open-standards/.github/workflows/wai-quantum-npu.yml
Transaction Science 6517401f01 CI: the gate that decides whether a workflow runs was queued behind the work it gates
Every path-gated workflow begins with one small job -- `does this change touch
it`, or `which workspaces this change touches` for the standards build -- that
all of its real jobs declare in `needs`. All 55 of those gate jobs ran on
`linux-arm64`.

That is the one label the fleet cannot spare. Sampled while 267 runs were
queued, every waiting run had the same shape:

    [waiting]  linux-arm64    does this change touch it
    [blocked]  linux          member · dataflow device, simulated
    [blocked]  macos          member · dataflow device absent

237 runs, each holding a gate in the longest queue, and every real job blocked
behind it. Not only `linux` jobs: `macos`, `linux-hw`, `linux-arm64-hw` and
`macos-m3-ultra` jobs were blocked too. The measured effect was `linux-arm64`
263 deep against `linux` **zero**, with jetson-hub's 20 cores taking no job for
a full hour -- starved not for want of work but because the work could not be
unblocked. A gate waiting on the resource it is rationing is head-of-line
blocking, and it makes every other runner in the fleet idle on its behalf.

The gates move to `linux`. Nothing else changes: the diff is 55 `runs-on` lines
and no others, and every touched file still parses.

Why `linux` and not elsewhere. The gate is not free -- it compares
`BASE...HEAD`, which needs the merge base, so `fetch-depth: 0` is a full clone
rather than a ten-second script. `macos` is served by hosts on a Mac Studio
sitting at load 185; `linux` is a separate 20-core box that was measurably
empty. This moves a serialisation point to the least contended label; it does
not remove it, and the gate is still a single point of failure for its
workflow. The structural fix is an early-exit first step inside each job, which
costs one checkout per leg and is a larger change.

Follow-up, deliberately not bundled: jetson-hub runs at capacity 2 with 15 GB of
RAM, so 55 clones still serialise two at a time. Raising capacity helps gates
and hurts the heavy emulated-architecture jobs that share the label, so it wants
measuring on its own.

Co-Authored-By: Claude Opus 5 <[email protected]>
2026-11-03 15:19:13 -05:00

119 lines
5.5 KiB
YAML
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

name: wai-quantum neural-processor member
# The fabric's neural-processor member: exact 32×32 products from 4-bit limbs,
# every limb product and limb sum formed in fp16 on the neural processor (each
# value an integer fp16 holds exactly), the host assembling; certified
# bit-exact against the CPU reference's pinned corpus digests.
#
# - npu-mac (the macOS runner): the platform's model runtime must place every
# operation of the model on the neural processor (its compute plan says so,
# or the member is refused); the device tests must find the declared
# processor, agree with the CPU on the probe and on random states, reproduce
# all four pinned certificates, and catch both model mutants.
# - npu-absent (a Linux container): the same build finds no member and says
# why; the device tests run over no member with the declaration `none`.
concurrency:
group: ${{ github.workflow }}-${{ github.ref }}
cancel-in-progress: true
on:
pull_request:
paths:
- 'wai/wai-quantum/**'
- 'rust-toolchain.toml'
- '.github/workflows/wai-quantum-npu.yml'
schedule:
- cron: '22 4 * * *'
jobs:
# Does this pull request's own diff touch the paths below? This forge matches a
# workflow's `paths:` against the push, so a merge of main into a branch fires the
# workflow whatever the pull request changes; every job here waits on this answer.
# A push to main, a schedule, or anything the script cannot determine answers
# `true` (scripts/pr-touches-paths.py).
relevant:
name: does this change touch it
runs-on: linux
outputs:
run: ${{ steps.gate.outputs.run }}
steps:
- uses: actions/checkout@v5
with:
fetch-depth: 0
- id: gate
run: |
set -euo pipefail
r=$(python3 scripts/pr-touches-paths.py .github/workflows/wai-quantum-npu.yml \
"${{ github.event.pull_request.base.sha }}" \
"${{ github.event.pull_request.head.sha }}")
echo "this change touches wai-quantum-npu: $r"
echo "run=$r" >> "$GITHUB_OUTPUT"
npu-mac:
name: member · neural processor (macos)
needs: relevant
if: needs.relevant.outputs.run == 'true'
runs-on: macos
timeout-minutes: 120
env:
WAI_QUANTUM_NPU_EXPECT: npu-0
steps:
- uses: actions/checkout@v5
# launchd's PATH omits cargo on the macOS host runner.
- name: Put cargo on PATH
shell: bash
run: echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
- name: The member's unit tests
shell: bash
working-directory: wai/wai-quantum
run: |
set -euo pipefail
nice -n 10 cargo test --locked --features quantum_fabric_npu --lib -- quantum_fabric_npu:: quantum_fabric::npu:: 2>&1 | tee "$RUNNER_TEMP/unit.log"
grep -qE '^test result: ok\. [1-9][0-9]* passed; 0 failed' "$RUNNER_TEMP/unit.log" || { echo "::error::no unit test ran"; exit 1; }
- name: The member's device tests
shell: bash
working-directory: wai/wai-quantum
run: |
set -euo pipefail
report="$RUNNER_TEMP/npu-report.txt"; : > "$report"
export WAI_QUANTUM_NPU_CACHE="$RUNNER_TEMP/npu-cache"
WAI_QUANTUM_NPU_REPORT="$report" nice -n 10 \
cargo test --locked --release --features quantum_fabric_npu --test npu_member -- --test-threads=1 2>&1 | tee "$RUNNER_TEMP/device.log"
grep -qE '^test result: ok\. 9 passed; 0 failed' "$RUNNER_TEMP/device.log" || { echo "::error::not every device test ran and passed"; exit 1; }
cat "$report"
key="$WAI_QUANTUM_NPU_EXPECT"
grep -qx "npu devices: as declared ($key)" "$report" || { echo "::error::the processor found is not the declared one"; exit 1; }
grep -qE "^npu placement: key=$key( [a-z0-9_.]+=MLNeuralEngineComputeDevice)+$" "$report" || { echo "::error::an operation was not placed on the neural processor"; exit 1; }
n="$(grep -c "^npu certified: key=$key .* parity=bit-exact " "$report")"
[ "$n" = 4 ] || { echo "::error::$n of 4 certificates were earned"; exit 1; }
grep -qx "npu mutants: key=$key killed 2 of 2" "$report" || { echo "::error::a model mutant survived"; exit 1; }
npu-absent:
name: member · neural processor absent (linux)
needs: relevant
if: needs.relevant.outputs.run == 'true'
runs-on: linux
timeout-minutes: 60
env:
WAI_QUANTUM_NPU_EXPECT: none
steps:
- uses: actions/checkout@v5
- name: Install Rust 1.98.0
shell: bash
run: |
export PATH="$HOME/.cargo/bin:$PATH"
command -v rustup >/dev/null 2>&1 || \
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y --profile minimal --default-toolchain 1.98.0
rustup toolchain install 1.98.0 --profile minimal
rustup default 1.98.0
echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
- name: No neural processor here, and the member says so
shell: bash
working-directory: wai/wai-quantum
run: |
set -euo pipefail
report="$RUNNER_TEMP/npu-report.txt"; : > "$report"
WAI_QUANTUM_NPU_REPORT="$report" cargo test --locked --features quantum_fabric_npu --test npu_member -- --test-threads=1 2>&1 | tee "$RUNNER_TEMP/device.log"
grep -qE '^test result: ok\. 9 passed; 0 failed' "$RUNNER_TEMP/device.log" || { echo "::error::not every device test ran and passed"; exit 1; }
grep -qx 'npu devices: as declared (none)' "$report" || { echo "::error::the absent case was not established"; exit 1; }