ScriptOptions.WithReferences(Assembly[]) resolves each assembly through
MetadataReference.CreateFromFile, which does not cache: every call mints a
fresh AssemblyMetadata -> PEReader -> NativeHeapMemoryBlock holding an
unmanaged copy of the assembly metadata that nothing disposes. Building the
options per compile therefore leaked native memory permanently — invisible to
the GC, to gcdump and to the managed allocation counters, so the working set
grew while the GC heap did not.
Diagnosed from a live dump of a wonder-app-vd03 Site node: 2,885 MB working
set 78 min after a cold start, only 150 MB live GC heap, ~2,469 MB on the
default process heap across ~6,700 undisposed AssemblyMetadata instances
against 473 DLLs on disk.
Three sites, all hoisted to static readonly:
- SiteRuntime ScriptCompilationService (the dumped one)
- InboundAPI InboundScriptExecutor — same defect on the central node; method
compiles recur on every re-registration and revision change
- CentralUI ScriptAnalysisService — CreateFromFile per sandbox run
ScriptAnalysis RoslynScriptCompiler also builds options per call but draws
from the static ScriptTrustPolicy.DefaultReferences, so it mints no metadata
and is left alone.
Guarded by reference-equality on the artifact rather than by watching memory:
a bytes-watching test would be flaky, and the leak is native so the managed
counters cannot see it at all. The test is proven to fail before the fix.
This does NOT close the AddTemplateScript OOM — that path was shown twice not
to compile scripts. It explains how a long-running node reaches a native
memory state where a large allocation fails with gigabytes free, which is a
lead worth re-testing, not a closure.
Also updates the now-obsolete CompileAndRegister_NonCompilingUpdate test
(InboundScriptExecutorTests.cs, not in the plan's Files list) to assert the
deliberate DB-authoritative behavior: a broken save now 500s consistently on
every node instead of the stale delegate silently serving.
Claude-Session: https://claude.ai/code/session_01MtdgwpEeCUn6cUA5f1LMPj
Remove project bookkeeping citations from shipped code comments across the
solution: hyphenated task IDs (WP-14, StoreAndForward-025), milestone/task/
issue refs (M3, Task 4, Audit Log #23, #21), Bundle X task-bundle labels,
and C/D/K/S/T phase labels.
Comment text only — no code logic, string/log literals, or XML-doc structure
changed. Genuine descriptions are preserved (only the citation is stripped),
and technical lookalikes are retained (UTF-8, SHA-256, T00:00:00, M365,
UTC-5, pre-C4/pre-C5 schema versions). Flagged by the new CommentChecker
TaskReferenceInComment / TrackingReferenceInComment checks plus targeted
grep passes; full solution builds clean, append-only guard tests pass.
CompileAndRegister only swaps the cached handler on a successful compile, so a
management (or DB) save of a non-compiling inbound script silently kept the
previously-registered handler serving — stale results, still HTTP 200 — while
UpdateApiMethod reported success. A broken save thus looked like a no-op, with
the real Roslyn error only in the central log.
Surface the diagnostics instead: a new CompileAndRegister(method, out errors)
overload reports the compile/trust-model errors (the bool overload is kept for
existing callers), and HandleCreate/UpdateApiMethod return them as a top-level
compileWarning on the management result (null when the script compiled). The
save still persists and the previously registered version keeps serving — the
warning just stops a non-compiling save from looking like a success.
- InboundScriptExecutor: Compile returns its diagnostics alongside the handler.
- ManagementActor: TryCompileAndDescribe + flat ApiMethodResult projection so
compileWarning rides at the top level without polluting the ApiMethod POCO.
- Component-InboundAPI.md: document the non-fatal-but-surfaced contract.
- Tests: executor diagnostics + non-compiling-update-keeps-previous-handler,
plus an end-to-end ManagementActor test asserting compileWarning is returned.
Resolves InboundAPI-026/027/028/029 (+ newly-surfaced -030).
- 026: authorize the scoped Database helper in the design doc; SQL-injection
protection is parameter binding (values never concatenated); allow writes via
ExecuteAsync; drop the false 'read-only' claim. Named connections only.
- 027: async ADO.NET end-to-end (no .GetAwaiter().GetResult()); honour the method
deadline token on ExecuteScalarAsync/ExecuteReaderAsync/ExecuteNonQueryAsync +
a CommandTimeout backstop derived from the method timeout.
- 028: negative-path tests (null-gateway, deadline cancellation, parameterization)
+ e2e Database + WaitForAttribute cases through the real endpoint.
- 029: WaitForAttribute is bounded by its WAIT timeout (per-wait CTS + client-abort
+ explicit token), NOT the method deadline (spec §6) — a long wait may outlive the
method timeout; WithRequestAborted threads the raw client-abort token separately.
- 030: Central UI compile-surface mirrors (InboundScriptHost / SandboxInboundScriptHost)
gained the Database member (drifted since the runtime helper was added) so the
authorized async API type-checks at the design-time gate.
Joseph Doherty