Skip to content

AGENTS.md

Latest commit

 

History

History
41 lines (29 loc) · 7.54 KB

AGENTS.md

File metadata and controls

41 lines (29 loc) · 7.54 KB
Raw

KeyLoad.AppHost

Purpose and entry points

  • Owns Aspire application composition for KeyLoad services and infrastructure resources.
  • Entry point: Program.cs; benchmark resource composition: Features/BenchmarkComparisons/Resources/BenchmarkResources.cs.

Ownership and boundaries

  • Keep host wiring and genuinely shared infrastructure composition here. Feature-specific deployment behavior belongs under Features/<SliceName>/ and the matching docs/Features/<SliceName>.md.
  • Aspire owns resource orchestration, not database storage state or atomic partition identity. Each node-local PartitionHost owns storage, journals, file locks and the apply gate; Orleans grains route commands and physical placement remains separate from atomic partition identity.
  • ClusterResources composes actual Docker RF3 resources. Do not replace the required Docker/Aspire RF3 topology with an in-memory or single-node demonstration. The accepted ADR-082 protocol-cohort test seam preserves fixed three-voter membership and node-local storage ownership.

Commands and evidence

  • The canonical Aspire-owned test entry is defined by root AGENTS.md and ADR-074. AppHost defaults to Release because the pinned Aspire CLI evaluates RunCommand without forwarding outer dotnet-run configuration; verify the actual launched path and native TUnit outcomes. Do not bypass CLI/DCP or count stale Debug output as current qualification.
  • GitHub Actions solution build: dotnet build KeyLoad.slnx --no-restore --configuration Release.
  • Owner correction2026-10-03 requires every test caller, including integration/RF3, to use dotnet run --project src/KeyLoad.AppHost --no-build --no-restore --configuration Release -- --KeyLoadTests:Suite=<suite>. Native dotnet test is only the owned child process; preserve all suites and original artifacts. Delivered-source qualification remains actual Linux GitHub Actions evidence.

Skills and protected risks

  • Applicable skills: none installed; skill installation is prohibited by owner direction.
  • Owner correction2026-10-03 explicitly permits local development through the same Aspire-owned entry. Preserve secret boundaries, resource health dependencies, scoped fault injection, complete resource shutdown and the required Orleans RF3 topology; local results do not qualify delivered-source Linux or public performance gates.

Read-first and canonical slice ownership

  • Read the root policy, architecture map, RepositoryGovernance feature, and ADR-032 first.
  • Owned composition slices: ClusterReplication, ClusterRouting, ClientApi, TestInfrastructure and BenchmarkComparisons; target paths use the same Features/<SliceName>/ convention. ClusterRouting's protocol image overrides follow NativeCqrsRequestV2 and ADR-082; they are allowed only in an explicitly selected ephemeral fixed-three-voter child test topology, never benchmark or ordinary production composition.
  • Features/ClientApi/Resources/ClusterConnectionEndpointResources.cs owns the named direct HTTP/2 container endpoint under ADR-125, preserving original HTTP/1.1 endpoints, peer identity, readiness and genuine RF3 resources. Actual parallel-connection SDK/official MCP tests discover this endpoint through Aspire; no proxy or synthetic session can substitute for its accepted TCP connection.
  • Keep Program.cs as the shared Aspire composition entry point; put feature-owned resource definitions under their named slice.
  • Features/CodeQuality/ owns the test-only native functional coverage configuration and Aspire preparation/resource dependencies under ADR-033 and CodeQuality AC-CQ-039..042. Use the same centrally pinned Microsoft engine as MTP, downloaded as a package; do not install a global tool. Preserve exact contributor/source/DLL/PDB/image identity, original process settlement and all full mandatory suites. Runtime coverage still comes from the original owned server processes, not the AppHost assembly.

Vertical-slice responsibility folders

  • Keep feature-owned implementation inside its canonical Features/<SliceName>/ and organize it in populated, feature-local responsibility folders (such as Models/, Contracts/, Commands/, Queries/, Transport/, Hosting/, Serialization/, or Validation/). Do not leave a flat dump of unrelated responsibilities at the slice root; keep only genuinely shared building blocks and executable/composition entry points outside feature slices. Preserve namespaces and runtime contracts during physical moves.

Native TUnit entry, owner correction 2026-10-07

  • ADR-117 supersedes the earlier outer AppHost caller requirements: CI starts TUnit directly after build with Detailed output. Test fixtures own Aspire infrastructure startup, readiness, client operations and cleanup. scripts/Features/TestInfrastructure/run-tests.mjs only selects native test arguments/environment; it cannot execute database workloads. RF3 coverage preparation belongs to the TUnit session lifecycle. Preserve every original qualification/artifact gate and separate Benchmarks ownership.

TASK-TEST-APPHOST-NATIVE-SIMD-001 — current normal-only suite composition

Owner2026-10-11 mandatory native SIMD supersedes every earlier active unit-scalar/runner-disabled-intrinsics clause, including REQ/AC-TEST-009 and historical suite examples. Immutable original scalar reports remain history, never current qualification. Current TestSuiteProtocol and ProjectForSuite admit only analyzers/unit/recovery/rf3/comparison/site; former unit-scalar is unknown and refuses before resource creation. TestSuiteResources removes only its explicit DOTNET_EnableHWIntrinsic=0 runner branch; native50 ordinary/exclusiveheavy1, explicit supported parallelism, original comparison isolation1, all project/argument/TRX/coverage/path selection and genuine Aspire preparation/readiness/clients/shutdown remain unchanged. No old ignored alias, scalar alternate or limits/configuration fallback.

Native audit proves TestSuiteApplication, TestSuiteSettingsReader and typed execution policy have no scalar-specific behavior, so their cancellation, output/process joins and failure ledgers remain byte-exact. Existing AspireTestEntryModelTests same declarations verify the complete six-suite model, exact native argument/project/results shape and absent injected disable environment. FilterIsOneNativeArgumentAndIsBoundedBeforeResourcesAreAdded now refuses old unit-scalar before its original genuine healthy model; full negative and healthy controls/disposal remain. Its independent default ordinary literal is50, while the explicit rejected comparison20 configuration remains independently20. This source-owned model remains in ComparisonTests and executes only through authorized GitHub Benchmarks under its local policy; no local benchmark workload is started. Existing FunctionalTestParallelismTests/NativeTUnitSelectionTests normal controls and final all-source compiler/format/native census remain mandatory.

Ordered source owners: this spec/local AppHost policy/ADR first; exact protocol and selection retirement; resource branch retirement; original complete model assertions. Root alone joins/builds/executes live inputs. No product/public/persisted ID or storage-format change, migration, clock/deadline/resource increase. Rollback cannot reinstate the superseded scalar mode; preserve originals and repair the same normal-native contract. Current execution/coverage/Linux qualification is OPEN until actual source/DLL/PDB/discovery/reports/cleanup bind. Native source-path/count arithmetic is not UID or PASS.