[CI] replace shellcheck script with shellcheck-py hook (#52572)

Signed-off-by: Turner <doubleujabbour@gmail.com>
Signed-off-by: Turner Jabbour <doubleujabbour@gmail.com>
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
Turner Jabbour
2026-08-20 10:09:59 +00:00
committed by GitHub
co-authored by Claude Sonnet 5
parent 5b1e7a812b
commit 1eab6fef01
20 changed files with 204 additions and 197 deletions
@@ -52,10 +52,10 @@ CUDA_VISIBLE_DEVICES=$PREFILL_GPU_ID \
VLLM_SSM_CONV_STATE_LAYOUT=DS \
VLLM_KV_CACHE_LAYOUT=HND \
VLLM_NIXL_SIDE_CHANNEL_PORT=5559 \
vllm serve $MODEL \
vllm serve "$MODEL" \
--port $PREFILL_PORT \
--enforce-eager \
--gpu-memory-utilization $GPU_MEMORY_UTILIZATION \
--gpu-memory-utilization "$GPU_MEMORY_UTILIZATION" \
--max-model-len 16384 \
--block-size 128 \
--trust-remote-code \
@@ -70,10 +70,10 @@ CUDA_VISIBLE_DEVICES=$DECODE_GPU_ID \
VLLM_SSM_CONV_STATE_LAYOUT=DS \
VLLM_KV_CACHE_LAYOUT=HND \
VLLM_NIXL_SIDE_CHANNEL_PORT=6000 \
vllm serve $MODEL \
vllm serve "$MODEL" \
--port $DECODE_PORT \
--enforce-eager \
--gpu-memory-utilization $GPU_MEMORY_UTILIZATION \
--gpu-memory-utilization "$GPU_MEMORY_UTILIZATION" \
--max-model-len 16384 \
--block-size 128 \
--trust-remote-code \
@@ -86,10 +86,10 @@ launch_baseline() {
--gpu-memory-utilization ${GPU_MEMORY_UTILIZATION} \
--dtype float16 \
--enforce-eager"
echo ${BASELINE_BASE_CMD}
echo "${BASELINE_BASE_CMD}"
bash -c "${BASELINE_BASE_CMD}" &
sleep 10
wait_for_server ${BASELINE_HOST} ${BASELINE_PORT}
wait_for_server "${BASELINE_HOST}" "${BASELINE_PORT}"
}
launch_pd() {
@@ -97,53 +97,50 @@ launch_pd() {
PREFILL_GPU_ID=$(compute_gpu_id 0 "${PREFILLER_TP_SIZE}")
local DECODE_GPU_ID
DECODE_GPU_ID=$(compute_gpu_id "${PREFILLER_TP_SIZE}" "${DECODER_TP_SIZE}")
local kv_config="{\"kv_connector\":\"NixlConnector\",\"kv_role\":\"kv_both\",\"kv_buffer_device\":\"${KV_BUFFER_DEVICE}\"}"
PREFILL_BASE_CMD="
ZE_AFFINITY_MASK=$PREFILL_GPU_ID \
VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=200 \
VLLM_NIXL_SIDE_CHANNEL_HOST=${PREFILL_HOST} \
VLLM_NIXL_SIDE_CHANNEL_PORT=${PREFILL_NIXL_SIDE_PORT} \
VLLM_WORKER_MULTIPROC_METHOD=spawn \
VLLM_ENABLE_V1_MULTIPROCESSING=1 vllm serve $MODEL_NAME \
--host ${PREFILL_HOST} \
--port ${PREFILL_PORT} \
--max-model-len ${MAX_MODEL_LEN}\
--seed 42 \
--block-size ${BLOCK_SIZE} \
--enforce-eager \
--dtype float16 \
-tp ${PREFILLER_TP_SIZE} \
--gpu-memory-utilization ${GPU_MEMORY_UTILIZATION} \
--kv-transfer-config '{\"kv_connector\":\"NixlConnector\",\"kv_role\":\"kv_both\",\"kv_buffer_device\":\"$KV_BUFFER_DEVICE\"}'"
DECODE_BASE_CMD="
ZE_AFFINITY_MASK=$DECODE_GPU_ID \
VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=200 \
VLLM_WORKER_MULTIPROC_METHOD=spawn \
VLLM_ENABLE_V1_MULTIPROCESSING=1 vllm serve $MODEL_NAME \
--host ${DECODE_HOST} \
--port ${DECODE_PORT} \
--max-model-len ${MAX_MODEL_LEN}\
--seed 42 \
--block-size ${BLOCK_SIZE} \
--enforce-eager \
-tp ${DECODER_TP_SIZE} \
--dtype float16 \
--gpu-memory-utilization ${GPU_MEMORY_UTILIZATION} \
--kv-transfer-config '{\"kv_connector\":\"NixlConnector\",\"kv_role\":\"kv_both\",\"kv_buffer_device\":\"$KV_BUFFER_DEVICE\"}'"
echo ${PREFILL_BASE_CMD}
echo ${DECODE_BASE_CMD}
sleep 2
# execute on hosts
bash -c "${PREFILL_BASE_CMD}" &
bash -c "${DECODE_BASE_CMD}" &
echo "Starting prefill on GPU ${PREFILL_GPU_ID}, port ${PREFILL_PORT}"
ZE_AFFINITY_MASK="${PREFILL_GPU_ID}" \
VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=200 \
VLLM_NIXL_SIDE_CHANNEL_HOST="${PREFILL_HOST}" \
VLLM_NIXL_SIDE_CHANNEL_PORT="${PREFILL_NIXL_SIDE_PORT}" \
VLLM_WORKER_MULTIPROC_METHOD=spawn \
VLLM_ENABLE_V1_MULTIPROCESSING=1 \
vllm serve "$MODEL_NAME" \
--host "${PREFILL_HOST}" \
--port "${PREFILL_PORT}" \
--max-model-len "${MAX_MODEL_LEN}" \
--seed 42 \
--block-size "${BLOCK_SIZE}" \
--enforce-eager \
--dtype float16 \
-tp "${PREFILLER_TP_SIZE}" \
--gpu-memory-utilization "${GPU_MEMORY_UTILIZATION}" \
--kv-transfer-config "${kv_config}" &
echo "Starting decode on GPU ${DECODE_GPU_ID}, port ${DECODE_PORT}"
ZE_AFFINITY_MASK="${DECODE_GPU_ID}" \
VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=200 \
VLLM_WORKER_MULTIPROC_METHOD=spawn \
VLLM_ENABLE_V1_MULTIPROCESSING=1 \
vllm serve "$MODEL_NAME" \
--host "${DECODE_HOST}" \
--port "${DECODE_PORT}" \
--max-model-len "${MAX_MODEL_LEN}" \
--seed 42 \
--block-size "${BLOCK_SIZE}" \
--enforce-eager \
-tp "${DECODER_TP_SIZE}" \
--dtype float16 \
--gpu-memory-utilization "${GPU_MEMORY_UTILIZATION}" \
--kv-transfer-config "${kv_config}" &
sleep 1
wait_for_server ${PREFILL_HOST} ${PREFILL_PORT}
wait_for_server "${PREFILL_HOST}" "${PREFILL_PORT}"
sleep 1
wait_for_server ${DECODE_HOST} ${DECODE_PORT}
wait_for_server "${DECODE_HOST}" "${DECODE_PORT}"
sleep 1
}
@@ -153,7 +150,7 @@ launch_pd_proxy(){
--prefiller-host ${PREFILL_HOST} --prefiller-port ${PREFILL_PORT} \
--decoder-host ${DECODE_HOST} --decoder-port ${DECODE_PORT} \
--host=${PROXY_HOST} --port ${PROXY_PORT}"
echo ${PROXY_BASE_CMD}
echo "${PROXY_BASE_CMD}"
bash -c "${PROXY_BASE_CMD}" &
sleep 2
}
@@ -161,7 +158,7 @@ launch_pd_proxy(){
run_tests(){
local service_url=$1
local mode=$2
python3 ${EXP_ROOT}/test_disagg_accuracy.py --service_url=${service_url} --model_name=${MODEL_NAME} --mode=${mode} --file_name=${OUTPUT_FILE}
python3 "${EXP_ROOT}/test_disagg_accuracy.py" --service_url="${service_url}" --model_name="${MODEL_NAME}" --mode="${mode}" --file_name="${OUTPUT_FILE}"
}
@@ -178,7 +175,7 @@ launch_pd_proxy
run_tests "http://${PROXY_HOST}:${PROXY_PORT}" "disagg"
echo "-----P/D success----"
rm ${OUTPUT_FILE}
rm "${OUTPUT_FILE}"
cleanup
exit 0
@@ -103,8 +103,10 @@ fi
cleanup_instances() {
echo ""
echo "Cleaning up..."
# shellcheck disable=SC2046 # word splitting is intentional for multiple PIDs
kill $(jobs -pr) 2>/dev/null || true
sleep 1
# shellcheck disable=SC2046
kill -9 $(jobs -pr) 2>/dev/null || true
pkill -9 -f "vllm serve.*${MODEL_NAME}" 2>/dev/null || true
pkill -9 -f "toy_proxy_server.*8192" 2>/dev/null || true
@@ -122,7 +124,7 @@ wait_for_server() {
local deadline=${5:-600}
local elapsed=0
echo "Waiting for ${server_name} on port ${port}..."
while [ $elapsed -lt $deadline ]; do
while [ "$elapsed" -lt "$deadline" ]; do
if ! ps -p "$server_pid" > /dev/null 2>&1; then
local status=0
wait "$server_pid" || status=$?
@@ -186,7 +188,7 @@ else
else
num=1
fi
for (( g=0; g<num; g++ )); do ALL_GPUS+=($g); done
for (( g=0; g<num; g++ )); do ALL_GPUS+=("$g"); done
fi
TOTAL_GPUS_NEEDED=$(( (NUM_PREFILL_INSTANCES * PREFILLER_TP_SIZE) + (NUM_DECODE_INSTANCES * DECODER_TP_SIZE) ))
@@ -246,22 +248,22 @@ run_test_for_device() {
echo "Starting prefill instance $i on GPU $GPU_ID, port $PORT"
env \
${GPU_DEVICE_VAR}=$GPU_ID \
${GPU_DEVICE_VAR}="$GPU_ID" \
VLLM_KV_CACHE_LAYOUT='HND' \
UCX_NET_DEVICES=all \
${VLLM_SSM_CONV_STATE_LAYOUT:+VLLM_SSM_CONV_STATE_LAYOUT=$VLLM_SSM_CONV_STATE_LAYOUT} \
VLLM_NIXL_SIDE_CHANNEL_HOST=$NIXL_SIDE_CHANNEL_HOST \
VLLM_NIXL_SIDE_CHANNEL_HOST="$NIXL_SIDE_CHANNEL_HOST" \
VLLM_NIXL_SIDE_CHANNEL_PORT=$SIDE_CHANNEL_PORT \
vllm serve $MODEL_NAME \
vllm serve "$MODEL_NAME" \
--port $PORT \
--enforce-eager \
--max-model-len $MAX_MODEL_LEN \
--block-size ${BLOCK_SIZE} \
--max-model-len "$MAX_MODEL_LEN" \
--block-size "${BLOCK_SIZE}" \
"${KV_CACHE_ARGS[@]}" \
--tensor-parallel-size $PREFILLER_TP_SIZE \
--tensor-parallel-size "$PREFILLER_TP_SIZE" \
--kv-transfer-config "$kv_config_p" \
--speculative-config "$PREFILL_SPEC_CONFIG" \
--attention-backend $ATTENTION_BACKEND \
--attention-backend "$ATTENTION_BACKEND" \
${EXTRA_SERVE_ARGS[@]+"${EXTRA_SERVE_ARGS[@]}"} &
local SERVER_PID=$!
@@ -285,22 +287,22 @@ run_test_for_device() {
echo "Starting decode instance $i on GPU $GPU_ID, port $PORT"
env \
${GPU_DEVICE_VAR}=$GPU_ID \
${GPU_DEVICE_VAR}="$GPU_ID" \
VLLM_KV_CACHE_LAYOUT='HND' \
UCX_NET_DEVICES=all \
${VLLM_SSM_CONV_STATE_LAYOUT:+VLLM_SSM_CONV_STATE_LAYOUT=$VLLM_SSM_CONV_STATE_LAYOUT} \
VLLM_NIXL_SIDE_CHANNEL_HOST=$NIXL_SIDE_CHANNEL_HOST \
VLLM_NIXL_SIDE_CHANNEL_HOST="$NIXL_SIDE_CHANNEL_HOST" \
VLLM_NIXL_SIDE_CHANNEL_PORT=$SIDE_CHANNEL_PORT \
vllm serve $MODEL_NAME \
vllm serve "$MODEL_NAME" \
--port $PORT \
--enforce-eager \
--max-model-len $MAX_MODEL_LEN \
--block-size ${BLOCK_SIZE} \
--max-model-len "$MAX_MODEL_LEN" \
--block-size "${BLOCK_SIZE}" \
"${KV_CACHE_ARGS[@]}" \
--tensor-parallel-size $DECODER_TP_SIZE \
--tensor-parallel-size "$DECODER_TP_SIZE" \
--kv-transfer-config "$kv_config_d" \
--speculative-config "$DECODE_SPEC_CONFIG" \
--attention-backend $ATTENTION_BACKEND \
--attention-backend "$ATTENTION_BACKEND" \
${EXTRA_SERVE_ARGS[@]+"${EXTRA_SERVE_ARGS[@]}"} &
local SERVER_PID=$!
@@ -315,10 +317,10 @@ run_test_for_device() {
echo "Starting proxy server on port $PROXY_PORT..."
python3 "${GIT_ROOT}/tests/v1/kv_connector/nixl_integration/toy_proxy_server.py" \
--port $PROXY_PORT \
--prefiller-hosts ${PREFILL_HOSTS[*]} \
--prefiller-ports ${PREFILL_PORTS[*]} \
--decoder-hosts ${DECODE_HOSTS[*]} \
--decoder-ports ${DECODE_PORTS[*]} &
--prefiller-hosts "${PREFILL_HOSTS[@]}" \
--prefiller-ports "${PREFILL_PORTS[@]}" \
--decoder-hosts "${DECODE_HOSTS[@]}" \
--decoder-ports "${DECODE_PORTS[@]}" &
local PROXY_PID=$!
wait_for_server "$PROXY_PORT" "$PROXY_PID" "proxy" "/healthcheck" 60