Wrap the script compile-cost guardrail block in its own inner try/catch so a
transient SQL failure on ToListAsync cannot fall through to the outer catch and
produce a Rejected reply for an otherwise-valid deploy. advisory is declared in
the outer scope so the Accepted StartDeploymentResult Message is unaffected on
the happy path; the inner catch logs a Warning and leaves advisory null.
Swap MemProbe's ProjectReference from Core.VirtualTags to
Core.Scripting.Abstractions so the heavy-mode globalsType
(ScriptGlobals<VirtualTagContext>) resolves from the post-A0
Roslyn-free assembly.
Measured results (2026-06-07, N=50, Release):
heavy (post-A0): 2.40 / 2.53 MiB/script (was ~18 MiB)
lean: 1.64 / 1.65 MiB/script
=> heavy ≈ lean; both well under the 3 MiB gate. PASS.
Filter ExternalIdReservations to WHERE ReleasedAt IS NULL so
DraftSnapshot.ActiveReservations matches its documented semantics and
ValidateReservationPreflight cannot emit spurious BadDuplicateExternalIdentifier
errors from already-released rows. Adds a focused unit test seeding one active
and one released reservation and asserting only the active row is returned.
ParseComposition(blob, nodeId, onInconsistency?) detects a kept equipment whose
UNS line belongs to another cluster (a same-cluster-invariant violation that
would orphan the equipment folder) and reports it via an optional callback,
wired to OpcUaPublishActor's logger. Detection-only; the upstream draft
validator remains the authority. Adds two unit tests.
Adds a 'migrator' Dockerfile stage + Compose service that runs 'dotnet ef
database update' once on bring-up, so a fresh SQL volume gets the schema with no
operator step (quirk 1). cluster-seed + every host node depend on it via
service_completed_successfully, so the seed never races an in-progress migration
(quirk 2). Host build pinned to target: runtime (the migrator is now the last
stage). entrypoint + README updated; the manual 'dotnet ef' first-time step is
gone. Verified: down -v + up --build self-bootstraps (migrator+seed exit 0,
6 nodes up), deploy Sealed 6/6.
Aligns ConfigPublishCoordinator's _acks/_expectedAcks with the case-insensitive
ClusterId/NodeId scoping in DeploymentArtifact.ResolveClusterScope, so an ack
from a node whose host:port differs only in case still matches its expected-ack
entry (SQL collation + DNS are case-insensitive).
Log a WARNING on startup when IVirtualTagEvaluator is not registered so a DI misconfig on a
driver-role node is visible in logs instead of silently evaluating all VirtualTags to NoChange.
Add a comment in PushDesiredSubscriptions noting that TryRecoverFromStale does not call this
method, so VirtualTags remain empty after a Stale recovery until the next deployment dispatch
(intentional, consistent with driver recovery).
Context.Watch each spawned child; OnChildTerminated evicts it from _children so the
next ApplyVirtualTags (still containing that vtagId) falls through the ContainsKey
guard and re-spawns a fresh VirtualTagActor. Adds a spawn-site Debug log, moves the
TODO about in-place plan mutation to the skip-existing branch where it belongs, and
adds a deterministic TestKit test (Child_is_respawned_after_unexpected_termination)
that kills the first child, drains its UnregisterInterest from the mux probe, re-applies,
and asserts a second distinct RegisterInterest arrives.
IReadOnlyList<string> DependencyRefs compared by reference in the auto-generated record
equality, causing every VirtualTag with dependencies to be flagged "Changed" on every parse
(fresh list instances from composer and artifact-decoder). Add Equals/GetHashCode overrides
with element-wise ordinal comparison so Phase7Plan.IsEmpty short-circuits a no-op redeploy.
Add regression test Identical_virtualtag_snapshots_diff_to_empty_plan (separate list instances,
same contents → IsEmpty true). Add TODO comment in Phase7Applier near needsRebuild predicate.
Joseph Doherty