Files
openclaw/.github/workflows/openclaw-performance.yml
2026-05-03 14:32:35 +01:00

569 lines
22 KiB
YAML

name: OpenClaw Performance
on:
schedule:
- cron: "11 5 * * *"
workflow_dispatch:
inputs:
target_ref:
description: OpenClaw ref to benchmark; defaults to the workflow ref
required: false
default: ""
type: string
profile:
description: Kova profile to run
required: false
default: diagnostic
type: choice
options:
- smoke
- diagnostic
- soak
- release
repeat:
description: Repeat count for non-profiled Kova runs
required: false
default: "3"
type: string
deep_profile:
description: Run the deep-profile lane with CPU/heap/trace artifacts
required: false
default: false
type: boolean
live_gpt54:
description: Run the live OpenAI GPT 5.4 agent-turn lane
required: false
default: false
type: boolean
fail_on_regression:
description: Fail the workflow when Kova exits non-zero
required: false
default: false
type: boolean
kova_ref:
description: openclaw/Kova Git ref to install
required: false
default: b63b6f9e20efb23641df00487e982230d81a90ac
type: string
permissions:
contents: read
concurrency:
group: ${{ github.event_name == 'workflow_dispatch' && format('{0}-{1}', github.workflow, github.run_id) || format('{0}-{1}', github.workflow, github.ref) }}
cancel-in-progress: false
env:
FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
OCM_VERSION: v0.2.15
KOVA_REPOSITORY: openclaw/Kova
PERFORMANCE_MODEL_ID: gpt-5.4
jobs:
kova:
name: ${{ matrix.title }}
runs-on: blacksmith-16vcpu-ubuntu-2404
timeout-minutes: 240
strategy:
fail-fast: false
matrix:
include:
- lane: mock-provider
title: Kova mock provider performance
auth: mock
repeat: input
deep_profile: "false"
live: "false"
include_filters: "scenario:fresh-install scenario:gateway-performance scenario:bundled-plugin-startup scenario:bundled-runtime-deps scenario:agent-cold-warm-message"
- lane: mock-deep-profile
title: Kova mock provider deep profile
auth: mock
repeat: "1"
deep_profile: "true"
live: "false"
include_filters: "scenario:fresh-install scenario:gateway-performance scenario:agent-cold-warm-message"
- lane: live-gpt54
title: Kova live OpenAI GPT 5.4 agent turn
auth: live
repeat: "1"
deep_profile: "false"
live: "true"
include_filters: "scenario:agent-cold-warm-message"
env:
KOVA_REF: ${{ inputs.kova_ref || 'b63b6f9e20efb23641df00487e982230d81a90ac' }}
KOVA_HOME: ${{ github.workspace }}/.artifacts/kova/home/${{ matrix.lane }}
PERFORMANCE_HELPER_DIR: ${{ github.workspace }}/.artifacts/performance-workflow
REPORT_DIR: ${{ github.workspace }}/.artifacts/kova/reports/${{ matrix.lane }}
BUNDLE_DIR: ${{ github.workspace }}/.artifacts/kova/bundles/${{ matrix.lane }}
SUMMARY_DIR: ${{ github.workspace }}/.artifacts/kova/summaries
SOURCE_PERF_DIR: ${{ github.workspace }}/.artifacts/openclaw-performance/source/${{ matrix.lane }}
LANE_ID: ${{ matrix.lane }}
TARGET_REF: ${{ inputs.target_ref || github.ref_name }}
PROFILE: ${{ inputs.profile || 'diagnostic' }}
REQUESTED_REPEAT: ${{ inputs.repeat || '3' }}
FAIL_ON_REGRESSION: ${{ inputs.fail_on_regression || 'false' }}
INCLUDE_FILTERS: ${{ matrix.include_filters }}
AUTH_MODE: ${{ matrix.auth }}
MATRIX_REPEAT: ${{ matrix.repeat }}
MATRIX_DEEP_PROFILE: ${{ matrix.deep_profile }}
MATRIX_LIVE: ${{ matrix.live }}
steps:
- name: Decide lane
id: lane
shell: bash
run: |
set -euo pipefail
run_lane=true
reason=""
if [[ "$LANE_ID" == "mock-deep-profile" && "${{ github.event_name }}" != "schedule" && "${{ inputs.deep_profile || 'false' }}" != "true" ]]; then
run_lane=false
reason="deep_profile input is false"
fi
if [[ "$LANE_ID" == "live-gpt54" && "${{ github.event_name }}" != "schedule" && "${{ inputs.live_gpt54 || 'false' }}" != "true" ]]; then
run_lane=false
reason="live_gpt54 input is false"
fi
echo "run=$run_lane" >> "$GITHUB_OUTPUT"
if [[ "$run_lane" != "true" ]]; then
echo "Skipping ${LANE_ID}: ${reason}" >> "$GITHUB_STEP_SUMMARY"
fi
- name: Detect clawgrit report token
id: clawgrit
if: steps.lane.outputs.run == 'true'
env:
CLAWGRIT_REPORTS_TOKEN: ${{ secrets.CLAWGRIT_REPORTS_TOKEN }}
shell: bash
run: |
set -euo pipefail
if [[ -n "${CLAWGRIT_REPORTS_TOKEN:-}" ]]; then
echo "present=true" >> "$GITHUB_OUTPUT"
else
echo "present=false" >> "$GITHUB_OUTPUT"
fi
- name: Checkout OpenClaw
if: steps.lane.outputs.run == 'true'
uses: actions/checkout@v6
with:
ref: ${{ inputs.target_ref || github.ref }}
fetch-depth: 1
persist-credentials: false
- name: Checkout performance workflow helpers
if: steps.lane.outputs.run == 'true'
uses: actions/checkout@v6
with:
ref: ${{ github.sha }}
path: .artifacts/performance-workflow
fetch-depth: 1
persist-credentials: false
- name: Record tested revision
if: steps.lane.outputs.run == 'true'
shell: bash
run: |
set -euo pipefail
tested_sha="$(git rev-parse HEAD)"
echo "TESTED_REF=${TARGET_REF}" >> "$GITHUB_ENV"
echo "TESTED_SHA=${tested_sha}" >> "$GITHUB_ENV"
{
echo "Tested ref: ${TARGET_REF}"
echo "Tested SHA: ${tested_sha}"
echo "Workflow ref: ${GITHUB_REF_NAME}"
echo "Workflow SHA: ${GITHUB_SHA}"
} >> "$GITHUB_STEP_SUMMARY"
- name: Set up Node environment
if: steps.lane.outputs.run == 'true'
uses: ./.github/actions/setup-node-env
with:
install-bun: "false"
- name: Install OCM and Kova
if: steps.lane.outputs.run == 'true'
shell: bash
run: |
set -euo pipefail
KOVA_SRC="${RUNNER_TEMP}/kova-src"
echo "KOVA_SRC=$KOVA_SRC" >> "$GITHUB_ENV"
mkdir -p "$HOME/.local/bin" "$(dirname "$KOVA_SRC")"
curl -fsSL https://raw.githubusercontent.com/shakkernerd/ocm/main/install.sh \
| bash -s -- --version "$OCM_VERSION" --prefix "$HOME/.local" --force
git clone --filter=blob:none "https://github.com/${KOVA_REPOSITORY}.git" "$KOVA_SRC"
git -C "$KOVA_SRC" checkout "$KOVA_REF"
cat > "$HOME/.local/bin/kova" <<EOF
#!/usr/bin/env bash
export KOVA_HOME="${KOVA_HOME}"
exec node "${KOVA_SRC}/bin/kova.mjs" "\$@"
EOF
chmod 0755 "$HOME/.local/bin/kova"
echo "$HOME/.local/bin" >> "$GITHUB_PATH"
- name: Pin Kova OpenAI model to GPT 5.4
if: steps.lane.outputs.run == 'true'
shell: bash
run: |
set -euo pipefail
node - <<'NODE'
const fs = require("node:fs");
const path = require("node:path");
const root = process.env.KOVA_SRC;
const files = [
"support/configure-openclaw-mock-auth.mjs",
"support/configure-openclaw-live-auth.mjs",
"support/mock-openai-server.mjs",
"states/mock-openai-provider.json"
];
for (const rel of files) {
const file = path.join(root, rel);
const before = fs.readFileSync(file, "utf8");
const after = before.replaceAll("gpt-5.5", process.env.PERFORMANCE_MODEL_ID);
fs.writeFileSync(file, after, "utf8");
}
NODE
- name: Kova version and plan sanity
if: steps.lane.outputs.run == 'true'
shell: bash
run: |
set -euo pipefail
kova version --json
kova matrix plan \
--profile "$PROFILE" \
--target "local-build:${GITHUB_WORKSPACE}" \
--include scenario:fresh-install \
--json >/tmp/kova-plan.json
- name: Configure live OpenAI auth
if: ${{ steps.lane.outputs.run == 'true' && matrix.live == 'true' }}
env:
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
OPENAI_BASE_URL: ${{ secrets.OPENAI_BASE_URL }}
shell: bash
run: |
set -euo pipefail
if [[ -z "${OPENAI_API_KEY:-}" ]]; then
echo "OPENAI_API_KEY is not configured; live GPT 5.4 lane will be skipped." >> "$GITHUB_STEP_SUMMARY"
exit 0
fi
kova setup --ci --json
kova setup --non-interactive --auth env-only --provider openai --env-var OPENAI_API_KEY --json
- name: Run Kova
id: kova
if: steps.lane.outputs.run == 'true'
env:
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
OPENAI_BASE_URL: ${{ secrets.OPENAI_BASE_URL }}
CLAWGRIT_REPORTS_TOKEN_PRESENT: ${{ steps.clawgrit.outputs.present || 'false' }}
shell: bash
run: |
set -euo pipefail
mkdir -p "$REPORT_DIR" "$BUNDLE_DIR" "$SUMMARY_DIR"
if [[ "$MATRIX_LIVE" == "true" && -z "${OPENAI_API_KEY:-}" ]]; then
echo "skipped=true" >> "$GITHUB_OUTPUT"
exit 0
fi
repeat="$REQUESTED_REPEAT"
if [[ "$MATRIX_REPEAT" != "input" ]]; then
repeat="$MATRIX_REPEAT"
fi
args=(
matrix run
--profile "$PROFILE"
--target "local-build:${GITHUB_WORKSPACE}"
--auth "$AUTH_MODE"
--parallel 1
--repeat "$repeat"
--report-dir "$REPORT_DIR"
--execute
--json
)
for filter in $INCLUDE_FILTERS; do
args+=(--include "$filter")
done
if [[ "$MATRIX_DEEP_PROFILE" == "true" ]]; then
args+=(--deep-profile)
fi
if [[ "$FAIL_ON_REGRESSION" == "true" ]]; then
args+=(--gate)
fi
log_path="$REPORT_DIR/${LANE_ID}.log"
set +e
kova "${args[@]}" 2>&1 | tee "$log_path"
status=${PIPESTATUS[0]}
set -e
report_json="$(find "$REPORT_DIR" -maxdepth 1 -type f -name '*.json' -print | sort | tail -n 1)"
if [[ -z "$report_json" ]]; then
echo "Kova did not write a JSON report." >&2
exit 1
fi
report_md="${report_json%.json}.md"
echo "status=$status" >> "$GITHUB_OUTPUT"
echo "report_json=$report_json" >> "$GITHUB_OUTPUT"
echo "report_md=$report_md" >> "$GITHUB_OUTPUT"
kova report bundle "$report_json" --output-dir "$BUNDLE_DIR" --json | tee "$BUNDLE_DIR/bundle.json"
ref_slug="$(printf '%s' "${TESTED_REF}" | tr -c 'A-Za-z0-9._-' '-')"
run_slug="${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
report_url=""
if [[ "${CLAWGRIT_REPORTS_TOKEN_PRESENT:-false}" == "true" ]]; then
report_url="https://github.com/openclaw/clawgrit-reports/tree/main/openclaw-performance/${ref_slug}/${run_slug}/${LANE_ID}"
fi
summary_path="$SUMMARY_DIR/${LANE_ID}.md"
summary_args=(node "$PERFORMANCE_HELPER_DIR/scripts/kova-ci-summary.mjs" --report "$report_json" --output "$summary_path" --lane "$LANE_ID")
if [[ -n "$report_url" ]]; then
summary_args+=(--report-url "$report_url")
fi
"${summary_args[@]}"
cat >> "$summary_path" <<EOF
## Test scope
- Repository: ${GITHUB_REPOSITORY}
- Tested ref: ${TESTED_REF}
- Tested SHA: ${TESTED_SHA}
- Workflow ref: ${GITHUB_REF_NAME}
- Workflow SHA: ${GITHUB_SHA}
- Kova repository: ${KOVA_REPOSITORY}
- Kova ref: ${KOVA_REF}
- Kova profile: ${PROFILE}
- Lane auth: ${AUTH_MODE}
- Lane model: ${PERFORMANCE_MODEL_ID}
- Lane repeat: ${repeat}
- Include filters: ${INCLUDE_FILTERS}
EOF
cat "$summary_path" >> "$GITHUB_STEP_SUMMARY"
if [[ "$FAIL_ON_REGRESSION" == "true" && "$status" != "0" ]]; then
exit "$status"
fi
- name: Run OpenClaw source performance probes
if: ${{ steps.lane.outputs.run == 'true' && matrix.lane == 'mock-provider' }}
shell: bash
run: |
set -euo pipefail
source_runs="$REQUESTED_REPEAT"
if ! [[ "$source_runs" =~ ^[0-9]+$ ]] || [[ "$source_runs" -lt 1 ]]; then
source_runs=3
fi
mkdir -p "$SOURCE_PERF_DIR/mock-hello"
if ! node -e "const fs=require('node:fs'); const scripts=require('./package.json').scripts||{}; process.exit(scripts['test:gateway:cpu-scenarios'] && scripts.openclaw && fs.existsSync('scripts/bench-cli-startup.ts') ? 0 : 1)"; then
cat > "$SOURCE_PERF_DIR/index.md" <<EOF
# OpenClaw Source Performance
Generated: $(date -u +%Y-%m-%dT%H:%M:%SZ)
Source probes skipped for this tested ref because one or more probe entry points are not present in the checked-out source tree.
## Test scope
- Tested ref: ${TESTED_REF}
- Tested SHA: ${TESTED_SHA}
- Required scripts: test:gateway:cpu-scenarios, openclaw, scripts/bench-cli-startup.ts
EOF
cat "$SOURCE_PERF_DIR/index.md" >> "$GITHUB_STEP_SUMMARY"
exit 0
fi
pnpm build
pnpm test:gateway:cpu-scenarios \
--output-dir "$SOURCE_PERF_DIR/gateway-cpu" \
--runs "$source_runs" \
--warmup 1 \
--skip-qa \
--startup-case default \
--startup-case skipChannels \
--startup-case oneInternalHook \
--startup-case allInternalHooks \
--startup-case fiftyPlugins \
--startup-case fiftyStartupLazyPlugins
for run_index in $(seq 1 "$source_runs"); do
run_dir="$SOURCE_PERF_DIR/mock-hello/run-$(printf '%03d' "$run_index")"
pnpm openclaw qa suite \
--provider-mode mock-openai \
--model "mock-openai/${PERFORMANCE_MODEL_ID}" \
--concurrency 1 \
--output-dir "$(realpath --relative-to="$GITHUB_WORKSPACE" "$run_dir")" \
--scenario channel-chat-baseline
done
gateway_home="$(mktemp -d)"
gateway_port="$(node -e "const net=require('node:net'); const s=net.createServer(); s.listen(0,'127.0.0.1',()=>{ console.log(s.address().port); s.close(); });")"
gateway_state="$gateway_home/.openclaw"
gateway_config="$gateway_state/openclaw.json"
gateway_log="$SOURCE_PERF_DIR/cli-gateway.log"
gateway_pid=""
mkdir -p "$gateway_state"
cat > "$gateway_config" <<EOF
{
"browser": { "enabled": false },
"gateway": {
"mode": "local",
"port": ${gateway_port},
"bind": "loopback",
"auth": { "mode": "none" },
"controlUi": { "enabled": false },
"tailscale": { "mode": "off" }
},
"plugins": {
"enabled": true,
"entries": { "browser": { "enabled": false } }
}
}
EOF
cleanup_gateway() {
if [[ -n "${gateway_pid:-}" ]] && kill -0 "$gateway_pid" 2>/dev/null; then
kill "$gateway_pid" 2>/dev/null || true
wait "$gateway_pid" 2>/dev/null || true
fi
rm -rf "$gateway_home"
}
trap cleanup_gateway EXIT
OPENCLAW_HOME="$gateway_home" OPENCLAW_STATE_DIR="$gateway_state" OPENCLAW_CONFIG_PATH="$gateway_config" OPENCLAW_GATEWAY_PORT="$gateway_port" OPENCLAW_SKIP_CHANNELS=1 \
node dist/entry.js gateway run --bind loopback --port "$gateway_port" --auth none --allow-unconfigured --force \
>"$gateway_log" 2>&1 &
gateway_pid="$!"
for _ in $(seq 1 120); do
if curl -fsS "http://127.0.0.1:${gateway_port}/healthz" >/dev/null; then
break
fi
if ! kill -0 "$gateway_pid" 2>/dev/null; then
cat "$gateway_log" >&2
exit 1
fi
sleep 1
done
curl -fsS "http://127.0.0.1:${gateway_port}/healthz" >/dev/null
OPENCLAW_HOME="$gateway_home" OPENCLAW_STATE_DIR="$gateway_state" OPENCLAW_CONFIG_PATH="$gateway_config" OPENCLAW_GATEWAY_PORT="$gateway_port" \
node --import tsx scripts/bench-cli-startup.ts \
--case gatewayHealthJson \
--case configGetGatewayPort \
--runs "$source_runs" \
--warmup 1 \
--output "$SOURCE_PERF_DIR/cli-startup.json"
cleanup_gateway
trap - EXIT
node "$PERFORMANCE_HELPER_DIR/scripts/openclaw-performance-source-summary.mjs" \
--source-dir "$SOURCE_PERF_DIR" \
--output "$SOURCE_PERF_DIR/index.md"
cat "$SOURCE_PERF_DIR/index.md" >> "$GITHUB_STEP_SUMMARY"
- name: Upload Kova artifacts
if: ${{ always() && steps.lane.outputs.run == 'true' }}
uses: actions/upload-artifact@v5
with:
name: openclaw-performance-${{ matrix.lane }}-${{ github.run_id }}-${{ github.run_attempt }}
path: |
.artifacts/kova/reports/${{ matrix.lane }}
.artifacts/kova/bundles/${{ matrix.lane }}
.artifacts/kova/summaries/${{ matrix.lane }}.md
.artifacts/openclaw-performance/source/${{ matrix.lane }}
if-no-files-found: ignore
retention-days: ${{ matrix.deep_profile == 'true' && 14 || 30 }}
- name: Prepare clawgrit reports checkout
if: ${{ steps.kova.outputs.report_json != '' && steps.clawgrit.outputs.present == 'true' }}
env:
CLAWGRIT_REPORTS_TOKEN: ${{ secrets.CLAWGRIT_REPORTS_TOKEN }}
shell: bash
run: |
set -euo pipefail
reports_root=".artifacts/clawgrit-reports"
mkdir -p "$reports_root"
git -C "$reports_root" init -b main
git -C "$reports_root" remote add origin https://github.com/openclaw/clawgrit-reports.git
auth_header="$(printf 'x-access-token:%s' "$CLAWGRIT_REPORTS_TOKEN" | base64 -w0)"
git -C "$reports_root" config http.https://github.com/.extraheader "AUTHORIZATION: basic ${auth_header}"
if git -C "$reports_root" ls-remote --exit-code --heads origin main >/dev/null 2>&1; then
git -C "$reports_root" fetch --depth=1 origin main
git -C "$reports_root" checkout -B main FETCH_HEAD
else
git -C "$reports_root" checkout -B main
fi
- name: Publish to clawgrit reports
if: ${{ steps.kova.outputs.report_json != '' && steps.clawgrit.outputs.present == 'true' }}
shell: bash
run: |
set -euo pipefail
reports_root=".artifacts/clawgrit-reports"
ref_slug="$(printf '%s' "${TESTED_REF}" | tr -c 'A-Za-z0-9._-' '-')"
run_slug="${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
dest="${reports_root}/openclaw-performance/${ref_slug}/${run_slug}/${LANE_ID}"
mkdir -p "$dest"
cp "${{ steps.kova.outputs.report_json }}" "$dest/report.json"
if [[ -f "${{ steps.kova.outputs.report_md }}" ]]; then
cp "${{ steps.kova.outputs.report_md }}" "$dest/report.md"
fi
cp "$SUMMARY_DIR/${LANE_ID}.md" "$dest/index.md"
if [[ -d "$BUNDLE_DIR" ]]; then
mkdir -p "$dest/bundles"
cp -R "$BUNDLE_DIR"/. "$dest/bundles/"
fi
if [[ -d "$SOURCE_PERF_DIR" ]]; then
mkdir -p "$dest/source"
cp -R "$SOURCE_PERF_DIR"/. "$dest/source/"
if [[ -f "$SOURCE_PERF_DIR/index.md" ]]; then
cat >> "$dest/index.md" <<'EOF'
## Source probes
Additional gateway boot, memory, plugin pressure, mock hello-loop, and CLI startup numbers are in [source/index.md](source/index.md).
EOF
fi
fi
cat > "${reports_root}/openclaw-performance/${ref_slug}/latest-${LANE_ID}.json" <<EOF
{
"repository": "${GITHUB_REPOSITORY}",
"ref": "${TESTED_REF}",
"sha": "${TESTED_SHA}",
"tested_ref": "${TESTED_REF}",
"tested_sha": "${TESTED_SHA}",
"workflow_ref": "${GITHUB_REF_NAME}",
"workflow_sha": "${GITHUB_SHA}",
"workflow": "${GITHUB_WORKFLOW}",
"run_id": "${GITHUB_RUN_ID}",
"run_attempt": "${GITHUB_RUN_ATTEMPT}",
"lane": "${LANE_ID}",
"path": "openclaw-performance/${ref_slug}/${run_slug}/${LANE_ID}"
}
EOF
git -C "$reports_root" config user.name "openclaw-performance[bot]"
git -C "$reports_root" config user.email "openclaw-performance[bot]@users.noreply.github.com"
git -C "$reports_root" add openclaw-performance
if git -C "$reports_root" diff --cached --quiet; then
echo "No clawgrit report changes to publish."
exit 0
fi
git -C "$reports_root" commit -m "perf: add OpenClaw ${LANE_ID} report ${TESTED_SHA::12}"
for attempt in 1 2 3 4 5; do
if git -C "$reports_root" push origin HEAD:main; then
exit 0
fi
if [[ "$attempt" == "5" ]]; then
exit 1
fi
sleep $((attempt * 2))
git -C "$reports_root" fetch --depth=1 origin main
git -C "$reports_root" rebase FETCH_HEAD
done