mt-tenants campaign tooling: awsbench product-surface MT mode + feeds… #354

Workflow file for this run

	name: ci

	on:
	push:
	branches: [main, slate]
	pull_request:
	schedule:
	# SR2-5: the noisy-neighbor campaign runs NIGHTLY (03:17 UTC), not
	# as a one-time recorded receipt. Locked thresholds live in the
	# script; a regression fails this workflow.
	- cron: "17 3 * * *"

	# Product-surface spec appendix §7: the required CI topology. The
	# dual-surface equivalence corpus, DST scenarios, and product
	# conformance run inside `cargo test --release` (dst::dst_tests);
	# request-cost budgets are asserted by the postings/planner tests in
	# the same suite. The upstream Durable Streams server suite runs as
	# its own job against a fresh namespace.

	jobs:
	# Job: rust build + full suite (product conformance, dual-surface
	# equivalence corpus, DST safety/liveness, request-cost budgets).
	rust:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- uses: dtolnay/rust-toolchain@stable
	- uses: Swatinem/rust-cache@v2
	- name: Check
	run: cargo check --all-targets
	# SR-6: the two multitenancy gates run as NAMED steps so a
	# failure is legible at a glance. The audit is the fingerprint
	# inventory (identity-neutral categories, baseline-diffed); the
	# syn lint is the zero-unmarked-residual security gate and also
	# runs inside the full suite — here it prints the reviewed
	# exemption-marker inventory into the CI log for certification.
	- name: Multitenancy audit (bare, never piped)
	run: bash scripts/multitenancy-audit.sh
	- name: Multitenancy identity lint (exemption inventory)
	run: cargo test --release --bin streams-slate multitenancy_identity_lint -- --nocapture
	- name: Test (product conformance + dual-surface + DST + cost budgets)
	run: \|
	cargo test --release 2>&1 \| tee /tmp/suite.log
	grep -E "^test result" /tmp/suite.log
	if grep -E "^test result" /tmp/suite.log \| grep -qv " 0 failed"; then
	echo "suite has failures"; exit 1
	fi
	# Job: the REAL enforce-mode noisy-neighbor campaign (SR2-5),
	# nightly. Reduced herd for the 4-core runner (24 noisy, 100 idle)
	# with the SAME locked thresholds; the contract-scale 48/200 run is
	# part of RC promotion (scripts/promote-rc.sh).
	noisy-campaign:
	if: github.event_name == 'schedule'
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- uses: dtolnay/rust-toolchain@stable
	- uses: Swatinem/rust-cache@v2
	- uses: actions/setup-node@v4
	with:
	node-version: 22
	- name: Noisy-neighbor campaign (locked thresholds)
	run: NOISY=24 IDLE=100 WINDOW_SECS=20 node scripts/mt-noisy-campaign.mjs

	# Job: the platform end-to-end battery (docs/CONTROL-PLANE-
	# INTEGRATION.md §15) — the TS emulator mints real credentials,
	# tokens and feed files; a REAL streams-slate consumes them under
	# the full release posture (workload identity, no static token);
	# the SDK exchanges through tokenProvider. Phase-A scenarios.
	platform-e2e:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- uses: dtolnay/rust-toolchain@stable
	- uses: Swatinem/rust-cache@v2
	- uses: actions/setup-node@v4
	with:
	node-version: 22
	- name: Build SDK dist
	run: cd sdk && npm ci && npm run build
	- name: Platform end-to-end battery
	run: node scripts/platform-e2e.mjs

	# Job: the shared-cell certification smoke at CONTRACT scale
	# (docs/MULTITENANCY.md §18 Stage 8 / SR-6). The suite's default run
	# exercises the same test at 128 projects; this job runs the
	# 1,000-project posture (48 noisy, shared workspaces, same-name
	# streams, suspension/revocation cutoffs, reconciliation, journaled
	# denials) on every push so the at-scale posture cannot rot.
	mt-cert-1000:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- uses: dtolnay/rust-toolchain@stable
	- uses: Swatinem/rust-cache@v2
	- name: Certification smoke at 1,000 projects
	run: MT_CERT_PROJECTS=1000 cargo test --release --bin streams-slate shared_cell_certification_smoke -- --nocapture

	# Job: the SDK ships as an npm package, so CI validates the PACKAGE,
	# not the source tree: build, pack, install the tarball into a clean
	# project, and run the end-to-end smoke through the package specifier
	# on the oldest supported runtime (engines: >=18) and on current Node.
	sdk-package:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- uses: dtolnay/rust-toolchain@stable
	- uses: Swatinem/rust-cache@v2
	- uses: actions/setup-node@v4
	with:
	node-version: 22
	- name: Typecheck + build + pack
	run: \|
	cd sdk
	npm ci
	npx tsc --noEmit -p tsconfig.json
	npm run build
	# npm >=10 does not create the destination directory; without
	# it, pack dies with ENOENT on the tarball path.
	mkdir -p /tmp/pkg
	npm pack --pack-destination /tmp/pkg
	- name: Install the tarball into a clean project
	run: \|
	mkdir -p /tmp/consumer && cd /tmp/consumer
	npm init -y >/dev/null
	npm install /tmp/pkg/prisma-streams-*.tgz
	node -e "require('node:assert').ok(require('node:fs').existsSync('node_modules/@prisma/streams/dist/index.js'))"
	- name: Build server
	run: cargo build --release --bin s3lite --bin streams-slate
	- name: Smoke through the package (current Node)
	run: \|
	./target/release/s3lite --listen 127.0.0.1:9500 --latency-ms 2 &
	sleep 1
	./target/release/streams-slate --listen 127.0.0.1:8090 \
	--s3-endpoint http://127.0.0.1:9500 --bucket sdk22-$RANDOM \
	--max-unflushed-bytes 67108864 &
	sleep 2
	cd /tmp/consumer
	STREAMS_SDK=@prisma/streams node $GITHUB_WORKSPACE/sdk/smoke.mjs http://127.0.0.1:8090
	- uses: actions/setup-node@v4
	with:
	node-version: 18
	- name: Smoke through the package (Node 18, the engines floor)
	run: \|
	./target/release/streams-slate --listen 127.0.0.1:8091 \
	--s3-endpoint http://127.0.0.1:9500 --bucket sdk18-$RANDOM \
	--max-unflushed-bytes 67108864 &
	sleep 2
	cd /tmp/consumer
	STREAMS_SDK=@prisma/streams node $GITHUB_WORKSPACE/sdk/smoke.mjs http://127.0.0.1:8091
	# The SDK claims Bun and Deno too, so gate them rather than assert
	# them. Both run the same package through the same smoke.
	- uses: oven-sh/setup-bun@v2
	- name: Smoke through the package (Bun)
	run: \|
	./target/release/streams-slate --listen 127.0.0.1:8092 \
	--s3-endpoint http://127.0.0.1:9500 --bucket sdkbun-$RANDOM \
	--max-unflushed-bytes 67108864 &
	sleep 2
	cd /tmp/consumer
	STREAMS_SDK=@prisma/streams bun $GITHUB_WORKSPACE/sdk/smoke.mjs http://127.0.0.1:8092
	- uses: denoland/setup-deno@v2
	with:
	deno-version: v2.x
	- name: Smoke through the package (Deno)
	run: \|
	./target/release/streams-slate --listen 127.0.0.1:8093 \
	--s3-endpoint http://127.0.0.1:9500 --bucket sdkdeno-$RANDOM \
	--max-unflushed-bytes 67108864 &
	sleep 2
	cd /tmp/consumer
	STREAMS_SDK=@prisma/streams deno run -A --node-modules-dir \
	$GITHUB_WORKSPACE/sdk/smoke.mjs http://127.0.0.1:8093

	# Job: the pinned upstream Durable Streams server conformance suite,
	# UNMODIFIED, against a fresh namespace (spec appendix §7). The
	# version here must match src/protocol_pin.rs.
	durable-streams-server-conformance:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- uses: dtolnay/rust-toolchain@stable
	- uses: Swatinem/rust-cache@v2
	- uses: actions/setup-node@v4
	with:
	node-version: 22
	- name: Build server
	run: cargo build --release --bin s3lite --bin streams-slate --bin streams-keys
	- name: Run pinned suite (fresh namespace)
	run: \|
	./target/release/s3lite --listen 127.0.0.1:9500 --latency-ms 2 &
	sleep 1
	KEY=$(./target/release/streams-keys generate)
	# Group commit is required: the suite's property tests run
	# ~240 sequential appends inside vitest's 5 s budget, and the
	# default flush cadence makes them latency-marginal. This is
	# what field deployments run anyway (see CONFORMANCE.md).
	./target/release/streams-slate --listen 127.0.0.1:8090 \
	--s3-endpoint http://127.0.0.1:9500 --bucket conf-$RANDOM \
	--max-unflushed-bytes 67108864 \
	--flush-interval-ms 1 --wal-flush-gap-ms 2 \
	--conformance-default-key "$KEY" &
	sleep 2
	# The runner lives in conformance/ with a committed lock and an
	# exact expected outcome. It does NOT use the package's npx
	# CLI, whose include glob misses its own runner and exits 0
	# having run nothing.
	cd conformance
	npm ci
	CONFORMANCE_TEST_URL=http://127.0.0.1:8090 npm test

	# Job: the product field corpus (scripts/field-gate.mjs) — the same
	# 20 checks the cloud campaign runs, against a local server with
	# aggressive split thresholds so a real scaler split fires inside CI.
	# Negative auth, raw create replay, product seal replay, fork replay,
	# a real automatic split, and raw/product default-key isolation.
	product-field-gate:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- uses: dtolnay/rust-toolchain@stable
	- uses: Swatinem/rust-cache@v2
	- uses: actions/setup-node@v4
	with:
	node-version: 22
	- name: Build server
	run: cargo build --release --bin s3lite --bin streams-slate
	- name: Run the field gate (auth enabled, deterministic split)
	run: \|
	./target/release/s3lite --listen 127.0.0.1:9500 --latency-ms 2 &
	sleep 1
	# Low, deterministic scaler thresholds: one hot evaluation
	# over a 10 s window splits, so the "load drove a real split"
	# check fires within the gate's own load phase instead of
	# racing production cadence.
	SCALE_EVAL_SECS=5 SCALE_RATE_WINDOW_SECS=10 \
	SCALE_HOT_PCT=1 SCALE_HOT_EVALS=1 SCALE_COOLDOWN_SECS=5 \
	./target/release/streams-slate --listen 127.0.0.1:8092 \
	--s3-endpoint http://127.0.0.1:9500 --bucket fg-$RANDOM \
	--max-unflushed-bytes 67108864 \
	--flush-interval-ms 1 --wal-flush-gap-ms 2 \
	--auth-token ci-gate-token &
	sleep 2
	STREAMS_TOKEN=ci-gate-token node scripts/field-gate.mjs http://127.0.0.1:8092

Navigation Menu

Uh oh!

mt-tenants campaign tooling: awsbench product-surface MT mode + feeds… #354

Workflow file

mt-tenants campaign tooling: awsbench product-surface MT mode + feeds… #354

Uh oh!

Workflow file for this run