Compare commits

...

2 Commits
v5.13 ... v6.0

Author SHA1 Message Date
Mikhail Chusavitin
d2eadedff2 Default NVIDIA ISO to open modules and add nvidia-legacy 2026-04-06 16:27:13 +03:00
Mikhail Chusavitin
a98c4d7461 Include terminal charts in benchmark report 2026-04-06 12:34:57 +03:00
6 changed files with 185 additions and 42 deletions

View File

@@ -243,7 +243,7 @@ func (s *System) RunNvidiaBenchmark(ctx context.Context, baseDir string, opts Nv
return "", fmt.Errorf("write result.json: %w", err)
}
report := renderBenchmarkReport(result)
report := renderBenchmarkReportWithCharts(result, loadBenchmarkReportCharts(runDir, selected))
if err := os.WriteFile(filepath.Join(runDir, "report.txt"), []byte(report), 0644); err != nil {
return "", fmt.Errorf("write report.txt: %w", err)
}

View File

@@ -2,11 +2,25 @@ package platform
import (
"fmt"
"os"
"path/filepath"
"regexp"
"strings"
"time"
)
func renderBenchmarkReport(result NvidiaBenchmarkResult) string {
return renderBenchmarkReportWithCharts(result, nil)
}
type benchmarkReportChart struct {
Title string
Content string
}
var ansiEscapePattern = regexp.MustCompile(`\x1b\[[0-9;]*m`)
func renderBenchmarkReportWithCharts(result NvidiaBenchmarkResult, charts []benchmarkReportChart) string {
var b strings.Builder
fmt.Fprintf(&b, "Bee NVIDIA Benchmark Report\n")
fmt.Fprintf(&b, "===========================\n\n")
@@ -93,6 +107,20 @@ func renderBenchmarkReport(result NvidiaBenchmarkResult) string {
b.WriteString("\n")
}
if len(charts) > 0 {
fmt.Fprintf(&b, "Terminal Charts\n")
fmt.Fprintf(&b, "---------------\n")
for _, chart := range charts {
content := strings.TrimSpace(stripANSIEscapeSequences(chart.Content))
if content == "" {
continue
}
fmt.Fprintf(&b, "%s\n", chart.Title)
fmt.Fprintf(&b, "%s\n", strings.Repeat("~", len(chart.Title)))
fmt.Fprintf(&b, "%s\n\n", content)
}
}
fmt.Fprintf(&b, "Methodology\n")
fmt.Fprintf(&b, "-----------\n")
fmt.Fprintf(&b, "- Profile %s uses standardized baseline, warmup, steady-state, interconnect, and cooldown phases.\n", result.BenchmarkProfile)
@@ -117,6 +145,36 @@ func renderBenchmarkReport(result NvidiaBenchmarkResult) string {
return b.String()
}
func loadBenchmarkReportCharts(runDir string, gpuIndices []int) []benchmarkReportChart {
phases := []struct {
name string
label string
}{
{name: "baseline", label: "Baseline"},
{name: "steady", label: "Steady State"},
{name: "cooldown", label: "Cooldown"},
}
var charts []benchmarkReportChart
for _, idx := range gpuIndices {
for _, phase := range phases {
path := filepath.Join(runDir, fmt.Sprintf("gpu-%d-%s-metrics-term.txt", idx, phase.name))
raw, err := os.ReadFile(path)
if err != nil || len(raw) == 0 {
continue
}
charts = append(charts, benchmarkReportChart{
Title: fmt.Sprintf("GPU %d %s", idx, phase.label),
Content: string(raw),
})
}
}
return charts
}
func stripANSIEscapeSequences(raw string) string {
return ansiEscapePattern.ReplaceAllString(raw, "")
}
func renderBenchmarkSummary(result NvidiaBenchmarkResult) string {
var b strings.Builder
fmt.Fprintf(&b, "run_at_utc=%s\n", result.GeneratedAt.Format(time.RFC3339))

View File

@@ -145,3 +145,35 @@ func TestRenderBenchmarkReportIncludesFindingsAndScores(t *testing.T) {
}
}
}
func TestRenderBenchmarkReportIncludesTerminalChartsWithoutANSI(t *testing.T) {
t.Parallel()
report := renderBenchmarkReportWithCharts(NvidiaBenchmarkResult{
BenchmarkProfile: NvidiaBenchmarkProfileStandard,
OverallStatus: "OK",
SelectedGPUIndices: []int{0},
Normalization: BenchmarkNormalization{
Status: "full",
},
}, []benchmarkReportChart{
{
Title: "GPU 0 Steady State",
Content: "\x1b[31mGPU 0 chart\x1b[0m\n 42┤───",
},
})
for _, needle := range []string{
"Terminal Charts",
"GPU 0 Steady State",
"GPU 0 chart",
"42┤───",
} {
if !strings.Contains(report, needle) {
t.Fatalf("report missing %q\n%s", needle, report)
}
}
if strings.Contains(report, "\x1b[31m") {
t.Fatalf("report should not contain ANSI escapes\n%s", report)
}
}

View File

@@ -41,15 +41,15 @@ while [ $# -gt 0 ]; do
;;
*)
echo "unknown arg: $1" >&2
echo "usage: $0 [--cache-dir /path] [--rebuild-image] [--clean-build] [--authorized-keys /path/to/authorized_keys] [--variant nvidia|amd|all]" >&2
echo "usage: $0 [--cache-dir /path] [--rebuild-image] [--clean-build] [--authorized-keys /path/to/authorized_keys] [--variant nvidia|nvidia-legacy|amd|nogpu|all]" >&2
exit 1
;;
esac
done
case "$VARIANT" in
nvidia|amd|nogpu|all) ;;
*) echo "unknown variant: $VARIANT (expected nvidia, amd, nogpu, or all)" >&2; exit 1 ;;
nvidia|nvidia-legacy|amd|nogpu|all) ;;
*) echo "unknown variant: $VARIANT (expected nvidia, nvidia-legacy, amd, nogpu, or all)" >&2; exit 1 ;;
esac
if [ "$CLEAN_CACHE" = "1" ]; then
@@ -61,8 +61,13 @@ if [ "$CLEAN_CACHE" = "1" ]; then
"${CACHE_DIR:?}/lb-packages"
echo "=== cleaning live-build work dirs ==="
rm -rf "${REPO_ROOT}/dist/live-build-work-nvidia"
rm -rf "${REPO_ROOT}/dist/live-build-work-nvidia-legacy"
rm -rf "${REPO_ROOT}/dist/live-build-work-amd"
rm -rf "${REPO_ROOT}/dist/live-build-work-nogpu"
rm -rf "${REPO_ROOT}/dist/overlay-stage-nvidia"
rm -rf "${REPO_ROOT}/dist/overlay-stage-nvidia-legacy"
rm -rf "${REPO_ROOT}/dist/overlay-stage-amd"
rm -rf "${REPO_ROOT}/dist/overlay-stage-nogpu"
echo "=== caches cleared, proceeding with build ==="
fi
@@ -180,6 +185,9 @@ case "$VARIANT" in
nvidia)
run_variant nvidia
;;
nvidia-legacy)
run_variant nvidia-legacy
;;
amd)
run_variant amd
;;
@@ -188,6 +196,7 @@ case "$VARIANT" in
;;
all)
run_variant nvidia
run_variant nvidia-legacy
run_variant amd
run_variant nogpu
;;

View File

@@ -1,8 +1,10 @@
#!/bin/sh
# build-nvidia-module.sh — compile NVIDIA proprietary driver modules for Debian 12
# build-nvidia-module.sh — compile NVIDIA kernel modules for Debian 12
#
# Downloads the official NVIDIA .run installer, extracts kernel modules and
# userspace tools (nvidia-smi, libnvidia-ml). Everything is proprietary NVIDIA.
# userspace tools (nvidia-smi, libnvidia-ml). Supports both:
# - open -> kernel-open/ sources from the .run installer
# - proprietary -> traditional proprietary kernel sources from the .run installer
#
# Output is cached in DIST_DIR/nvidia-<version>-<kver>/ so subsequent builds
# are instant unless NVIDIA_DRIVER_VERSION or kernel version changes.
@@ -17,10 +19,19 @@ set -e
NVIDIA_VERSION="$1"
DIST_DIR="$2"
DEBIAN_KERNEL_ABI="$3"
NVIDIA_FLAVOR="${4:-open}"
[ -n "$NVIDIA_VERSION" ] || { echo "usage: $0 <nvidia-version> <dist-dir> <debian-kernel-abi>"; exit 1; }
[ -n "$DIST_DIR" ] || { echo "usage: $0 <nvidia-version> <dist-dir> <debian-kernel-abi>"; exit 1; }
[ -n "$DEBIAN_KERNEL_ABI" ] || { echo "usage: $0 <nvidia-version> <dist-dir> <debian-kernel-abi>"; exit 1; }
[ -n "$NVIDIA_VERSION" ] || { echo "usage: $0 <nvidia-version> <dist-dir> <debian-kernel-abi> [open|proprietary]"; exit 1; }
[ -n "$DIST_DIR" ] || { echo "usage: $0 <nvidia-version> <dist-dir> <debian-kernel-abi> [open|proprietary]"; exit 1; }
[ -n "$DEBIAN_KERNEL_ABI" ] || { echo "usage: $0 <nvidia-version> <dist-dir> <debian-kernel-abi> [open|proprietary]"; exit 1; }
case "$NVIDIA_FLAVOR" in
open|proprietary) ;;
*)
echo "unsupported NVIDIA flavor: $NVIDIA_FLAVOR (expected open or proprietary)" >&2
exit 1
;;
esac
KVER="${DEBIAN_KERNEL_ABI}-amd64"
# On Debian, kernel headers are split into two packages:
@@ -31,22 +42,13 @@ KVER="${DEBIAN_KERNEL_ABI}-amd64"
KDIR_ARCH="/usr/src/linux-headers-${KVER}"
KDIR_COMMON="/usr/src/linux-headers-${DEBIAN_KERNEL_ABI}-common"
echo "=== NVIDIA ${NVIDIA_VERSION} (proprietary) for kernel ${KVER} ==="
echo "=== NVIDIA ${NVIDIA_VERSION} (${NVIDIA_FLAVOR}) for kernel ${KVER} ==="
if [ ! -d "$KDIR_ARCH" ] || [ ! -d "$KDIR_COMMON" ]; then
echo "=== installing linux-headers-${KVER} ==="
DEBIAN_FRONTEND=noninteractive apt-get install -y \
"linux-headers-${KVER}" \
gcc make perl
fi
echo "kernel headers (arch): $KDIR_ARCH"
echo "kernel headers (common): $KDIR_COMMON"
CACHE_DIR="${DIST_DIR}/nvidia-${NVIDIA_VERSION}-${KVER}"
CACHE_DIR="${DIST_DIR}/nvidia-${NVIDIA_FLAVOR}-${NVIDIA_VERSION}-${KVER}"
CACHE_ROOT="${BEE_CACHE_DIR:-${DIST_DIR}/cache}"
DOWNLOAD_CACHE_DIR="${CACHE_ROOT}/nvidia-downloads"
EXTRACT_CACHE_DIR="${CACHE_ROOT}/nvidia-extract"
CACHE_LAYOUT_VERSION="2"
CACHE_LAYOUT_VERSION="3"
CACHE_LAYOUT_MARKER="${CACHE_DIR}/.cache-layout-v${CACHE_LAYOUT_VERSION}"
if [ -d "$CACHE_DIR/modules" ] && [ -f "$CACHE_DIR/bin/nvidia-smi" ] \
&& [ -f "$CACHE_LAYOUT_MARKER" ] \
@@ -57,6 +59,15 @@ if [ -d "$CACHE_DIR/modules" ] && [ -f "$CACHE_DIR/bin/nvidia-smi" ] \
exit 0
fi
if [ ! -d "$KDIR_ARCH" ] || [ ! -d "$KDIR_COMMON" ]; then
echo "=== installing linux-headers-${KVER} ==="
DEBIAN_FRONTEND=noninteractive apt-get install -y \
"linux-headers-${KVER}" \
gcc make perl
fi
echo "kernel headers (arch): $KDIR_ARCH"
echo "kernel headers (common): $KDIR_COMMON"
# Download official NVIDIA .run installer with sha256 verification
BASE_URL="https://download.nvidia.com/XFree86/Linux-x86_64/${NVIDIA_VERSION}"
mkdir -p "$DOWNLOAD_CACHE_DIR" "$EXTRACT_CACHE_DIR"
@@ -90,12 +101,18 @@ EXTRACT_DIR="${EXTRACT_CACHE_DIR}/nvidia-extract-${NVIDIA_VERSION}"
rm -rf "$EXTRACT_DIR"
"$RUN_FILE" --extract-only --target "$EXTRACT_DIR"
# Find kernel source directory (proprietary: kernel/, open: kernel-open/)
# Find kernel source directory for the selected flavor.
KERNEL_SRC=""
for d in "$EXTRACT_DIR/kernel" "$EXTRACT_DIR/kernel-modules-sources" "$EXTRACT_DIR/kernel-source"; do
[ -f "$d/Makefile" ] && KERNEL_SRC="$d" && break
done
[ -n "$KERNEL_SRC" ] || { echo "ERROR: kernel source dir not found in:"; ls "$EXTRACT_DIR/"; exit 1; }
if [ "$NVIDIA_FLAVOR" = "open" ]; then
for d in "$EXTRACT_DIR/kernel-open" "$EXTRACT_DIR/kernel-open/"*; do
[ -f "$d/Makefile" ] && KERNEL_SRC="$d" && break
done
else
for d in "$EXTRACT_DIR/kernel" "$EXTRACT_DIR/kernel-modules-sources" "$EXTRACT_DIR/kernel-source"; do
[ -f "$d/Makefile" ] && KERNEL_SRC="$d" && break
done
fi
[ -n "$KERNEL_SRC" ] || { echo "ERROR: kernel source dir not found for flavor ${NVIDIA_FLAVOR} in:"; ls "$EXTRACT_DIR/"; exit 1; }
echo "kernel source: $KERNEL_SRC"
# Build kernel modules

View File

@@ -15,26 +15,46 @@ DIST_DIR="${REPO_ROOT}/dist"
VENDOR_DIR="${REPO_ROOT}/iso/vendor"
CACHE_ROOT="${BEE_CACHE_DIR:-${DIST_DIR}/cache}"
AUTH_KEYS=""
BUILD_VARIANT="nvidia"
BEE_GPU_VENDOR="nvidia"
BEE_NVIDIA_MODULE_FLAVOR="open"
# parse args
while [ $# -gt 0 ]; do
case "$1" in
--authorized-keys) AUTH_KEYS="$2"; shift 2 ;;
--variant) BEE_GPU_VENDOR="$2"; shift 2 ;;
--variant) BUILD_VARIANT="$2"; shift 2 ;;
*) echo "unknown arg: $1"; exit 1 ;;
esac
done
case "$BEE_GPU_VENDOR" in
nvidia|amd|nogpu) ;;
*) echo "unknown variant: $BEE_GPU_VENDOR (expected nvidia, amd, or nogpu)" >&2; exit 1 ;;
case "$BUILD_VARIANT" in
nvidia)
BEE_GPU_VENDOR="nvidia"
BEE_NVIDIA_MODULE_FLAVOR="open"
;;
nvidia-legacy)
BEE_GPU_VENDOR="nvidia"
BEE_NVIDIA_MODULE_FLAVOR="proprietary"
;;
amd)
BEE_GPU_VENDOR="amd"
BEE_NVIDIA_MODULE_FLAVOR=""
;;
nogpu)
BEE_GPU_VENDOR="nogpu"
BEE_NVIDIA_MODULE_FLAVOR=""
;;
*)
echo "unknown variant: $BUILD_VARIANT (expected nvidia, nvidia-legacy, amd, or nogpu)" >&2
exit 1
;;
esac
BUILD_WORK_DIR="${DIST_DIR}/live-build-work-${BEE_GPU_VENDOR}"
OVERLAY_STAGE_DIR="${DIST_DIR}/overlay-stage-${BEE_GPU_VENDOR}"
BUILD_WORK_DIR="${DIST_DIR}/live-build-work-${BUILD_VARIANT}"
OVERLAY_STAGE_DIR="${DIST_DIR}/overlay-stage-${BUILD_VARIANT}"
export BEE_GPU_VENDOR
export BEE_GPU_VENDOR BEE_NVIDIA_MODULE_FLAVOR BUILD_VARIANT
. "${BUILDER_DIR}/VERSIONS"
export PATH="$PATH:/usr/local/go/bin"
@@ -627,7 +647,7 @@ recover_iso_memtest() {
AUDIT_VERSION_EFFECTIVE="$(resolve_audit_version)"
ISO_VERSION_EFFECTIVE="$(resolve_iso_version)"
ISO_BASENAME="easy-bee-${BEE_GPU_VENDOR}-v${ISO_VERSION_EFFECTIVE}-amd64"
ISO_BASENAME="easy-bee-${BUILD_VARIANT}-v${ISO_VERSION_EFFECTIVE}-amd64"
# Versioned output directory: dist/easy-bee-v4.1/ — all final artefacts live here.
OUT_DIR="${DIST_DIR}/easy-bee-v${ISO_VERSION_EFFECTIVE}"
mkdir -p "${OUT_DIR}"
@@ -801,7 +821,7 @@ if [ ! -d "/usr/src/linux-headers-${KVER}" ]; then
apt-get install -y "linux-headers-${KVER}"
fi
echo "=== bee ISO build (variant: ${BEE_GPU_VENDOR}) ==="
echo "=== bee ISO build (variant: ${BUILD_VARIANT}) ==="
echo "Debian: ${DEBIAN_VERSION}, Kernel ABI: ${DEBIAN_KERNEL_ABI}, Go: ${GO_VERSION}"
echo "Audit version: ${AUDIT_VERSION_EFFECTIVE}, ISO version: ${ISO_VERSION_EFFECTIVE}"
echo ""
@@ -871,7 +891,7 @@ if [ "$BEE_GPU_VENDOR" = "nvidia" ]; then
fi
fi
echo "=== preparing staged overlay (${BEE_GPU_VENDOR}) ==="
echo "=== preparing staged overlay (${BUILD_VARIANT}) ==="
mkdir -p "${BUILD_WORK_DIR}" "${OVERLAY_STAGE_DIR}"
# Sync builder config into variant work dir, preserving lb cache.
@@ -981,10 +1001,10 @@ done
# --- NVIDIA kernel modules and userspace libs ---
if [ "$BEE_GPU_VENDOR" = "nvidia" ]; then
run_step "build NVIDIA ${NVIDIA_DRIVER_VERSION} modules" "40-nvidia-module" \
sh "${BUILDER_DIR}/build-nvidia-module.sh" "${NVIDIA_DRIVER_VERSION}" "${DIST_DIR}" "${DEBIAN_KERNEL_ABI}"
sh "${BUILDER_DIR}/build-nvidia-module.sh" "${NVIDIA_DRIVER_VERSION}" "${DIST_DIR}" "${DEBIAN_KERNEL_ABI}" "${BEE_NVIDIA_MODULE_FLAVOR}"
KVER="${DEBIAN_KERNEL_ABI}-amd64"
NVIDIA_CACHE="${DIST_DIR}/nvidia-${NVIDIA_DRIVER_VERSION}-${KVER}"
NVIDIA_CACHE="${DIST_DIR}/nvidia-${BEE_NVIDIA_MODULE_FLAVOR}-${NVIDIA_DRIVER_VERSION}-${KVER}"
# Inject .ko files into overlay at /usr/local/lib/nvidia/
OVERLAY_KMOD_DIR="${OVERLAY_STAGE_DIR}/usr/local/lib/nvidia"
@@ -1055,13 +1075,14 @@ GIT_COMMIT="$(git -C "${REPO_ROOT}" rev-parse --short HEAD 2>/dev/null || echo u
if [ "$BEE_GPU_VENDOR" = "nvidia" ]; then
GPU_VERSION_LINE="NVIDIA_DRIVER_VERSION=${NVIDIA_DRIVER_VERSION}
NVIDIA_KERNEL_MODULES_FLAVOR=${BEE_NVIDIA_MODULE_FLAVOR}
NCCL_VERSION=${NCCL_VERSION}
NCCL_CUDA_VERSION=${NCCL_CUDA_VERSION}
CUBLAS_VERSION=${CUBLAS_VERSION}
CUDA_USERSPACE_VERSION=${CUDA_USERSPACE_VERSION}
NCCL_TESTS_VERSION=${NCCL_TESTS_VERSION}
JOHN_JUMBO_COMMIT=${JOHN_JUMBO_COMMIT}"
GPU_BUILD_INFO="nvidia:${NVIDIA_DRIVER_VERSION}"
GPU_BUILD_INFO="nvidia-${BEE_NVIDIA_MODULE_FLAVOR}:${NVIDIA_DRIVER_VERSION}"
elif [ "$BEE_GPU_VENDOR" = "amd" ]; then
GPU_VERSION_LINE="ROCM_VERSION=${ROCM_VERSION}"
GPU_BUILD_INFO="rocm:${ROCM_VERSION}"
@@ -1073,6 +1094,7 @@ fi
cat > "${OVERLAY_STAGE_DIR}/etc/bee-release" <<EOF
BEE_ISO_VERSION=${ISO_VERSION_EFFECTIVE}
BEE_AUDIT_VERSION=${AUDIT_VERSION_EFFECTIVE}
BEE_BUILD_VARIANT=${BUILD_VARIANT}
BEE_GPU_VENDOR=${BEE_GPU_VENDOR}
BUILD_DATE=${BUILD_DATE}
GIT_COMMIT=${GIT_COMMIT}
@@ -1083,6 +1105,11 @@ EOF
# Write GPU vendor marker for hooks
echo "${BEE_GPU_VENDOR}" > "${OVERLAY_STAGE_DIR}/etc/bee-gpu-vendor"
if [ "$BEE_GPU_VENDOR" = "nvidia" ]; then
echo "${BEE_NVIDIA_MODULE_FLAVOR}" > "${OVERLAY_STAGE_DIR}/etc/bee-nvidia-modules-flavor"
else
rm -f "${OVERLAY_STAGE_DIR}/etc/bee-nvidia-modules-flavor"
fi
# Patch motd with build info
BEE_BUILD_INFO="${BUILD_DATE} git:${GIT_COMMIT} debian:${DEBIAN_VERSION} ${GPU_BUILD_INFO}"
@@ -1153,10 +1180,10 @@ fi
# --- build ISO using live-build ---
echo ""
echo "=== building ISO (live-build, variant: ${BEE_GPU_VENDOR}) ==="
echo "=== building ISO (variant: ${BUILD_VARIANT}) ==="
# Export for auto/config
BEE_GPU_VENDOR_UPPER="$(echo "${BEE_GPU_VENDOR}" | tr 'a-z' 'A-Z')"
BEE_GPU_VENDOR_UPPER="$(echo "${BUILD_VARIANT}" | tr 'a-z-' 'A-Z_')"
export BEE_GPU_VENDOR_UPPER
cd "${LB_DIR}"
@@ -1191,7 +1218,7 @@ if [ -f "$ISO_RAW" ]; then
validate_iso_nvidia_runtime "$ISO_RAW"
cp "$ISO_RAW" "$ISO_OUT"
echo ""
echo "=== done (${BEE_GPU_VENDOR}) ==="
echo "=== done (${BUILD_VARIANT}) ==="
echo "ISO: $ISO_OUT"
if command -v stat >/dev/null 2>&1; then
ISO_SIZE_BYTES="$(stat -c '%s' "$ISO_OUT" 2>/dev/null || stat -f '%z' "$ISO_OUT")"