DriverInstanceDiagnostics.Name carries the string driver-instance id
(what /hosts, telemetry and the artifact key on) and DriverInstanceId
the DB row Guid; the wire record has no slot for the authored display
name. The write-cap test asserts entry order per slot pair instead of
completion order of two concurrent writes.
Refs #731
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
DriverHostActor wraps the health publisher it hands its children in a
ChildHealthLedger (last snapshot per driver id, forwarded to the real
publisher), so GetDiagnostics now answers with each child's row id,
authored name, last-published state, device counts (host statuses when
the driver probes hosts, else one device) and last-change time — instead
of Guid.Empty / "Spawned" / 0 / now. Entries are forgotten on StopChild
and operator Restart.
IFleetDiagnosticsClient gains its first consumer: the new
/fleet/diagnostics page (node selector + ?node= query, revision,
running-from-cache notice, per-driver table). FleetDiagnosticsClient
logs the degrade-to-empty path at Warning.
Review ref: 02/U1.
Refs #731
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
- DriverHostActor.StopChild removes the driver's _lastReprovisionUtc stamp
(and its deferred-reprovision timer): the map grew per removed id and a
re-added driver inherited a backoff it never earned (02/L2).
- Stale now registers explicit drops for AttributeValueBatchPublished,
AttributeAlarmPublished, ArtifactForApply and ApplyWedged instead of
letting them hit Unhandled (02/H5).
- DbHealthProbeActor runs its SELECT on the thread pool and pipes the
outcome back (single-flight), with a 2 s command timeout on relational
providers, so a hung SQL Server no longer pins the actor thread and the
cached GetStatus answer stays reachable (02/P7).
- TelemetryLocalHub eviction tombstones are timestamped and swept past a
10 min TTL on the next Evict; a TimeProvider ctor is the test seam (02/L4).
- Tests: ValuePublishes.Single helper replaces the per-value
AttributeValuePublished in CalculationDependencyFlowTests (02/U4, test
half; the production receive is owned by Lane E).
- R05/8 audit: PollGroupEngine.Unsubscribe has no production caller; every
driver already uses UnsubscribeAsync (no change).
Refs #731
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
Clears CommentChecker findings in src/ (3,280 -> 209, the remainder being false
positives like PLC-5/IEEE-754/M0.0) by adding missing summaries/params/returns,
normalising inheritdoc usage, and removing Gitea/review tracking IDs while keeping
the rationale they annotated. Comment-only: comment-stripped sources are identical
to the parent commit.
- HistoryUpdate Primary gate now reads IHistoryUpdateRoleView (PrimaryGatePolicy.ShouldServeHistoryUpdate):
the data-plane verdict but CLOSED until this node's first redundancy snapshot, so both pair nodes cannot
accept in the pre-join boot window. Single nodes open on their own singleton's first snapshot; admin-only
nodes register no view => no gate. DriverHostActor publishes it at every Primary-verdict publish point.
- SqlServer v1: the updater throws NotSupportedException on every arm; the node manager maps it to a
node-level BadHistoryOperationUnsupported (was per-item under a Good node for UpdateData/DeleteAtTime).
HistoryWrite bit still advertised on v1 (not knowable at materialise time) — documented.
- The update plan is evaluated before the Primary gate: an unservable request (Remove, IsDeleteModified,
not historized, no updater) is BadHistoryOperationUnsupported off-Primary, not BadInvalidState.
- Local-kind SourceTimestamp converted to UTC before the future-skew compare and sent as UTC.
- Hosted-service wiring extracted to BuildHistoryUpdatePosture and tested (gate source, skew, clock).
- Docs: Historian.md (boot window, ordering, node-level v1, HistoryWrite bit), CLAUDE.md.
Co-Authored-By: Claude <noreply@anthropic.com>
A SIGKILLed Primary stays an unreachable-but-not-downed member until
auto-down, and the redundancy singleton usually dies with it, so the
standby's cached snapshot kept naming it Primary and nobody captured for
~30 s. DriverHostActor now holds its own ReachabilityEvent (+MemberUp)
subscription, caches the last snapshot, and excludes unreachable peers
from the historization peer-is-Primary predicate only, republishing just
the capture verdict on a reachability change. Alarm drain and the Primary
data-plane verdict are unchanged. Gated by
ContinuousHistorization:CaptureWhilePrimaryUnreachable (default true).
Tests: message-driven cases with real ReachabilityEvents (incl. a
falsification control on the drain/data-plane verdicts) plus a real
two-node cluster crash test with downing disabled.
Since #516 a DriverConfig change is stop + respawn, so DriverSpawnPlan.ToApplyDelta
was always empty and the in-place path had no producer. #663 and #664 were both
latent bugs in it. This deletes it:
- host: ApplyChildDelta, _pendingDelta, HandleApplyResult and HandleDeltaApplied
- plan: DriverSpawnPlan.ToApplyDelta
- actor: the ApplyDelta, ApplyResult and DeltaApplied messages, the Connected
reinit handler and the mid-connect AdoptConfigDuringInit
- tests: the actor tests that drove the seam
IDriver.ReinitializeAsync and each driver's implementation and unit tests stay.
Comments and docs that cited AdoptConfigDuringInit as the reason a driver must
release prior state at init now cite the IDriver contract and overlapping
retries, which still apply.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
After an OpcUaClient respawn an upstream inactive+unacked condition went out as
Retrigger+Clear (duplicate Activated+Cleared rows, Active flicker), and a
first-sight inactive+acked one as Unspecified (an Activated row for an inactive
alarm; a retained-Active host condition never cleared).
- New AlarmTransitionKind.InactiveUnacknowledged: one projection, rows only for
what changed (Activated+Cleared from acked, Cleared from active+unacked, none
when already inactive+unacked).
- OpcUaClient uses it for moves into inactive+unacked; first sight of an
inactive condition is InactiveUnacknowledged or Clear+Acknowledge, flagged.
- Projector: a flagged Acknowledge on an already-acked condition writes no row.
- E2E tests through DriverHostActor.ForwardNativeAlarm (red-proven).
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
- AlarmTransitionKind.ActiveAcknowledged (appended): active AND acked applied as ONE projection.
Galaxy ActiveAcked snapshot entries and OpcUaClient inactive->(active,acked) use it; the
projector writes 'Activated' only if not already active, 'Acknowledged' only if not acked.
- AlarmEventArgs.IsRestatement (optional, default false): set only by Galaxy snapshot entries
and OpcUaClient first sight. A flagged Retrigger on an Active+unacked condition with the same
message and severity, or a flagged Clear on an inactive condition, writes no /alerts row.
- NativeAlarmProjector reports the history rows per transition; ForwardNativeAlarm writes one
row per entry. Unflagged sources (FOCAS/MTConnect/AbCip, live Galaxy) are unchanged.
- Tests: projector rules each with a must-write control; OpcUaClient transition table;
Galaxy feed/driver mapping; Galaxy + OpcUaClient end-to-end through DriverHostActor.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
MAJOR-1: a deploy that changes a driver's config respawns it, and the fresh instance's empty told
record re-sent Raise for every still-active alarm, un-acking acknowledged conditions and writing a
duplicate "Activated" alerts row. NativeAlarmProjector now treats Raise on an already-active
condition as a re-statement (keeps Acked; ForwardNativeAlarm skips the alerts row); only Retrigger
(or inactive->active) un-acks. Sources whose upstream ack state is authoritative send Retrigger for
an unacked re-statement: OpcUaClient on first sight (no prior told), Galaxy for an unacked active
snapshot entry. Runtime E2E drives a FRESH FOCAS driver after a redeploy and asserts the ack survives.
MAJOR-2: a failed FOCAS alarm read annotates each subscribed condition BadNoCommunication once per
outage (#626 quality-only path) with one Warning (MINOR-6), lifted to Good on recovery; failed alarm
reads degrade driver health with LastSuccessfulRead frozen (successful ones advance it), so an
alarm-only driver trips the #527 comms-loss watchdog.
MINOR-3 CLAUDE.md #630 bullet; MINOR-4 split line; MINOR-5 a non-empty cnc_rdalmmsg2 payload that
fits neither stride throws; MINOR-7 value-tag + parsed-address tables are built and swapped, never
cleared in place under the retained poll engine. Docs corrected (FOCAS.md, AlarmTracking.md, CLAUDE.md).
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
60 s labelled a slow-but-clean large rebuild degraded. The deadline moves to a
shared Commons DeployTimings constant that ConfigPublishCoordinator reads, and
the node's timeout is deadline - 15 s margin, so the held ack still lands in
time. A timed-out ack's reason now says the outcome was not reported in time,
distinct from a real sink failure.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
The apply writes the node's row Applied/null before the address-space outcome
is known; central adds the reason only when it ACCEPTS the ack, so a discarded
(stale/timed-out) ack left a fused node's row reading clean. The prefix moves
to ApplyAck.DegradedReasonPrefix so both writers spell the row identically.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
DriverHostActor no longer acks Applied the moment the revision advances. It
asks OpcUaPublishActor for the outcome (RebuildAddressSpace.ReportOutcome) and
acks Applied+Degraded(reason) when the apply was degraded; a missing outcome
times out (60 s, inside the coordinator's 2 min deadline) to a degraded ack,
never a clean one. Re-dispatching a revision whose apply was degraded or is
still unconfirmed re-applies instead of taking the clean idempotent
short-circuit - that re-apply is also what repairs the degraded nodes.
Tests with a bare publish probe use a clean-outcome auto-pilot probe.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
#641 keeps the projector's prior state across a deploy instead of resetting it to the
acked default. The projector never saw a client's ack (the SDK applies it to the node and
the host only routes it), so a Clear after ack+deploy now projected Acked=false and
un-acknowledged the operator's ack - previously masked by the per-deploy reset.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
DriverHostActor cleared NativeAlarmProjector's per-condition prior state on
EVERY apply, so after any deploy a driver Acknowledge of a still-active alarm
projected Active=false (writing it inactive) and a Clear of an unacknowledged
alarm projected Acked=true (silently acknowledging it). The projector is now
pruned to the conditions the new composition still carries (RetainOnly).
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
- Unavailable keeps an ACTIVE condition active: quality-only BadNoCommunication; on data
return same statement => quality-only Good, Normal => Clear, different => Retrigger.
- A changed alarm reference set (alarm-bearing delta = full rebuild) clears the whole
last-told record so every active condition is re-asserted; the same set keeps it.
- Quality-only events go through the host's AlarmQualityUpdate path, never a projected
snapshot (which could ack a retained unacked condition or rewrite message/severity).
- Deterministic tie-break between equal-level conditions (ordinal nativeCode, then type).
- Dedup key is level+nativeCode+nativeSeverity+qualifier; a text-only change updates nothing.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
AlarmEventArgs gains optional OpcUaSeverity (supersedes the four-bucket mapping, clamped
1..1000) and QualityStatusCode (projected onto the condition's Quality). An Unspecified-kind
event carrying a quality is a quality-only annotation: the host updates the condition but
publishes no alerts row. Both default null, so every existing native source is unchanged.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
DriverHostActor.StopChild released the tracker (Invalidate) and evicted the hub synchronously, but
Context.Stop is asynchronous: the stopping child's in-flight SubscribeBulk completed through its
capability invoker AFTER the release, and its RecordCallStart/RecordSuccess/RecordCallComplete
re-created the tracker entry, which DriverResilienceStatusPublisherService then re-emitted every tick
forever. Confirmed with a temporary stack capture in the tracker (removed): every post-Release write
came from DriverInstanceActor.HandleSubscribeAsync -> CapabilityInvoker.ExecuteAsync; the test failed
whenever it landed before the second publisher tick.
- DriverResilienceStatusTracker.Retire (Release) drops the id's snapshots and ignores every later
Record* for it; one guarded write path re-checks after the upsert so a write racing Retire cannot
survive it. Reinstate (called by DriverCapabilityInvokerFactory.Create) lifts it, so the same id
re-added in a later deployment records again.
- TelemetryLocalHub.Evict now also holds the id (late Health/Resilience dropped, neither cached nor
fanned out); ITelemetryLocalHub.Readmit (default no-op), called at the top of SpawnChild, lifts it.
- The removal test is now deterministic: an invoker decorator holds the child's subscribe inside the
real invoker's bracket across the removal and signals when its late writes are done (positive
control) — red 5/5 without the fix, green 20/20 with it. New unit tests for the tracker guard
(incl. a Retire race), the factory Release/Create pair, the hub hold/readmit, and a re-add-same-id
host test.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
The drain gate's 'is my queue-sharing peer Primary?' used only the host of
LocalDb:Replication:PeerAddress, which the passive half (ApiKey +
SyncListenPort) does not set - so as Secondary it kept draining beside its
Primary partner and every alarm event reached the historian twice.
DriverHostActor takes a sharesAlarmQueueWithLocal predicate used when no
peer host is known; WithOtOpcUaRuntimeActors builds it
(BuildAlarmQueuePeerPredicate) for a SharesRowsWithPeer node with no
PeerAddress as 'a member of my own cluster-{ClusterId} mesh other than
me'. A node sharing no rows, or with no mesh role, gets none and keeps
draining. AlarmHistorian.md / Redundancy.md / pair-replication runbook
updated.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
MINOR-3: new DriverHostActorPrimaryRoleSourceTests case drives the
deny flip with a real ClusterEvent.MemberUp on the EventStream (how the
cluster's domain-event publisher reaches a Cluster.Subscribe subscriber)
instead of a direct RepublishPrimaryVerdict tell; it fails when the
Cluster.Subscribe(IMemberEvent) call is deleted.
MINOR-2: the published verdict MIRRORS the live gate; before this node has
joined its member count is 0 and both allow (seconds; up to the bootstrap
guard's 25 s probe), so the Sparkplug primary-host claim can be held on
both nodes in that window (native ack is unaffected - view and route
agree). Behaviour deliberately unchanged: denying until self is Up would
change the live gate too and delay writes/acks on a single-node cluster's
own join. Comments + AlarmTracking.md / Redundancy.md / Mqtt.md corrected.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
Review R1:
- M1 (#609): DriverHostActor.StopChild releases the instance's resilience state before evicting the hub.
- m2 (#597): DriverHostActor publishes ShouldServiceAsPrimary() to IDriverPrimaryRoleSource at start and on
cluster membership changes (deduped), so the native-ack gate denies in a pair's boot window.
- m4 (#601): ApplyCachedArtifact returns false on empty/malformed bytes; the cached revision is not adopted,
Direct mode goes Stale (no RUNNING FROM CACHE), FetchAndCache lands Steady-with-no-revision and a
re-dispatch of the same revision really applies.
- m3 (#603): LocalDbReplicationPolicy splits into CapturesChanges (PeerAddress||ApiKey, registration) and
SharesRowsWithPeer (ApiKey && (PeerAddress||SyncListenPort>0), alarm drain gate).
- m5 (#618): Program.cs catch-all filters HostAbortedException (design-time tooling).
- m6 (Cluster-002): ClusterRoleInfo seeds only Up/WeaklyUp members.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
The default decider restarts a faulting child immediately and without limit, and
the hook makes every restart do ShutdownAsync+InitializeAsync+SubscribeAsync, so a
driver faulting on every value became a reconnect storm. Per-driver last-reprovision
stamp + DefaultReprovisionBackoff (10s, matched to DefaultReconnectInterval);
a held request is re-delivered by a single per-driver timer carrying the original
incarnation ref, so the F2 identity check still applies.
Seen RED with the backoff disabled: InitializeCount == 3 (second applied
immediately) where the test requires 2-then-3.
Also: the _children.Keys union-arm test (seeder gained taglessDrivers) and the
_desiredByDriver xmldoc note that it lingers on last-known-good after a failed apply.
A ReprovisionRequested in flight from an old incarnation could reach the NEW child
after HandleRestartDriver swapped it in. In Connecting that child DOES handle
InitializeRequested, so it started a second concurrent InitializeAsync, racing the
#527 dispose-before-overwrite.
Seen RED: InitializeCount == 2 on a child held in Connecting by a gated init
(an already-connected child swallows InitializeRequested and passes vacuously).
Seen RED against a T3-reverted build: all 4 failed, Created[1].SubscribeCount == 0
on the headline case. 20x isolation run green after the fix was restored.
Both apply handlers (DriverHostActor.ApplyAndAck + HandleFetchedForApply)
and OpcUaPublishActor.HandleRebuild used to rebuild a JsonDocument over the
same artifact bytes for every question they asked of it — driver instances,
composition, cluster scope, data-plane authority. Each now opens ONE
ParsedDeploymentArtifact per apply and threads it through:
- ReconcileDriversForApply / ReconcileDriversFromParsed return the blob
plus the parse the rest of the apply reads; the handler owns it and
disposes it in its finally.
- PushDesiredSubscriptionsFromParsed and CacheAppliedArtifact take that
parse instead of re-parsing the bytes.
- HandleRebuild opens once after the #485 guard and feeds both the
data-plane authority and the composition from it.
The blob-based variants stay as the lenient wrappers the boot paths need
(BootstrapFromCache, RestoreApplied), so the three remaining static Parse*
call sites are boot-only — none on an apply path.
The RebuildAddressSpace message still carries raw BYTES, deliberately: the
publish actor is a different actor with its own mailbox and
ParsedDeploymentArtifact is a disposable, non-thread-safe object owned by
the sending handler. It opens its own single parse on the far side.
FeedDataPlaneAuthority still SHA-256s the RAW bytes for the revision id — a
re-serialisation of the parsed document would be a different byte sequence
and would not match the deployment's RevisionHash.
Behaviour note: Open() throws JsonException on bytes the static wrappers
swallowed to an empty result. That is routed into each path's EXISTING
failure handling — ApplyAndAck's catch (Failed ack, children left running)
and HandleRebuild's catch (Error log, last-known-good still served) — not
as a new outcome. An empty composition is not a benign no-op there: the
planner diffs it as a PureRemove and tears the address space down, so the
throw is the honest reading of unusable bytes.
Replaces the temporary per-item unwrap bridge with a real batched handler: one
AttributeValueBatchPublished burst now costs ONE hop to the dependency mux and ONE
OpcUaPublishActor.AttributeValueBatchUpdate, instead of one message per value.
Routing per item is unchanged — the v3 Batch 4 single-source fan-out (raw NodeId +
every referencing UNS NodeId, identical value/status/timestamp), with an unmapped ref
Debug-logged and dropped without disturbing the rest of the burst. A burst in which
nothing resolves sends no publish message at all rather than an empty batch.
Indexed for-loop over Items is deliberate: no interface-enumerator allocation per
burst on the hottest routing seam. ForwardToMux and its registrations stay for the
comms-loss fan and per-tag drivers.
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.
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.
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
Joseph Doherty