VirtualTag.Enabled was authorable but ignored by both runtime-plan seams.
AddressSpaceComposer.Compose and DeploymentArtifact.BuildEquipmentVirtualTagPlans
now drop disabled rows (byte-parity), so no node is materialised and no
VirtualTagActor is spawned - the same outcome as a disabled scripted alarm.
DraftValidator exempts a disabled VT from VirtualTagNoEffectiveTrigger.
Calculation tags have no Enabled flag, so no equivalent gap. docs/VirtualTags.md updated.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
- A self-contained Host publish (-r win-x64 --self-contained, the Windows installs) now publishes the
ScriptWorker self-contained for the same RID into script-worker/. The launcher starts its apphost
directly when runtimeconfig.json lists includedFrameworks. A framework-dependent worker (docker-dev)
keeps running through the dotnet muxer.
- docker-dev/docker-compose.script-runner.yml: OutOfProcess for the central pair's VT + Calc seams.
Live drill recorded in design §12.1 (runaway kill + quarantine, kill -9 recovery, open circuit
with /health/ready still 200, metrics).
- ScriptRunnerPerfTests (Category=Perf, opt-in via OTOPCUA_SCRIPT_PERF=1): InProcess vs OutOfProcess
at K=1/2/4. The accepted 5k evals/s/node gate passes on macOS and windev. The host-CPU, RSS and
Windows latency budget misses are tracked as #789.
- Fix: the single-flight slot is released before the awaiting caller resumes. Before, a caller that
started its next run right after the previous one returned could be refused with "previous
evaluation still running", which the seams publish as Bad.
- Fix: the worker compiles on dedicated threads, not the thread pool. A burst of cold loads starved
the heartbeat, and the host killed the shard (found on windev). Load-timeout queueing is #790.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Second P1 slice of docs/plans/2026-11-09-out-of-process-script-runner-design.md; still opt-in behind
Scripting:Runner=OutOfProcess (InProcess unchanged).
- Compiled-PE cache (design §7, §14 #8): Scripting:Worker:PeCacheDir holds <dir>/pe/<id>.pe entries authenticated
with HMAC-SHA256 under a per-node random key (<dir>/pe-cache.key, created by the host, 0600/0700 on Unix), handed to
the worker in Hello. Entry id = SHA-256(compile fingerprint + kind + source); the fingerprint covers the Roslyn
version, the Core.Scripting build (analyzer, wrapper, stack guard), the sandbox reference set and the runtime. Edited,
truncated, other-key or renamed entries are rejected, deleted and recompiled; the worker never loads unverified bytes.
Pruned by recency at startup (30 d / 50 000). ScriptEvaluator splits into CompileToImage + FromImage. Measured on
40 scripts, restart to every first result: macOS 820 ms no cache / 840 ms empty / 85 ms warm;
windev (Windows) 9.5 s / 9.4 s / 1.0 s.
- Quarantine race: WorkerConnection runs exit attribution synchronously BEFORE completing the evaluations/loads it held
(and before Exited), and the shard passes it straight to the runner — a kill strike is in place before the killed
key's slot is released.
- Supervision: a dead worker restarts by itself after its backoff; Unavailable outcomes carry ScriptRunOutcome.Recovery
(-> VirtualTagEvalResult.RetryWhen) and the VT actor / Calculation driver re-evaluate once against current inputs when
it completes. Isolation worker stops after Scripting:Worker:IsolationIdleTimeout (10 min). Keys compiling when a
worker dies are suspects too (isolation finds a poison compile and quarantines it). Startup orphan sweep by launch
marker: kills only a process whose marker's host is gone and whose pid + start time + image match.
- Crash-loop circuit (Scripting:Worker:CircuitOpenAfter, 2 min): runner Unavailable, one Error, evaluations answered
at once while shards keep retrying at the cap.
- Observability (§10): otopcua.script.runner.state, worker.up/inflight/queue_depth/heap_bytes (shard),
ipc.roundtrip + eval.duration (kind), compile{cache}, isolation.suspects, exits reason=idle. `script-runner` health
check on the ready tier, at worst Degraded (§14 #12: degraded but still ready).
Left on #780: self-contained worker publish, installer (account, pipe ACL, firewall), docker-dev live drill,
scripted-alarm seam, the §4.5 per-seam batch submit, the §8 perf harness.
Tests: SupervisionTests, PeCacheStoreTests/PeCacheWorkerTests (real workers), ScriptRunnerHealthCheckTests, VT + Calc
recovery seam tests; 14 mutations each turned their test red.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Item 2: a tag's optional `write` block picks the statement the driver builds from its own
catalog-validated identifiers (no authored SQL; values always bound):
- Update (default, unchanged back-compat UPDATE ... WHERE key; 0 rows = BadNoData)
- Insert (key+value[+stamp]; append for a WideRow topByTimestamp tag)
- Upsert: SQL Server MERGE WITH (HOLDLOCK), PostgreSQL INSERT ... ON CONFLICT, MySQL
ON DUPLICATE KEY UPDATE (1-or-2 rows), Oracle MERGE from DUAL
- Procedure: named stored procedure resolved against the routine catalog by the catalog gate
(INFORMATION_SCHEMA.ROUTINES / ALL_PROCEDURES), value/key/timestamp params by name (MySQL positional)
- stampTimestamp sets the tag's timestamp column to the write time (UTC)
Odbc serves Update/Insert only (shared SqlWriteSupport table). Constraint failures map to
BadEntryExists (unique) / BadOutOfRange (other integrity, data exception); everything else,
permission denied included, stays BadCommunicationError. WriteIdempotent stays false.
The parser refuses malformed blocks (strict keys/enums/param charset); DraftValidator emits
blocking SqlWriteBlockInvalid / SqlWriteModeUnsupported from the driver's own parse; the AdminUI
editor surfaces the same message.
Items 1 + 4 (docs/drivers/Sql.md): ODBC stays the ANSI SQL:2008 profile (native providers for
SQL Server/MySQL, DB2 unsupported) with an ops note on driver manager + vendor driver on every
Odbc and AdminUI host; read-only drivers require a SELECT-only login, writable drivers a login
limited to their write targets.
Tests: SqlWriteModeTests (parse, every dialect's SQL, gate, status map, SQLite end-to-end),
DraftValidatorSqlWriteBlockTests, AdminUI round-trip; live SqlWriteModesLiveTests on SQL Server,
PostgreSQL, MySQL, Oracle (seeds gain write_targets + set_target) and an Odbc write leg.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
(a) ScriptRunGuard logs its quarantine Error through the seam's script root
logger, which carries no ScriptName, so ScriptLogCompanionSink mirrored it to
the main log as "[Script] unknown emitted Error ..." and rate-limited every
quarantine on the node under one shared "unknown" key. The guard now binds the
script instance key as ScriptName (ForContext) on the quarantine Error and the
subscriber-threw Warning, so the mirror names the script and limits each one
separately. Test drives two real quarantines through the companion sink with a
1-per-window limiter: both mirrored, each named, none "unknown". Mutation-
checked: dropping the ForContext fails it.
(b) Evaluated a quarantine-specific OPC UA status (BadOutOfService) and kept
BadInternalError: it is the documented code for any script-attributable fault
on virtual tags, Calculation tags and scripted-alarm quality (and the
out-of-process runner design keeps it), BadOutOfService reads as a deliberately
disabled point, a new code is a ScadaBridge-visible contract change, and the
once-per-Good->Bad dedup means a timeout-driven quarantine would rarely show a
different code anyway. Documented in docs/VirtualTags.md ("Quarantine: what a
client sees"), with pointers from Calculation.md, ScriptedAlarms.md and the
OpcUaStatus.BadInternalError doc comment, plus where a quarantine IS visible.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
In the normal timeout branch (some acks landed) the deployment row was set to
TimedOut with a null FailureReason, so GET /api/deployments/{id} could not say
which node to fix; only the central log named it. The branch now persists a
reason naming the silent nodes ("Timed out waiting for acks (n/m landed). No ack
from: ..."), bounded to 20 named nodes plus a count of the rest, and still
backstopped by the 2048-char Truncate.
Tests: the reason names the silent node and not the acked one; on a 25-node
silence it names 20, counts 5 and fits the column. Mutation-checked: dropping
the SetFailureReason fails both, dropping the bound fails the second.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The docker-dev centrals logged "Mesh PSK secret ot-mesh-psk-<cluster> ... is
absent (or blank) in the store; the key is REVOKED" for every cluster on every
contact refresh, while every gRPC dispatch and ack authenticated.
Investigated: not an auth bypass. The rig supplies each cluster's key through
MeshTransport:ClusterPsks (central side, equal to that cluster's nodes'
MeshTransport:GrpcPsk) and seeds nothing in the store. TryGetPsk/Matches consult
the override first on every call, so enforcement was on; only RefreshAsync read
the store for those clusters regardless and logged a revocation that never took
effect. RefreshAsync now skips clusters covered by a non-blank override, so the
REVOKED warning is only logged for a cluster that genuinely has no key.
Test pins it with a positive control (a cluster with no override is still read,
warned about and refused) and enforcement on the override key (wrong token
rejected). Mutation-checked: removing the skip fails the test.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Choice: implement the timer (option a). The field was designed and promised as a
working trigger: the entity doc says ChangeTriggered and TimerIntervalMs together
realize the trigger model, the DB check constraint requires ChangeTriggered OR a
timer >= 50, the AdminUI virtual-tag modal exposes "TimerIntervalMs (optional)" and
UnsTreeService rejects a tag with neither, and the 2026-06-08 design specified
"ChangeTriggered OR TimerIntervalMs>=50". Only the runtime never honoured it (the
standalone TimerTriggerScheduler was deleted in #603), so a timer-only tag sat at
BadWaitingForInitialData forever.
- EquipmentVirtualTagPlan carries ChangeTriggered/TimerIntervalMs (in equality, so
an edit respawns the child); both the composer and the artifact decode fill them.
- VirtualTagActor schedules a repeating tick (PreStart, cancelled in PostStop,
clamped to the 50 ms floor). ChangeTriggered=false makes the timer the only
trigger. A tick during an in-flight async run (#779) is dropped, not stashed, so
slow scripts never build a backlog of runs.
- DraftValidator VirtualTagNoEffectiveTrigger (hard error, same severity as
CalculationNoEffectiveTrigger): no timer and no change trigger with a literal
ctx.GetTag read.
Tests: VirtualTagActorTimerTests, DraftValidatorVirtualTagTriggerTests, artifact
decode test. Mutation-checked: no scheduling, stashing ticks, no cancel in
PostStop, and removing the validator call each fail a test.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Gates A (#731), B (#732), C (#733) and E (#734) run on the local docker-dev
rig against origin/master 1b9c36f3 on 2026-11-09: all four PASS. Records the
procedure, the measured thread/fd/RSS, subscription and deploy-status
numbers per gate, plus six non-blocking observations.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Catalogue of what breaks moving from 1.5.378.106 to 2.0.1: restore
break (MEL 10.0.12), 42 -> 103 compile errors in the first two waves,
root-cause groups with API mappings, PeerAddress replacement for the
reflection-based remote endpoint resolver, effort and risks.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The Lane H2 remediation made Periodic(100 ms) the options-class default, but
appsettings.json, docker-dev and Program.cs's bad-value fallback all used
PerEntry, so effective behaviour never changed and the docs contradicted the
shipped config. The options default is now PerEntry; Periodic stays opt-in
(<= CommitIntervalMs of crash loss in exchange for throughput). Program.cs now
uses ContinuousHistorizationOptions.ResolveCommitMode() so the fallback shares
the single DefaultCommitMode constant. docs/Historian.md updated.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The docs said script authoring was Admin-gated (VirtualTags.md "Known
resource limits") and that a Designer "cannot publish" (security.md roles
table). The code admits Designer to both: Scripts, ScriptEdit, EquipmentPage
(virtual tags, scripted alarms), /api/script-analysis/* and Deployments all
carry AdminUiPolicies.ConfigEditor = RequireRole("Administrator",
"Designer"). That is a recorded decision, not drift: AdminUI-001 (98b27fc1)
loosened the analysis backend from FleetAdmin to Administrator,Designer to
match the page gate, and R2-05 converged the same pages on ConfigEditor
"with the same semantics". So the docs follow the code:
- VirtualTags.md: the primary control is ConfigEditor (Administrator or
Designer), which also gates deploy.
- security.md roles table: Designer = the whole ConfigEditor surface incl.
scripts and deploy; Administrator = Designer + the FleetAdmin-only surface
(role grants, API keys, certificate actions, manual failover) — the old
"publish, cluster/node CRUD" were ConfigEditor all along.
- 2026-06-09 Monaco plan: its FleetAdmin gate is marked superseded.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Reproduced first, out of process (ScriptRecursionContainmentTests re-launches
the xunit v3 test assembly to run one recursive script through
ScriptEvaluator + ScriptRunGuard, so the runner never executes one): before
this change every shape — local function, mutual local functions,
self-referencing Func<>, anonymous self-application, recursion through LINQ,
an expression-bodied void lambda, an anonymous method — ended the child with
exit 134 (SIGABRT, "Stack overflow."). A StackOverflowException cannot be
caught, so in production one script takes the whole node down.
Fix: ScriptEvaluator.Compile now rewrites the validated wrapper
(ScriptStackGuardInjector, after ForbiddenTypeAnalyzer) so the script body and
every lambda, anonymous method and local function starts with
ScriptStackGuard.Check(), which throws ScriptStackExhaustedException when
RuntimeHelpers.TryEnsureSufficientExecutionStack reports the stack nearly
spent. The script then fails like any throwing script (Bad, script log) and
the process lives.
Why a runtime check, not an analyzer rule: recursion is not decidable from
syntax (anonymous self-application never names itself), and bounded recursion
must keep working. Expression-bodied lambdas/local functions become blocks
(`return e;` or `e;`, chosen from the bound symbol, so the same overload
binds); expression-tree and async lambdas are left alone. Limits (documented
on ScriptStackGuard and in docs/VirtualTags.md): depth depends on the thread's
stack size; a recursion that never re-enters script code is not seen; a script
may catch the exception, which only unwinds it.
Tests: ScriptRecursionContainmentTests (7 shapes, child process each; all 7
fail with exit 134 when the rewrite is disabled), ScriptStackGuardInjectionTests
(bounded recursion and every rewritten lambda/local-function shape keep their
value; the expression-tree case fails if those lambdas are rewritten).
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
VirtualTagActor called the blocking IVirtualTagEvaluator.Evaluate, and
RoslynVirtualTagEvaluator blocked on the run guard with
.GetAwaiter().GetResult(), so every script run parked a dispatcher thread for
up to its 2 s budget (a first-use Roslyn compile did too). This is the #750 P0
refactor for virtual tags.
- IVirtualTagEvaluator.EvaluateAsync (default: a completed task over the
snapshot overload, so the null evaluator and fakes are unchanged).
- RoslynVirtualTagEvaluator implements it without blocking: passthrough and
compile failures complete synchronously, a first-use compile goes to the
pool, the run is awaited. The blocking overload now delegates to it.
- VirtualTagActor: a completed task is applied in the same turn, as before;
an incomplete one switches to an "evaluating" behaviour that stashes every
dependency/reassert message until a private completion message arrives,
then applies the result and unstashes. Each trigger is still evaluated
against its own inputs, results are applied/published in trigger order and
the dedup/Bad-transition state sees the same sequence. The dependency map is
not touched while a run reads it.
Calculation and scripted-alarm serialisation is NOT relaxed: neither blocks a
thread, and running independent scripts in parallel is not semantics-neutral
(pool consumption bound, cross-tag publish order, timer/pump exclusion for
Calc; state-machine/persist boundary, transition timestamps and the
store-failure path for alarms). Recorded in docs/VirtualTags.md and the #750
design's P0 entry.
Tests: VirtualTagActorAsyncEvaluationTests — on a one-thread dispatcher a slow
script no longer starves a sibling actor and the slow actor answers Identify
(fails when the actor calls the blocking Evaluate); triggers during a run are
evaluated afterwards in order against their own inputs; a reassert during a
run is applied after it; a faulted evaluation takes the evaluator-threw path
(the ordering/reassert tests fail with the stash removed).
RoslynVirtualTagEvaluatorAsyncTests — a slow script and a cold compile return
an incomplete task (fail without the override); cheap outcomes and timeouts
match the blocking overload.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
DriverHostActor built ScriptedAlarmEngine without a scriptTimeout, so every
production alarm predicate fell back to TimedScriptEvaluator.DefaultTimeout
(250 ms) while virtual tags and Calculation tags ran on 2 s. History shows no
decision behind it: 250 ms is the generic Phase-7 A.4 default, the VT
evaluator moved to 2 s in F8b (219d10a2) and the alarm engine was simply never
given a value.
- New node-wide Scripting:RunTimeout (default 2 s, clamped to (0, 60 s],
non-positive -> default, each clamp a startup Warning) on ScriptingOptions.
- The VT evaluator and the alarm engine default to it; DriverHostActor now
builds the engine through CreateScriptedAlarmEngine, which passes it
explicitly. Calculation keeps its per-driver runTimeoutMs.
- Docs: ScriptEditor (config table), ScriptedAlarms, VirtualTags.
Tests: ScriptedAlarmEngineBudgetTests (a 600 ms predicate now completes and
activates; default and configured budgets), ScriptedAlarmEngineBudgetWiringTests
(the host-built engine), ScriptingOptionsBindingTests (bind + clamps). The
engine and wiring tests fail with the old 250 ms default restored.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
allowWrites is no longer inert: SqlDriver implements IWritable, default off.
- Gates (outermost first): node Writable + WriteOperate (server), driver
allowWrites, the tag's own TagConfig "writable": true (strict JSON boolean;
anything else refuses the tag, and the deploy gate says why), and a model that
names exactly one row — KeyValue, or WideRow by where-pair. topByTimestamp and
Query are BadNotWritable.
- SqlWritePlanner: the only statement a write can become is
UPDATE <table> SET <value col> = @v WHERE <key col> = @k, built from the
catalog-gate-canonical identifiers (dialect-quoted) with value and key bound;
value-before-key marker order so positional ODBC binds correctly. No authored
write SQL, INSERT, UPSERT or procedure.
- Executed in a transaction that commits only when exactly one row was touched:
0 rows -> BadNoData (never inserts), >1 -> rolled back, BadConfigurationError,
unknown count -> rolled back, BadNotSupported. Bounded by operationTimeout;
provider errors -> BadCommunicationError; health untouched.
- Discovery advertises Operate only where a write could succeed.
- AdminUI: driver form's Allow writes switch is live; tag editor gains a
per-tag write switch; model validates writable strictly (parity corpus).
- Factory no longer warns "allowWrites ignored"; it logs whether writes are on.
- Tests: SqlWriteTests (planner SQL/bind order/injection/coercion, every gate,
readback, absent key, two-row rollback, gate-rejected tag, discovery);
MySqlReadTests write legs against InnoDB (keyed row only, two-row rollback,
server-side grant refusal, default-off). Seed adds a setpoints table and the
one UPDATE grant.
(lmxopcua#753)
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
SqlProvider.Odbc is now constructed; every enum member is a working backend.
- OdbcDialect: System.Data.Odbc 10.0.14 (no vulnerable packages), ANSI double-
quote quoting, positional `?` markers, INFORMATION_SCHEMA catalog (system
schemas of the common backends excluded by name — the live gate caught
psqlODBC listing pg_catalog), FETCH FIRST n ROWS ONLY limits, a conservative
1000-parameter chunk. Positional binding is safe because every caller adds
parameters in marker order (planner IN list, catalog schema-then-table).
- Named queries on a positional provider are checked by `?` count against the
declared params (literals/comments ignored), since there is no name to match.
- Supported backends are those whose SQL accepts that profile (PostgreSQL via
psqlODBC proven, Snowflake-like); SQL Server/MySQL/DB2 should use their native
provider. The host needs the ODBC driver manager + vendor driver.
- Reserved-provider controls in factory/probe/browser tests now use an
undeclared ordinal (99); the AdminUI form offers Odbc and a test pins the
dropdown theory to SqlDialects.Supported.
- Tests: OdbcDialectTests (unit, any host). OdbcReadTests + Docker/odbc/
run-odbc-gate.sh run the built suite in a throwaway Linux SDK container with
unixODBC + psqlODBC against a throwaway seeded PostgreSQL: KV IN-list
positional binding, where-pair, topByTimestamp via FETCH FIRST, a two-
parameter named query, and the browse walk. macOS has no vendor ODBC driver,
so there the suite skips.
(lmxopcua#753)
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
SqlProvider.MySql is now a constructed backend instead of a reserved name.
- MySqlDialect (Driver.Sql.Dialect): MySqlConnector 2.6.2 (MIT, fully managed,
no vulnerable packages), backtick quoting with MySQL's 64-char / no-trailing-
space rules, information_schema catalog SQL with the database as the schema
(system databases excluded, aliases pinned), trailing LIMIT, a type map that
also reads the reader-side spellings (INT UNSIGNED -> UInt32, BOOL -> Boolean).
- SqlDialects: the one provider -> dialect map. The factory, the Test Connect
probe and the browse picker each carried their own switch (and drifted once);
they now all read this, and their refusals list the supported/reserved sets
from it.
- AdminUI driver form offers MySql.
- Tests: MySqlDialectTests (golden quoting/catalog/type map/planner SQL); the
provider-gate tests in factory, probe, browser and form now treat MySql as
supported and Odbc as the reserved control. MySqlReadTests + Docker/mysql
(throwaway mysql:8.4, SELECT-only login): liveness, KV/wide-row/topByTimestamp,
INT keys vs authored text, case folding against case-sensitive table names,
catalog rejection, browse walk, factory+probe, named query with a bound
parameter, and a writing stored function whose effect the rolled-back
transaction discards.
(lmxopcua#753)
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
SqlTagModel.Query (design §5.4) is now served end to end instead of refused.
The driver config declares `queries` (name -> sql, params, maxRows); a tag names
the query, the result `column` it reads and optionally a `rowKey` picking its row.
- Parser/deploy gate: query + column required, rowKey all-or-nothing; Inspect
explains each refusal. Table is not read for this model.
- SqlReadOnlyStatement: single SELECT/WITH only, no data/schema-changing keyword
outside literals/comments (refuses SELECT INTO, FOR UPDATE, DML CTEs). Shared by
the factory and the AdminUI driver form (SqlNamedQueryRules.Describe), so the
form refuses what the driver refuses. One bad entry fails the driver.
- Parameters are always bound; the factory refuses a declared parameter whose
dialect marker the statement never uses.
- Reader: one execution per query per poll, inside a transaction that is always
rolled back; result columns matched exact-then-unique-CI; no row -> BadNoData,
ambiguous row / missing column / over maxRows -> BadConfigurationError.
- Catalog gate passes Query tags through (nothing is quoted into SQL); a
Query-only driver skips the catalog load. A tag naming an undeclared query is
skipped with the reason on /hosts.
- AdminUI: typed editor offers Query (query/column/rowKey); driver form edits
`queries` as JSON and keeps the last valid set on a typo.
(lmxopcua#753)
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Design only. Documents why today's in-process ScriptRunGuard can abandon but
not stop a runaway script (file:line evidence), and proposes a sharded
worker pool with an isolation worker under a ScriptRunnerActor, a
named-pipe + protobuf frame protocol carrying full input snapshots,
soft/hard deadlines with process kill, kill attribution and quarantine,
worker sandboxing (scrubbed env, pinned assemblies, Job Object, heap
limit), an HMAC'd content-addressed PE cache, performance budgets,
failure-quality semantics, a flag-gated InProcess/Shadow/OutOfProcess
rollout, a test plan, open decisions and an assumptions register.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Item 3 of #765. A keyed (API-key) session whose key had no Galaxy ack identity
got Good and then a revert, because whether the key has an identity lives in
the Secrets store, which must not be read under the SDK lock.
GalaxyAckIdentityDirectory (Driver.Galaxy) keeps a node-wide set of the key ids
that have a galaxy-ack-user secret. It lists this node's Secrets store
(metadata only, nothing decrypted) and is read lock-free.
GalaxyDriver.GetAcknowledgeRefusal (IAlarmAcknowledgePreflight, published
through NativeAlarmAckPreflightView as in #760) now refuses a keyed session
whose key the directory reports Absent, when the driver has no fallback ack
account. A refusal also asks for a refresh.
Fails open: before the first refresh, once the snapshot is older than
StaleAfter, and from a signalled identity-secret change until the next
refresh, the lookup answers Unknown and the ack takes the old Good-then-revert
path.
Refresh: GalaxyAckIdentityRefreshService runs on driver nodes, every
Drivers:Galaxy:AckIdentityCache:RefreshIntervalSeconds (default 30;
StaleAfterSeconds 0 = three intervals; validated at start). It also refreshes
early on secret changes. The Secrets library has no change event, so
SecretsRegistration decorates ISecretCacheInvalidator, which is called by
AdminUI /api-keys, /admin/secrets and the replication appliers. API-key
changes need no event: a revoked key is already denied before the pre-check.
Tests: Galaxy.Tests GalaxyAckIdentityDirectoryTests (16) and Host.Tests
GalaxyAckIdentityCacheWiringTests (9). Seven separate mutations each turn at
least one test red: pre-check ignores the directory, factory drops it, no
invalidator decoration, no staleness check, no change-pending check, no
refresh on refusal, loop never wakes early.
Docs: Galaxy.md, AlarmTracking.md, Configuration.md, .claude/rules/alarms.md.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Item 1 of #776 (the B2 over-the-wire ask): Client.CLI had no way to issue a
HistoryReadEvents. `historyread --events` now sends ReadEventDetails to an
event notifier and prints one line per event in select-clause order.
--fields picks the select clauses (default EventId, EventType, Time,
SourceName, Severity, Message, ConditionName (on ConditionType), SourceNode);
--of-type adds an OfType where-clause. --events is exclusive with --modified /
--aggregate, and --fields / --of-type require it.
Client.Shared gains IOpcUaClientService/ISessionAdapter.HistoryReadEventsAsync
(default-implemented as NotSupported for other implementations) backed by the
real session adapter through HistoryReadPaging, so continuation points are
followed up to --max and a Bad node status is surfaced, not printed as empty.
Tests: CLI tests over the fake service (default filter, --fields/--of-type
filter, option conflicts); an over-the-wire test drives the shared client's
event read against the in-process server — the default fields come back
non-null, OfType filters, an invalid where-clause is BadEventFilterInvalid.
The other #776 items are decisions and are left open.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
QuestDB modified-history reads dropped every still-pending trail row that was
not the latest at its source time, so V0->V1 (applied marker never written)
followed by V1->V2 hid the V0->V1 change for good. A pending row below the
latest is now served when the NEXT trail row at that time recorded a different
prior value (kind, value, status or the prior's server_ts) - the next change saw
what this one wrote; an orphan of a change that never landed leaves the same
prior for the next row and stays hidden. QuestDB window functions reject
VARCHAR, so the next row's prior comes from a LEFT JOIN on (ts, lead(id)).
The updater's comment and marker-failure log no longer claim modified reads
"still serve it" unconditionally.
Regression tests (QuestDB 10.0.1 throwaway container): marker lost then a
replace / delete at the same time; an orphan followed by a marker-lost retry
then a later change (orphan still dropped); the random marker-mix fuzz now
varies the recorded priors. All four fail on the old filter.
NITs:
- SqlServerHistorianValueWriter: a SqlException raised while the token is
cancelled (SqlClient's "Operation cancelled by user") is rethrown as
OperationCanceledException instead of booked as a failed write. Integration
test holds a TABLOCKX on Sample and cancels mid-INSERT; it returned false before.
- SqlServerHistorianDataUpdater.DeleteRawBatches is an Interlocked counter.
- HistoryReadEvents: an event filter whose SDK validation throws (unknown
operator) failed with the generic BadUnexpectedError; the node manager now
re-throws any failed event-filter validation as BadEventFilterInvalid.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
HistoryReadEvents projected only EventId / SourceName / Time / ReceiveTime /
Message / Severity: a select clause asking for EventType, ConditionName or
SourceNode (what most generic alarm viewers ask for) got nulls, and the
event filter's where-clause (an OfType) was ignored.
- HistoricalEvent gains optional EventTypeName, ConditionName and AlarmId.
SqlServer, QuestDB and Gateway event reads fill them from what their
alarm writers already store (AlarmTypeName / AlarmName / AlarmId; the
gateway event Type and its AlarmName / AlarmId properties).
- The node manager maps the type name to its Part 9 event type (unknown or
non-event => AlarmConditionType), projects ConditionName, and resolves
SourceNode to the materialised condition the alarm id names (the live
events' SourceNode), else null.
- The where-clause is evaluated per event with the SDK content filter
through an IFilterTarget (OfType via the type tree; other operators
against the projected fields), after the page's cursor is taken, so
paging is unchanged.
Tested in the node manager, over a real client session (the Client.CLI has
no `historyread --events` mode yet; filed as a follow-up), and by alarm
writer -> event read round trips on SqlServer and QuestDB.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The Gateway backend's gaps against SqlServer/QuestDb were documented
piecemeal. The backend-selection table in docs/Historian.md now also covers
the aggregate set (and the native time-weighted opt-in), emulated reverse
and end-only reads, the client-side events cap, retention and the health
probe.
Continuous historization on Gateway silently loses sample quality (stored
Good), comms-loss outage markers and non-numeric values. Building its live
writer, which only happens when ContinuousHistorization is enabled, now
logs one startup Warning that names the gaps and the alternatives.
The review asked for the matrix in the AdminUI backend picker, but no such
picker exists: the backend is configuration-only. Filed as a follow-up.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The continuous-historization recorder awaited every outbox append (an fsync
under PerEntry) on its mailbox, so ingestion was capped at 1/fsync_latency
bursts per second while the mux kept telling into an unbounded, unmetered
mailbox. GetStatus also suspended the mailbox on an unbounded outbox count.
- Gated ingest bursts go to a bounded channel (64 bursts) drained by one
ordered append loop, which commits every burst waiting behind a slow
append in a single AppendBatchAsync (FIFO kept). Only a full queue holds
the mailbox (backpressure, bounded memory). Counters are Interlocked; the
loop nudges the drain. PostStop flushes queued bursts within the shared
5 s stop budget.
- GetStatus reads counters on the actor thread and pipes a count bounded by
a 10 s timeout (-1 when it does not answer); it no longer suspends the
mailbox. RecorderStatus gains PendingAppends, MailboxDepth and
AppendBackpressureWaits; two gauges publish pending appends and mailbox
depth.
- ContinuousHistorization:CommitMode defaults to Periodic (100 ms). The
Host's appsettings.json, its unparseable-value fallback and docker-dev
still say PerEntry: filed as a Host follow-up (outside this lane).
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
QuestDB has no multi-table transaction, so a HistoryUpdate writes its trail
row before the sample change. A failed or crashed change left an orphan
trail row that modified reads served as a change that happened, and a trail
INSERT re-sent after a client timeout duplicated (the table had no key).
- sample_modified gains `pending` and DEDUP UPSERT KEYS(ts, tag,
modified_id). The trail row is written pending; after the change lands
the same row is re-upserted pending = false (the applied marker).
- A transient trail INSERT failure is re-sent once with the same ids; a
first attempt that did apply converges on the key.
- ReadModifiedAsync serves a pending row only when it is the latest trail
row at its time and the sample there was written at or after the
modification (a crash between change and marker). The orphan of a change
that never landed, typically followed by the client's retry, is dropped.
ASOF JOIN over the window's samples of one tag; the resume filter sits
outside the window function so it never cuts a time's partition.
- The bootstrapper migrates an older trail (ADD COLUMN pending, DEDUP
ENABLE) and waits for it to apply (bounded 10 s); old rows read
pending = false and stay visible. Until applied, or with
AutoCreateSchema=false, the old format is used.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
DeleteRaw deleted an unbounded range in one statement and transaction with
OUTPUT INTO the modification trail. Over years of a 1 Hz tag that is tens of
millions of rows deleted and audited at once: lock escalation on Sample
(blocking the live writer), transaction-log growth, and a 30 s
CommandTimeout that rolled the whole thing back.
It now runs DELETE TOP (4000) batches, each its own transaction with its
trail rows in the same statement (every deleted row is still audited),
cancellable between batches. A failure part-way leaves the committed batches
deleted and reports the failure; re-issuing completes it (documented, same
as QuestDB's paged tombstoning). The existing atomicity tests still hold per
batch.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Event reads filtered on source and every modified-value read filtered on tag
scanned that symbol column in every partition of the window; only
sample.tag carried an index (#683).
Both columns are now SYMBOL INDEX in the DDL, and the bootstrapper's index
migration is generalised over QuestDbTables.SymbolIndexes (sample.tag,
event.source, sample_modified.tag): a missing index is added with the same
idempotent ALTER ... ADD INDEX, Warning-only when AutoCreateSchema is false
or the ALTER is refused. Integration tests: new tables carry all three; a
pre-index table set gains them on boot and event reads answer the same.
This reverses #683's documented choice to leave the two unindexed;
docs/Historian.md carries the reasoning.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The SqlServer purge read every TagId and issued at least one DELETE TOP per
table per tag on every pass, even when nothing was eligible: 50000 tags made
100000 statements per hourly purge, serialised on one connection under the
applock.
One pre-filter query per table (EXISTS seek on the clustered key) now picks
the tags holding a row older than the cutoff; only those get deletes. A pass
with nothing eligible sends just the Event delete. The mid-command
cancellation test now accepts the pre-filter SELECT as the blocked command,
since it is the first statement to touch Sample.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
All three alarm-history writers inserted one event per round trip, serially
(SqlServer also opened a pooled connection per event), so a 10000-event
alarm storm drained at 10000 x RTT while the backlog sat in LocalDb.
- Sql.Common.AlarmBatchWrite: one statement per chunk; a chunk that fails
PermanentFail-class is re-sent event by event so only the poison event
dead-letters; a transient failure retries the chunk; a cancellation never
dead-letters.
- SqlServer: multi-row insert-if-absent over a VALUES set (150 events per
statement, under the 2100-parameter limit), duplicate ids in a batch
written once; a duplicate-key race isolates per event, where it is Ack.
No new table type, so no schema migration (the review suggested a TVP;
the VALUES form gets the same one-round-trip-per-chunk without one).
- QuestDb: multi-row INSERTs of 500 events; pre-1970 events still
dead-letter unsent.
- Gateway: no batch RPC exists, so sends are pipelined, at most 8 in
flight, outcomes kept in input order.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The QuestDB live writer waited for WAL visibility after every tag group, and
the wait polled wal_tables() every 10 ms. With DrainConcurrency = 8 that was
up to eight pollers at ~100 queries/s each against a database whose WAL
apply was already behind.
New Runtime seam IHistorianPassWriter (WriteLiveValuesInPassAsync +
FlushAsync). The recorder writes every group of a pass through it and calls
FlushAsync once after all groups report, before the ack, whenever any group
wrote (an un-acked pass is redelivered and its insert-if-absent check must
see this pass's rows). A flush failure does not fail the pass. The QuestDB
writer implements it: in-pass writes only mark a flush as owed; FlushAsync
waits once. The poll now backs off 10 -> 20 -> ... -> 200 ms, never past the
timeout. Direct WriteLiveValuesAsync callers still wait per call.
Also updates the QuestDB cancelled-write integration test to the rethrow
contract from the previous commit.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Every IHistorianValueWriter caught Exception and returned false, including
OperationCanceledException from the drain's own token. The recorder's guard
("a cancelled pass is not a tag failure") was therefore dead code: at shutdown
mid-pass each cancelled group could be booked as a consecutive failure and,
with a sibling that already wrote, tip a healthy tag toward quarantine.
SqlServer, QuestDb and Gateway writers now rethrow a cancellation of the
caller's token; a cancellation-shaped error on a live token is still an
ordinary failed write.
The Gateway writer relabelled a Local-kind timestamp as UTC instead of
converting it, so the same value could land at two different times on two
backends. Stamped samples now go through LiveWritePreparer (Local -> UTC,
non-finite doubles not sent, duplicate timestamps keep the first); unstamped
samples are still left for the gateway to stamp, in batch order.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Every Gateway HistoryRead was preceded by a BrowseTagNames catalog stream
keyed by a glob filter (a name with '*' or '?' streamed every match), and the
4096-entry presence cache evicted an arbitrary entry (Keys.First()) when full,
so a sweep over more tags thrashed it.
The presence check is now one exact unary GetTagMetadata (null = absent).
Only the deployed gateway's known misparse of a missing tag (Internal /
Unknown / InvalidDataException) falls back to the exact-name catalog check;
any other failure propagates and never becomes absence. The positive TTL is 5
minutes (was 60 s); a full cache drops expired entries first, else the one
closest to expiry. Wire tests cover the working, the misparsing and the
refusing gateway.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The Gateway historian's ProcessConnectionOpen / EventConnectionOpen were
derived from read outcomes only, so a freshly booted node with a healthy
gateway reported the historian closed until a client read, and a node that
only writes (continuous historization) never flipped it.
The data source now runs a ProbeAsync loop (ServerHistorian:Gateway:
ProbeInterval, default 30 s, first probe at start-up; 0 = off). Each probe,
bounded at 10 s, is recorded as the latest connectivity observation; the
latest observation (probe or read) is what the snapshot reports. Probes never
touch the query/failure counters. DisposeAsync stops the loop before
disposing the client.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
AVEVA's native TimeAverage / Total are answered with Good-status wrong values
while the gateway sends the interval in the wrong unit (historiangw#27): one
whole-span value on the earliest interval, the rest absent. Nothing on the
wire marks the answer as degraded.
Both are now refused BadAggregateNotSupported (unsent, not counted as a
backend failure, not advertised in AggregateFunctions) unless
ServerHistorian:Gateway:EnableNativeTimeWeighted is true. The default
Gateway set is the ten raw-evaluated aggregates
(GatewayHistorianDataSource.RawEvaluatedAggregates); ServedAggregates stays
the 12 the backend can serve with the flag on. The native-path tests opt in.
Also documents the ServerHistorian:Gateway sub-section (EventDrainBudget from
the previous commit) and the events drain in docs/Historian.md.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The SDK keeps sending publish requests on a session that holds no subscription,
and the server answers each with an anonymous (id 0) BadNoSubscription. With
nothing tracked there is nothing to lose, but every answer was counted on
publish_error, so the counter climbed on any idle driver. HandlePublishError now
returns before counting for an anonymous BadNoSubscription while neither the data
nor the alarm subscription set tracks anything. Any other PublishError, and an
anonymous one while something is tracked, is counted and handled as before.
Not addressed from #764: the Republish/state probe for a delete slower than the
5 s grace, and the #728 wire test's fixed delays.
Tests: OpcUaClientIdlePublishErrorTests (idle session: not counted, with a
BadTooManyPublishRequests positive control; tracked subscription: still counted).
The idle test fails without the fix. Driver.OpcUaClient.Tests 295/0.
Docs: drivers/OpcUaClient.md.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The #733 miss budget (MissBurst / MissRefillPerSecond) was process-wide, because
the impersonation seam did not carry the client's address, so one noisy client
could drain it for everybody while the known-id set was unavailable.
- IOpcUaUserAuthenticator gains AuthenticateUserNameAsync(username, password,
IPEndPoint? remoteEndpoint, ct) (default: ignore it and delegate); the server
calls it. IApiKeyStore gains the matching FindAsync(..., IPAddress?, ...).
- The SDK (1.5.378) exposes no client address publicly. SecureChannelRemoteEndpoint
reads it from the TCP listener's private channel map by reflection (through the
protected TransportListeners, exposed by OtOpcUaSdkServer), keyed by the
session's secure channel id. Fail-soft: null => process-wide accounting only.
- BoundedApiKeyStore charges a miss to the client's own bucket (keyed by IP
address) and then to the process-wide one, which stays as the backstop. A
refusal by either takes from neither; a hit refunds both. New options
Security:ApiKeys:MissBurstPerEndpoint (10) / MissRefillPerSecondPerEndpoint (2),
0 disables; telemetry outcome endpoint_miss_budget. At most 4096 addresses are
tracked; idle buckets are pruned and an address beyond the cap is charged
process-wide only.
Tests: BoundedApiKeyStorePerEndpointTests (noisy client vs quiet one, process-wide
backstop, refund + refill, off switch, authenticator forwards the address) and
over the wire ApiKeyMissBudgetPerEndpointTests (activation hands the authenticator
the loopback endpoint; real authenticator + bounded store refuse a client's 3rd
and 4th unknown key without a store read). Each new behaviour test fails with its
fix mutated out. Host.Tests 224/0, OpcUaServer.Tests 1019/1 (the one failure is the
wall-clock NodeManagerLockTimeBenchmarkTests under suite load; passes alone).
Docs: security.md.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Residuals of #760:
1. An ack routed to a disabled / not-yet-spawned driver, or to a stubbed child
(driver does not run on this node, or its factory rejected the config), was
dropped and left the condition Acknowledged. DriverHostActor now reverts it
through the same NativeAlarmProjector.TryRevertAcknowledge path as a refused
ack.
2. A native ack arriving while the host is Stale was dropped before the projector
recorded it, so it could never be reverted. Stale now marks the whole
NativeAlarmAckPreflightView unavailable (SetHostUnavailable), so the node
manager refuses every native ack synchronously; BecomeSteady clears it.
3. (keyed sessions without a Galaxy identity) is not addressed.
Tests: DriverHostActorNativeAlarmTests' client-ack test and the FOCAS respawn
theory now start the host's own driver (AckingAlarmDriver; a factory building a
real FocasDriver per spawn over per-instance fake CNCs, with a real config-change
respawn) - both fail against the fix in their old shape. New: ack to a disabled
driver reverts, ack to a stubbed driver reverts, a Stale host refuses through the
pre-check view - all three fail with the fix mutated out. Runtime.Tests 943/0/3.
Docs: AlarmTracking.md, .claude/rules/alarms.md.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
R06:M-5 made Sparkplug *Array leaves unbindable because the ingestor refused
them. The ingestor now decodes the packed bytes_value of all 13 array types
(new SparkplugArrayCodec: LE fixed widths, int64-ms DateTime, uint32-count +
MSB-first BooleanArray, NUL-terminated UTF-8 StringArray; golden vectors are
the spec's examples, reproduced byte-for-byte by SparkplugNet 1.3.10) and
publishes a typed 1-D array through the scalar coercion, all-or-nothing.
Shape uses the fleet's one array model: the tag's platform isArray/arrayLength
keys (ValueRank=1, ArrayDimensions=[arrayLength] or [0] = variable length).
The driver reads them with TagConfigIntent's exact semantics and never rejects
on them, so composer, driver, authoring model and deploy inspector agree
(MqttArrayShapeParityTests). Shape mismatch (array metric on a scalar tag, or
the reverse) is BadTypeMismatch - a String tag no longer gets the packed
buffer as base64; a malformed body or a length differing from a declared
arrayLength is BadDecodingError (MTConnect #627 verdict). Template/DataSet
members of an array type take the same path. Arrays stay read-only (encoder
refuses them; MQTT is not in DriverArrayWriteSupport).
Browser: array leaves are bindable again and state isArray in AddressFields;
the commit mapper writes "isArray": true (no arrayLength). Discovery reports
IsArray/ArrayDim. Docs: Mqtt.md "Array metrics", design table, drivers rule.
Tests: codec (golden + malformed), ingest through the real codec + Dispatch
(the project's MiniBroker never forwards a PUBLISH, so there is no in-process
broker harness), browse, discovery, commit mapper, parity corpus. Mutating the
fix back out fails 18 MQTT + 9 AdminUI tests.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Bounded, docs-only part of #717. The standard non-transparent metadata itself is
NOT implemented here: the owner-selected contract is Warm with an explicit
degraded standby (and None standalone), today's Secondary serves Good data at
ServiceLevel 240, and the design/spec/plan for that standby behaviour is still
pending on the issue - advertising Warm now would misdescribe the server.
- docs/Redundancy.md: the "deferred pending an SDK object-type upgrade" claim
was wrong; the pinned SDK 1.5.378.106 can build the graph
(CreateMainNodeManagerFactory + AddBehaviourToPredefinedNode). Records what is
and is not advertised (i=3709 SDK default, no i=11314) and the blocker.
- .claude/rules/cluster-topology.md: said each node exposes RedundancySupport
and ServerUriArray; it does not.
- docs/OpcUaServer.md: same false claim about the SDK exposing them.
Interop gates A4/A5b stay unmet; no product code changed.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The native-ack path is fire-and-forget after the Part 9 Acknowledge returns, so
a Galaxy ack refused for lack of a credential (#749), or rejected by MXAccess,
left the client with Good and the condition showing Acknowledged while Galaxy
was never acked. Both options from the issue, layered:
1. Synchronous refusal for what the driver can decide without I/O. New optional
IAlarmAcknowledgePreflight (Core.Abstractions); GalaxyDriver refuses an
anonymous session when the driver has no AlarmAck fallback account.
DriverHostActor publishes a condition NodeId -> owning-driver snapshot into
NativeAlarmAckPreflightView (lock-free, volatile); the node manager's new
NativeAlarmAckPreflight seam, consulted after the authz and Primary gates,
returns BadUserAccessDenied with the reason - nothing routed, applied or
reported. A keyed session is not pre-checked (its identity lives in the
Secrets store, which must not be read under the SDK lock).
2. Revert for everything else. RouteAlarmAck carries the condition NodeId;
DriverInstanceActor replies AlarmAckFailed on a refusal, a backend failure,
a driver without an alarm source, or an ack dropped while connecting /
reconnecting. DriverHostActor re-projects the condition unacknowledged via
NativeAlarmProjector.TryRevertAcknowledge, but only while the local ack is
still pending - any driver transition projected since then wins. A disabled
or unspawned driver is not reverted (unchanged).
GatewayGalaxyAlarmAcknowledger now throws DriverOperationRefusedException on a
non-zero native hresult (wrong password, no ack permission) instead of only
warning, so that case reverts too; as a refusal it neither retries nor trips
the breaker.
Tests (each fails with its fix mutated out): projector revert/supersede,
DriverHostActor refused + failed ack -> unacked AlarmStateUpdate (successful
ack control), preflight view snapshot, node-manager preflight unit test,
over-the-wire ConditionMethodEventTests (BadUserAccessDenied, no route, no
event; positive control), Galaxy preflight + hresult refusal.
Docs: Galaxy.md, AlarmTracking.md, ScriptedAlarms.md, .claude/rules/alarms.md.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
HandlePublishError treated any anonymous BadNoSubscription (subscription id 0)
as loss of a tracked subscription whenever anything was tracked. A publish the
SDK queued while the session briefly held no subscription (the driver had just
deleted its last one) is answered BadNoSubscription and can be handled after the
replacement is registered, so a routine unsubscribe->resubscribe declared the new
subscription lost: Degraded health, a BadNoCommunication quality-only annotation
on every alarm condition and ConnectionLost (host re-initialise). That annotation
is the extra Unspecified event the #728 ConditionRefresh wire test saw.
- A BadNoSubscription that names a subscription id now counts only if the id is
tracked (same rule as BadSubscriptionIdInvalid).
- An anonymous BadNoSubscription within StaleNoSubscriptionGrace (5 s) of a
delete that left the driver tracking nothing is ignored (Debug log). The window
is stamped on removal and restarted when the delete completes; failed-subscribe
removes stamp it too; it resets with the session. A delete that leaves another
subscription tracked opens no window. A genuine loss inside the window is still
caught by the PublishingStopped heuristic; only the immediate ConnectionLost is
forgone.
- OpcUaClientStaleNoSubscriptionTests: stale error after alarm and data
resubscribe (fail without the fix), id-attributed stale error, plus genuine-loss
controls (after the grace, no recent delete, non-last delete, session release),
each mutation-checked.
- The ConditionRefresh wire test now records Kind, AlarmType, message, severity,
quality, restatement and timestamp of every post-resubscribe event on failure.
Wire test locally: 1/10 passing before (all 9 failures the 0x80310000
ConditionType quality-only annotation), 12/12 after.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
IOpcUaEndpointState (review-remediation lane C3) recorded whether the SDK started, but only the
health check read it: a fused node whose endpoint failed (port 4840 taken, ...) still advertised
250/240 and stayed driver Primary, so the Primary-gated data plane ran on a node no client could
reach while the reachable peer refused writes as Secondary.
- Move IOpcUaEndpointState/OpcUaEndpointState/OpcUaEndpointStatus from Host to Commons.OpcUa so
the Runtime can consume it.
- ServiceLevelCalculator: new NodeHealthInputs.EndpointListening (default true). false -> 0,
ahead of every other input including the +10 Primary bonus.
- OpcUaPublishActor: takes the endpoint state, feeds EndpointListening (Failed only; NotStarted /
Stopped are boot/shutdown windows) on the DB-backed and DB-less paths, and the legacy role-only
seam returns 0 for a failed endpoint too.
- New OpcUaEndpointStepDownActor (one per driver node): a Primary (ClusterActiveNode.SelfIsActive,
the election's own primitive) with a Failed endpoint gracefully Leaves its mesh - the manual
failover mechanism - so the peer becomes the oldest Up driver member, hosts the singletons and is
Primary. The election itself is not bent. Guarded on a peer having recently reported its own
endpoint listening (EndpointReport on the new opcua-endpoint-state DPS topic), so a pair with
both endpoints failed does not bounce the role and restart nodes forever, and a lone node never
shuts itself down.
- Runtime now references ZB.MOM.WW.Health.Akka (ClusterActiveNode).
- Docs: Redundancy.md tier table + new "Endpoint start failure: Primary step-down" section;
cluster-topology rule; HealthEndpoints / OpcUaEndpointHealthCheck comments (F-C1 landed).
Tests: ServiceLevelCalculatorTests (+5), OpcUaPublishActorEndpointStateTests (7: DB-less,
DB-backed, legacy seam, NotStarted/Stopped do not demote, a later failure demotes on the next
tick), OpcUaEndpointStepDownActorTests (3, real two-node mesh: holds while the peer is not
listening then leaves and the peer becomes Primary; failed Secondary and healthy Primary never
leave). Each was mutation-checked: removing the calculator/legacy guard, the Leave, the
peer-listening guard or the Primary guard fails the corresponding tests.
mxaccessgw 488649c+ acknowledges by AuthenticateUser as the operator and
writing <alarm>.AckMsg, so AcknowledgeAlarmRequest now needs a real Galaxy
operator_user + operator_domain + operator_password. The OPC UA data plane is
API-key/anonymous and carries no Galaxy password, so the driver resolves the
account per ack:
1. the session's API key's Galaxy ack identity, set on AdminUI /apikeys and
held ONLY in the encrypted Secrets store
(otopcua/apikeys/<keyId>/galaxy-ack-{user,password}); a key with a user but
no password secret is refused, never silently downgraded;
2. else the driver instance's AlarmAck fallback account (DriverConfig
FallbackUser/FallbackDomain/FallbackPasswordSecretRef; secret:/${secret:}/
env:/file:/dev: only — a literal is refused);
3. else DriverOperationRefusedException: nothing is sent, never an empty
operator_user. Refusals are excluded from the resilience retry/breaker.
- Pins ZB.MOM.WW.MxGateway.Client/Contracts 0.2.1 -> 0.3.0 (vendored into
nuget-packages/ from the Gitea feed, per the local-mxgw source mapping).
- The session ApiKeyId (not a secret) rides NativeAlarmAck ->
RouteNativeAlarmAck -> RouteAlarmAck -> AlarmAcknowledgeRequest; the password
is resolved in the driver and never travels through actor messages.
- The OPC UA principal + key id are appended to the AckMsg comment
("[OPC UA: <principal>, key <id>]") and logged, so a shared fallback
account does not lose who acknowledged.
- Password never logged: GalaxyAckCredential.ToString omits it, the
acknowledger scrubs it from every gateway message it echoes and withholds
any exception that echoes it; audit rows carry only the key id.
- AdminUI: /apikeys "Ack identity" editor (set/remove; delete-key clears it),
Galaxy driver form gains the fallback section.
- Synchronous OPC UA status for a refusal is follow-up #760 (the native ack
path is fire-and-forget after the Part 9 method returns).
Tests: request mapping (password populated), password absent from logs and
exception text (mutation-checked), key-mapped / fallback / refused paths at
resolver and driver level, breaker non-trip, key-id propagation through node
manager and DriverHostActor, AdminUI service + form round-trip.
Closes#749
Joseph Doherty