Skip to content

armgit / cua

Public Imported from GitHub
0

Scale computer-use 2.0 with open-source drivers, cross-OS sandboxes, and benchmarks for training, evaluation, and data generation.

ci(fork-fanout): check 2-CPU hosted runners against their own measured baseline (#5913)

ddupont2a3975570a88
diff --git a/.github/workflows/nightly-qemu-fork.yml b/.github/workflows/nightly-qemu-fork.yml
index 75dada6e9c..78a8684a9e 100644
--- a/.github/workflows/nightly-qemu-fork.yml
+++ b/.github/workflows/nightly-qemu-fork.yml
@@ -144,9 +144,20 @@ jobs:
CUA_QEMU_FREE_PAGE_REPORTING: ${{ matrix.forkable == '0' && '0' || '' }}
run: libs/cua/bench/fork-fanout/run.sh "$RUNNER_TEMP/fork-fanout.json"
- name: Perf guard
- run: >
- python3 libs/cua/bench/fork-fanout/check.py "$RUNNER_TEMP/fork-fanout.json"
- libs/cua/bench/fork-fanout/baseline.json --host "linux-kvm-gha-${{ matrix.path }}"
+ # Hosted runners of private repositories have 2 CPUs (public ones 4):
+ # forks are about twice as slow there, so they are checked against
+ # their own measured baseline when one exists.
+ env:
+ HOST_CLASS: linux-kvm-gha-${{ matrix.path }}
+ run: |
+ set -euo pipefail
+ base=libs/cua/bench/fork-fanout/baseline.json
+ host="$HOST_CLASS"
+ if [ "$(nproc)" -lt 4 ] && python3 -c 'import json,sys; sys.exit(0 if sys.argv[2] in json.load(open(sys.argv[1]))["hosts"] else 1)' "$base" "$host-2cpu"; then
+ host="$host-2cpu"
+ fi
+ echo "baseline host class: $host ($(nproc) CPUs)"
+ python3 libs/cua/bench/fork-fanout/check.py "$RUNNER_TEMP/fork-fanout.json" "$base" --host "$host"
- uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
if: always()
with:
diff --git a/libs/cua/bench/fork-fanout/baseline.json b/libs/cua/bench/fork-fanout/baseline.json
index 282b4280df..ca4dc2c75b 100644
--- a/libs/cua/bench/fork-fanout/baseline.json
+++ b/libs/cua/bench/fork-fanout/baseline.json
@@ -47,6 +47,26 @@
}
}
},
+ "linux-kvm-gha-forkable-2cpu": {
+ "source": "GitHub-hosted ubuntu-24.04 for private repositories (2 vCPU, 7.8 GB, nested KVM), QEMU from apt, run.sh MEMORY_MB=1024 DIRTY_MB=200 FORKABLE=1, CUA_VM_VCPU_OVERCOMMIT=10, published slim disk; measured run 2026-10-24 (wave of 8 forks of 2 vCPUs on 2 CPUs)",
+ "factor": 2.0,
+ "forkable": true,
+ "checkpoint_ms": 7372,
+ "waves": {
+ "4": {
+ "fork_ms_p50": 11689,
+ "fork_ms_p95": 13967,
+ "wall_s": 13.97,
+ "private_mb_per_fork": 100
+ },
+ "8": {
+ "fork_ms_p50": 22810,
+ "fork_ms_p95": 23674,
+ "wall_s": 23.67,
+ "private_mb_per_fork": 107
+ }
+ }
+ },
"linux-kvm-gha-loaded-stream": {
"source": "GitHub-hosted ubuntu-24.04 (4 vCPU, nested KVM), QEMU from apt, run.sh MEMORY_MB=1024 DIRTY_MB=200, PR #74 run 39363060790, 2026-10-20, published slim disk, FORKABLE=0",
"factor": 2.0,