ci: shard the portability job into three so each fits the forge's 30-minute cap
Some checks failed
CultMesh portability and performance / ubuntu-latest cache owners (pull_request) Failing after 52s
CultMesh portability and performance / ubuntu-latest mesh authority and content parity (pull_request) Failing after 6m56s
CultMesh portability and performance / ubuntu-latest reactive semantics and leases (pull_request) Successful in 17m42s

Runs 902, 904 and 905 were cancelled at 30:00 with one job holding every step. The steps now split by owner:
reactive semantics and leases, cache owners (TypeScript, Python, Rust and C# caches, packed packages) and mesh authority and content parity.
Every step still runs; each shard repeats the setup.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This commit is contained in:
Libby 2026-11-13 22:08:05 +00:00
commit 5fb1a7a921

View file

@ -23,14 +23,19 @@ concurrency:
jobs:
reactive-and-resource-lifetime:
name: ${{ matrix.os }} reactive semantics and leases
name: ${{ matrix.os }} ${{ matrix.shard }}
runs-on: ${{ matrix.os }}
# The job dies at the lower of this value and the forge runner's own cap, which Eureka is raising from 30 to 60 minutes
# (run 1378 carried 60 and was still cancelled at 30:02); the two must both say 60 before a full run can finish.
timeout-minutes: 60
# The forge runner cancels a job at 30:00 and one undivided job ran 22 to 30 minutes, so the steps are sharded across
# three jobs (matrix.shard) that each fit well under it; the forge runs them one after another. Every shard repeats
# the checkout and toolchain setup (about 4 minutes) so that no shard depends on another's output.
timeout-minutes: 30
strategy:
fail-fast: false
matrix:
shard:
- reactive semantics and leases
- cache owners
- mesh authority and content parity
# The forge runner (label `ubuntu-latest`) is Linux-only and no Windows runner exists: the Windows leg runs on workflow_dispatch only (GitHub or a Windows runner), the ubuntu-latest leg gates push and PR. Windows push/PR triggers return in the PR that registers a Windows runner.
os: ${{ fromJSON(github.event_name == 'workflow_dispatch' && '["windows-latest", "ubuntu-latest"]' || '["ubuntu-latest"]') }}
@ -101,7 +106,7 @@ jobs:
# The forge runner images carry no pwsh; GitHub's Windows and hosted images do.
- name: Install PowerShell
if: runner.os == 'Linux'
if: runner.os == 'Linux' && matrix.shard == 'reactive semantics and leases'
shell: bash
run: |
set -euo pipefail
@ -117,14 +122,17 @@ jobs:
DOTNET_SYSTEM_GLOBALIZATION_INVARIANT=1 pwsh -Version
- name: Verify explicit reactive scheduling across runtimes
if: matrix.shard == 'reactive semantics and leases'
shell: pwsh
run: ./scripts/verify-reactive-document-scaling.ps1
- name: Measure representative reactive workload
if: matrix.shard == 'reactive semantics and leases'
shell: pwsh
run: ./scripts/measure-reactive-document-performance.ps1 -Quick -SkipRuntimeBuild
- name: Churn 100,000 distinct document leases
if: matrix.shard == 'reactive semantics and leases'
run: >-
dotnet test tests/GameCult.Mesh.Tests/GameCult.Mesh.Tests.csproj
--filter FullyQualifiedName~Document_OneHundredThousandDistinctLeaseCyclesLeaveNoResourcesOrCallbacks
@ -143,20 +151,24 @@ jobs:
# and unrun. The wiring check fails when any TypeScript test file is run by no script, and the cache's own suite
# runs here beside the packages built on it.
- name: Check that every TypeScript test file is run by its package
if: matrix.shard == 'cache owners'
run: node scripts/check-typescript-tests-wired.mjs
- name: Verify the CultCache document owner with the TypeScript cache
if: matrix.shard == 'cache owners'
run: npm run test --workspace packages/cultcache-ts
# The Python cache's unit suite ran only in cultnet-interop.yml (windows-latest, dispatch-only), so on Linux
# it ran nowhere. It sits after the TypeScript cache's test so the built cultcache-ts store holder exists for the
# cross-runtime store-lock test, which skips without it.
- name: Verify the CultCache document owner with the Python cache
if: matrix.shard == 'cache owners'
run: python -m unittest discover -s packages/cultcache-py/tests
# cargo test for packages/cultcache-rs ran in no workflow: the Rust cache's own unit and doc tests were unwatched
# (only the interop example built, inside the C# store tests). The target directory is under RUNNER_TEMP, away from the checkout.
- name: Verify the CultCache document owner with the Rust cache
if: matrix.shard == 'cache owners'
shell: bash
run: CARGO_TARGET_DIR="$RUNNER_TEMP/cultcache-rs-target" cargo test --locked --manifest-path packages/cultcache-rs/Cargo.toml
@ -166,14 +178,17 @@ jobs:
# culture and throws CultureNotFoundException; it is excluded here and still runs wherever ICU is present
# (variants:follow_up:culture-test-invariant-image).
- name: Verify the CultCache document owner with the C# reference
if: matrix.shard == 'cache owners'
run: dotnet test tests/GameCult.Caching.Tests/GameCult.Caching.Tests.csproj --filter "FullyQualifiedName!~ADeterministicIdIsTheSameUnderEveryCulture" --logger "console;verbosity=minimal"
# Packs the four TypeScript packages and installs them into a clean consumer, so a package that builds but does
# not install or load fails here.
- name: Verify the packed TypeScript packages install and load from a clean consumer
if: matrix.shard == 'cache owners'
run: npm run test:pack:ts
- name: Verify the CultMesh authority rules with the TypeScript verifier
if: matrix.shard == 'mesh authority and content parity'
run: npm run test --workspace packages/cultnet-ts
# cultmesh-browser's test compiles test/interop, which imports cultmesh-ts/content, so cultmesh-ts builds first,
@ -186,9 +201,11 @@ jobs:
# in CI on any OS until these two steps existed, so a codec that agreed
# only with itself would have gone green.
- name: Verify the realtime frame codec and the content provider with the TypeScript runtime
if: matrix.shard == 'mesh authority and content parity'
run: npm run test --workspace packages/cultmesh-ts
- name: Verify the browser client against the shared TypeScript verifier
if: matrix.shard == 'mesh authority and content parity'
run: npm run test --workspace packages/cultmesh-browser
# Both directions of the CultMesh authority vectors: the C#-written
@ -198,6 +215,7 @@ jobs:
# in the step above. CULTMESH_WRITE_VECTORS is never set in CI: the
# committed files are the contract.
- name: Verify the CultMesh authority vectors with the C# reference
if: matrix.shard == 'mesh authority and content parity'
run: >-
dotnet test tests/GameCult.Mesh.Tests/GameCult.Mesh.Tests.csproj
--filter FullyQualifiedName~AuthorityProof
@ -207,6 +225,7 @@ jobs:
# re-encodes the vector file the TypeScript codec wrote. A fixture
# authored on one side alone would only agree with itself.
- name: Verify the realtime frame vectors with the C# reference
if: matrix.shard == 'mesh authority and content parity'
run: >-
dotnet test tests/GameCult.Mesh.Tests/GameCult.Mesh.Tests.csproj
--filter FullyQualifiedName~RealtimeWireProtocol
@ -218,6 +237,7 @@ jobs:
# content parity, which no CI step ran before this one. CULTNET_WRITE_VECTORS
# is never set in CI: the committed files are the contract.
- name: Verify the content vectors with the C# reference
if: matrix.shard == 'mesh authority and content parity'
run: >-
dotnet test tests/GameCult.Mesh.Tests/GameCult.Mesh.Tests.csproj
--filter FullyQualifiedName~CultMeshContentVectorTests
@ -227,4 +247,5 @@ jobs:
# TypeScript reader fetch from the TypeScript provider in its own process. Both refuse bytes tampered in flight,
# and the provider refuses a corrupt store.
- name: Verify HTTPS content parity across C#, the TypeScript provider and the TypeScript reader
if: matrix.shard == 'mesh authority and content parity'
run: npm run test:interop --workspace packages/cultmesh-browser