Add NVIDIA stress loader selection and DCGM 4 support
This commit is contained in:
Mikhail Chusavitin
93
iso/overlay/usr/local/bin/bee-gpu-burn
Normal file
93
iso/overlay/usr/local/bin/bee-gpu-burn
Normal file
@@ -0,0 +1,93 @@
|
||||
#!/bin/sh
|
||||
set -eu
|
||||
|
||||
SECONDS=5
|
||||
SIZE_MB=64
|
||||
DEVICES=""
|
||||
EXCLUDE=""
|
||||
WORKER="/usr/local/lib/bee/bee-gpu-burn-worker"
|
||||
|
||||
usage() {
|
||||
echo "usage: $0 [--seconds N] [--size-mb N] [--devices 0,1] [--exclude 2,3]" >&2
|
||||
exit 2
|
||||
}
|
||||
|
||||
normalize_list() {
|
||||
echo "${1:-}" | tr ',' '\n' | sed 's/[[:space:]]//g' | awk 'NF' | sort -n | uniq | paste -sd, -
|
||||
}
|
||||
|
||||
contains_csv() {
|
||||
needle="$1"
|
||||
haystack="${2:-}"
|
||||
echo ",${haystack}," | grep -q ",${needle},"
|
||||
}
|
||||
|
||||
while [ "$#" -gt 0 ]; do
|
||||
case "$1" in
|
||||
--seconds|-t) [ "$#" -ge 2 ] || usage; SECONDS="$2"; shift 2 ;;
|
||||
--size-mb|-m) [ "$#" -ge 2 ] || usage; SIZE_MB="$2"; shift 2 ;;
|
||||
--devices) [ "$#" -ge 2 ] || usage; DEVICES="$2"; shift 2 ;;
|
||||
--exclude) [ "$#" -ge 2 ] || usage; EXCLUDE="$2"; shift 2 ;;
|
||||
*) usage ;;
|
||||
esac
|
||||
done
|
||||
|
||||
[ -x "${WORKER}" ] || { echo "bee-gpu-burn worker not found: ${WORKER}" >&2; exit 1; }
|
||||
|
||||
ALL_DEVICES=$(nvidia-smi --query-gpu=index --format=csv,noheader,nounits 2>/dev/null | sed 's/[[:space:]]//g' | awk 'NF' | paste -sd, -)
|
||||
[ -n "${ALL_DEVICES}" ] || { echo "nvidia-smi found no NVIDIA GPUs" >&2; exit 1; }
|
||||
|
||||
DEVICES=$(normalize_list "${DEVICES}")
|
||||
EXCLUDE=$(normalize_list "${EXCLUDE}")
|
||||
SELECTED="${DEVICES}"
|
||||
if [ -z "${SELECTED}" ]; then
|
||||
SELECTED="${ALL_DEVICES}"
|
||||
fi
|
||||
|
||||
FINAL=""
|
||||
for id in $(echo "${SELECTED}" | tr ',' ' '); do
|
||||
[ -n "${id}" ] || continue
|
||||
if contains_csv "${id}" "${EXCLUDE}"; then
|
||||
continue
|
||||
fi
|
||||
if [ -z "${FINAL}" ]; then
|
||||
FINAL="${id}"
|
||||
else
|
||||
FINAL="${FINAL},${id}"
|
||||
fi
|
||||
done
|
||||
|
||||
[ -n "${FINAL}" ] || { echo "no NVIDIA GPUs selected after filters" >&2; exit 1; }
|
||||
|
||||
echo "loader=bee-gpu-burn"
|
||||
echo "selected_gpus=${FINAL}"
|
||||
|
||||
TMP_DIR=$(mktemp -d)
|
||||
trap 'rm -rf "${TMP_DIR}"' EXIT INT TERM
|
||||
|
||||
WORKERS=""
|
||||
for id in $(echo "${FINAL}" | tr ',' ' '); do
|
||||
log="${TMP_DIR}/gpu-${id}.log"
|
||||
echo "starting gpu ${id}"
|
||||
"${WORKER}" --device "${id}" --seconds "${SECONDS}" --size-mb "${SIZE_MB}" >"${log}" 2>&1 &
|
||||
pid=$!
|
||||
WORKERS="${WORKERS} ${pid}:${id}:${log}"
|
||||
done
|
||||
|
||||
status=0
|
||||
for spec in ${WORKERS}; do
|
||||
pid=${spec%%:*}
|
||||
rest=${spec#*:}
|
||||
id=${rest%%:*}
|
||||
log=${rest#*:}
|
||||
if wait "${pid}"; then
|
||||
echo "gpu ${id} finished: OK"
|
||||
else
|
||||
rc=$?
|
||||
echo "gpu ${id} finished: FAILED rc=${rc}"
|
||||
status=1
|
||||
fi
|
||||
sed "s/^/[gpu ${id}] /" "${log}" || true
|
||||
done
|
||||
|
||||
exit "${status}"
|
||||
100
iso/overlay/usr/local/bin/bee-john-gpu-stress
Normal file
100
iso/overlay/usr/local/bin/bee-john-gpu-stress
Normal file
@@ -0,0 +1,100 @@
|
||||
#!/bin/sh
|
||||
set -eu
|
||||
|
||||
SECONDS=300
|
||||
DEVICES=""
|
||||
EXCLUDE=""
|
||||
FORMAT=""
|
||||
JOHN_DIR="/usr/local/lib/bee/john/run"
|
||||
JOHN_BIN="${JOHN_DIR}/john"
|
||||
|
||||
usage() {
|
||||
echo "usage: $0 [--seconds N] [--devices 0,1] [--exclude 2,3] [--format name]" >&2
|
||||
exit 2
|
||||
}
|
||||
|
||||
normalize_list() {
|
||||
echo "${1:-}" | tr ',' '\n' | sed 's/[[:space:]]//g' | awk 'NF' | sort -n | uniq | paste -sd, -
|
||||
}
|
||||
|
||||
contains_csv() {
|
||||
needle="$1"
|
||||
haystack="${2:-}"
|
||||
echo ",${haystack}," | grep -q ",${needle},"
|
||||
}
|
||||
|
||||
while [ "$#" -gt 0 ]; do
|
||||
case "$1" in
|
||||
--seconds|-t) [ "$#" -ge 2 ] || usage; SECONDS="$2"; shift 2 ;;
|
||||
--devices) [ "$#" -ge 2 ] || usage; DEVICES="$2"; shift 2 ;;
|
||||
--exclude) [ "$#" -ge 2 ] || usage; EXCLUDE="$2"; shift 2 ;;
|
||||
--format) [ "$#" -ge 2 ] || usage; FORMAT="$2"; shift 2 ;;
|
||||
*) usage ;;
|
||||
esac
|
||||
done
|
||||
|
||||
[ -x "${JOHN_BIN}" ] || { echo "john binary not found: ${JOHN_BIN}" >&2; exit 1; }
|
||||
|
||||
ALL_DEVICES=$(nvidia-smi --query-gpu=index --format=csv,noheader,nounits 2>/dev/null | sed 's/[[:space:]]//g' | awk 'NF' | paste -sd, -)
|
||||
[ -n "${ALL_DEVICES}" ] || { echo "nvidia-smi found no NVIDIA GPUs" >&2; exit 1; }
|
||||
|
||||
DEVICES=$(normalize_list "${DEVICES}")
|
||||
EXCLUDE=$(normalize_list "${EXCLUDE}")
|
||||
SELECTED="${DEVICES}"
|
||||
if [ -z "${SELECTED}" ]; then
|
||||
SELECTED="${ALL_DEVICES}"
|
||||
fi
|
||||
|
||||
FINAL=""
|
||||
for id in $(echo "${SELECTED}" | tr ',' ' '); do
|
||||
[ -n "${id}" ] || continue
|
||||
if contains_csv "${id}" "${EXCLUDE}"; then
|
||||
continue
|
||||
fi
|
||||
if [ -z "${FINAL}" ]; then
|
||||
FINAL="${id}"
|
||||
else
|
||||
FINAL="${FINAL},${id}"
|
||||
fi
|
||||
done
|
||||
|
||||
[ -n "${FINAL}" ] || { echo "no NVIDIA GPUs selected after filters" >&2; exit 1; }
|
||||
|
||||
JOHN_DEVICES=""
|
||||
for id in $(echo "${FINAL}" | tr ',' ' '); do
|
||||
opencl_id=$((id + 1))
|
||||
if [ -z "${JOHN_DEVICES}" ]; then
|
||||
JOHN_DEVICES="${opencl_id}"
|
||||
else
|
||||
JOHN_DEVICES="${JOHN_DEVICES},${opencl_id}"
|
||||
fi
|
||||
done
|
||||
|
||||
echo "loader=john"
|
||||
echo "selected_gpus=${FINAL}"
|
||||
echo "john_devices=${JOHN_DEVICES}"
|
||||
|
||||
cd "${JOHN_DIR}"
|
||||
|
||||
choose_format() {
|
||||
if [ -n "${FORMAT}" ]; then
|
||||
echo "${FORMAT}"
|
||||
return 0
|
||||
fi
|
||||
for candidate in sha512crypt-opencl pbkdf2-hmac-sha512-opencl 7z-opencl sha256crypt-opencl md5crypt-opencl; do
|
||||
if ./john --test=1 --format="${candidate}" --devices="${JOHN_DEVICES}" >/dev/null 2>&1; then
|
||||
echo "${candidate}"
|
||||
return 0
|
||||
fi
|
||||
done
|
||||
return 1
|
||||
}
|
||||
|
||||
CHOSEN_FORMAT=$(choose_format) || {
|
||||
echo "no suitable john OpenCL format found" >&2
|
||||
./john --list=opencl-devices >&2 || true
|
||||
exit 1
|
||||
}
|
||||
|
||||
echo "format=${CHOSEN_FORMAT}"
|
||||
exec ./john --test="${SECONDS}" --format="${CHOSEN_FORMAT}" --devices="${JOHN_DEVICES}"
|
||||
@@ -114,4 +114,19 @@ fi
|
||||
ldconfig 2>/dev/null || true
|
||||
log "ldconfig refreshed"
|
||||
|
||||
# Start DCGM host engine so dcgmi can discover GPUs.
|
||||
# nv-hostengine must run before any dcgmi command — without it, dcgmi reports
|
||||
# "group is empty" even when GPUs and modules are present.
|
||||
# Skip if already running (e.g. started by a dcgm systemd service or prior boot).
|
||||
if command -v nv-hostengine >/dev/null 2>&1; then
|
||||
if pgrep -x nv-hostengine >/dev/null 2>&1; then
|
||||
log "nv-hostengine already running — skipping"
|
||||
else
|
||||
nv-hostengine
|
||||
log "nv-hostengine started"
|
||||
fi
|
||||
else
|
||||
log "WARN: nv-hostengine not found — dcgmi diagnostics will not work"
|
||||
fi
|
||||
|
||||
log "done"
|
||||
|
||||
Reference in New Issue
Block a user