs the absence exactly once across two contact loads, not once
per refresh.
Claude-Session: https://claude.ai/code/session_014WNM4vjoVksyyBraTXSZE1
CRITICAL. The ack deadline was one absolute DateTime computed once for the whole
endpoint walk. Against the characteristic WAN failure — a central that HANGS
(SYN black-hole, frozen host) rather than refusing — endpoint 1 burned the whole
AckTimeoutSeconds and endpoint 2 was then dialled with a deadline already in the
past: instant DeadlineExceeded, no packet leaving the node, the surviving central
never really tried. Failover looked like it ran and did not. It also contradicted
MeshTransportOptions' own documented "applied per endpoint, not per ack". The
deadline is now computed fresh inside the loop; headers stay hoisted (read-only,
sequential walk).
Two test gaps that let that survive are closed: nothing pinned the ackRouter gate
(narrowing it back to ClusterClient-only compiled and passed every test while
dead-ending every Grpc ack to DPS), and the deadline fact's lower bound accepted
any positive timeout at all. Both new pins are mutation-verified.
Docs corrected where they overstated the case: duplicate acks CANNOT mis-seal a
deployment — ConfigPublishCoordinator.HandleAck keys _acks by NodeId behind an
expected-set guard, so a duplicate overwrites its own entry. The single-endpoint
rule is defence in depth and a bound on wasted work, not a correctness
dependency; and the walk is at-least-once on the deadline path, not at-most-once.
Task 1B.2 — the node→central half of the ClusterClient→gRPC migration.
GrpcApplyAckSender walks MeshTransport:CentralGrpcEndpoints in order and
STOPS AT THE FIRST SUCCESS. That single `return` is the whole single-count
guarantee: the transport it replaces was ClusterClient.Send, deliberately not
SendToAll, because both central nodes forward to the SAME
ConfigPublishCoordinator singleton proxy — an ack posted to both would have the
coordinator count one node twice and could seal a deployment on a doubled count.
The list is an availability list, not a fan-out list.
No retry beyond that one ordered pass, no queue: the receptionist buffered
nothing and retried nothing, and this is like-for-like, not an upgrade to
delivery guarantees. All endpoints failing is still a DROP — but a loud one,
naming the node, deployment, correlation id and the status each endpoint
answered with, which is what separates "central is down" from "central rejected
this node's key". The bearer never reaches a log; only endpoints and status
codes do. Per-endpoint GrpcChannel cache and the endpoint loop clone
GrpcDeploymentArtifactFetcher; the cluster id in x-otopcua-cluster comes from
this node's own cluster-<id> role via RoleParser, off the same bound
Cluster:Roles ClusterRoleInfo reads.
SendAsync never throws: the caller discards the task, so an escaping exception
would become an unobserved fault — a drop with no log line, the exact failure
mode this migration removes.
NodeCommunicationActor gains an OPTIONAL ackSender; every existing test compiles
and passes unedited. It is checked before the ClusterClient so the two can never
both fire. The registration arm creates no ClusterClient under Grpc.
NOTE — one wiring change beyond the actor: DriverHostActor's ackRouter is now
`useClusterClient || useGrpc ? nodeComm : null`. Left at ClusterClient-only, a
Grpc node's ack would have gone out on the DPS deployment-acks topic and never
reached the sender at all.
Two review items from Task 1B.1, both diagnostic:
1. CentralCommandGrpcService's "actor not ready" Warning omitted the peer and
the correlation id while the decode-failure Warning six lines above carried
both. These are the only two ways an ack fails to reach the coordinator, and
an operator has to correlate either against the node's own log. Decoding has
already succeeded on this path, so the DECODED correlation id is free.
2. A fused driver,admin,cluster-MAIN node under Mode=Grpc owes BOTH validator
surfaces at once (driver keys AND ConfigServe:GrpcListenPort). The two blocks
are independent `if`s; chained as `else if` a fused central node would be
graded on one surface only and boot with nowhere for any ack to land. Pinned
from both sides plus the fully-configured pass.
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.
Joseph Doherty