OpenCue Scheduler Stress Pipeline #4

Workflow file for this run

.github/workflows/scheduler-stress-pipeline.yml at e47e327

	name: OpenCue Scheduler Stress Pipeline

	# Runs the scheduler booking + accounting stress suite
	# (rust/crates/scheduler/tests/stress_tests.rs): a full pipeline::run against a
	# seeded farm, with an end-of-run audit that cross-checks the Redis acct:*
	# hashes against SUM(proc) in Postgres and asserts cap enforcement.
	#
	# When it runs — and when it deliberately doesn't:
	# - Pull requests: only when the scheduler crate, its proto dependency, the
	# DB migrations, or this workflow change. The suite needs a migrated
	# Postgres plus a Redis container and takes several minutes — running it
	# for Python/CueGUI/docs changes would burn runner time for zero signal.
	# - Nightly on master: catches drift from changes that slipped past the
	# paths filter (e.g. shared workspace dependencies) and gives a daily
	# throughput data point under fixed scale.
	# - Manually (workflow_dispatch): for benchmarking a branch at custom scale.
	#
	# What is a gate vs. what is informational:
	# - The job FAILS on correctness regressions: accounting drift between Redis
	# and Postgres, cap breaches (subscription burst / job max-cores), booking
	# liveness (<90% drain, no saturation rejections), or leftover test data.
	# - The throughput numbers (frames/s) are reported in the step summary but
	# are NOT asserted on: shared runners are too noisy for perf gating. For
	# real benchmarking run the suite locally in release mode (see
	# docs/_docs/developer-guide/scheduler-stress-testing.md).

	on:
	pull_request:
	branches: ["master"]
	paths:
	- "rust/crates/scheduler/**"
	- "rust/crates/opencue-proto/**"
	- "rust/Cargo.toml"
	- "cuebot/src/main/resources/conf/ddl/postgres/migrations/**"
	- ".github/workflows/scheduler-stress-pipeline.yml"
	schedule:
	# Nightly on master. Odd minute to avoid the top-of-hour scheduling rush.
	- cron: "23 9 * * *"
	workflow_dispatch:
	inputs:
	stress_jobs:
	description: "Drain-phase job count (default 300)"
	required: false
	stress_hosts:
	description: "Drain-phase host count (default 1200)"
	required: false
	stress_frames_per_layer:
	description: "Drain-phase frames per layer (default 5)"
	required: false
	stress_timeout_secs:
	description: "Per-phase hard timeout in seconds (default 600)"
	required: false

	concurrency:
	group: ${{ github.workflow }}-${{ github.ref }}
	cancel-in-progress: true

	env:
	CARGO_TERM_COLOR: always
	CARGO_INCREMENTAL: 0
	CARGO_NET_RETRY: 10
	RUSTUP_MAX_RETRIES: 10

	jobs:
	stress:
	runs-on: ubuntu-22.04
	timeout-minutes: 45

	# The suite's Postgres half. Redis is NOT listed here on purpose: the test
	# starts its own throwaway Redis via testcontainers, so all accounting
	# state is guaranteed to die with the test process.
	services:
	postgres:
	image: postgres:15.1
	env:
	POSTGRES_USER: cuebot
	POSTGRES_PASSWORD: cuebot_password
	POSTGRES_DB: cuebot
	ports:
	- 5432:5432
	options: >-
	--health-cmd "pg_isready -U cuebot -d cuebot"
	--health-interval 5s
	--health-timeout 5s
	--health-retries 10

	steps:
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
	with:
	# The suite only reads the repo (no git push), so don't leave the
	# token on disk for later steps.
	persist-credentials: false

	- name: Install build dependencies
	run: \|
	sudo apt-get update && sudo apt-get install -y libx11-dev protobuf-compiler libcurl4-openssl-dev postgresql-client
	curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs > rust-install.sh
	bash ./rust-install.sh -y

	- name: Cache cargo deps
	uses: Swatinem/rust-cache@c19371144df3bb44fab255c43d04cbc2ab54d1c4 # v2.9.1
	with:
	workspaces: rust
	cache-on-failure: true

	# Plain psql instead of the sandbox Flyway image: the migrations are
	# versioned plain-SQL files, so applying them in numeric order is exactly
	# what Flyway would do, without building a JDK image first.
	- name: Apply database migrations
	working-directory: cuebot/src/main/resources/conf/ddl/postgres/migrations
	env:
	PGPASSWORD: cuebot_password
	run: \|
	for f in $(ls V*.sql \| sort -t V -k2 -n); do
	echo "== $f"
	psql -q -v ON_ERROR_STOP=1 -h localhost -U cuebot -d cuebot -f "$f"
	done

	- name: Run stress suite
	working-directory: rust
	shell: bash
	env:
	STRESS_JOBS: ${{ inputs.stress_jobs }}
	STRESS_HOSTS: ${{ inputs.stress_hosts }}
	STRESS_FRAMES_PER_LAYER: ${{ inputs.stress_frames_per_layer }}
	STRESS_TIMEOUT_SECS: ${{ inputs.stress_timeout_secs }}
	run: \|
	cargo test -p scheduler --features stress-tests --test stress_tests -- --nocapture 2>&1 \| tee stress-output.log

	- name: Publish phase report
	if: always()
	working-directory: rust
	shell: bash
	run: \|
	if [ -f stress-output.log ] && grep -q "^================ phase" stress-output.log; then
	{
	echo "## Scheduler stress suite"
	echo '```'
	sed -n '/^================ phase/,$p' stress-output.log \| sed -n '1,80p'
	echo '```'
	echo "_Throughput numbers are informational; only the accounting/enforcement assertions gate this job._"
	} >> "$GITHUB_STEP_SUMMARY"
	else
	echo "## Scheduler stress suite: no phase report produced (failed before the run?)" >> "$GITHUB_STEP_SUMMARY"
	fi

	- name: Upload full output
	if: always()
	uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
	with:
	name: scheduler-stress-output
	path: rust/stress-output.log
	if-no-files-found: ignore
	retention-days: 30

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

OpenCue Scheduler Stress Pipeline #4

Workflow file

OpenCue Scheduler Stress Pipeline #4

Uh oh!

Workflow file for this run