Refactor conditional checks to use [[ ]] for improved readability and consistency

Author: AryanAhadinia

src/swiss_ai_model_launch/assets/script.sh

@@ -10,7 +10,7 @@ OCF_SERVICE_NAME="llm"
 OCF_SERVICE_PORT=8080
 OCF_BOOTSTRAP_ADDR="/ip4/148.187.108.178/tcp/43905/p2p/QmbUKJkCfotDzbFE5uoTsXD4GRyPHjzZC1f2yAGLoeBMn9"
 
-if [ -n "$TELEMETRY_ENDPOINT" ]; then
+if [[ -n "$TELEMETRY_ENDPOINT" ]]; then
     curl -sf -X POST "$TELEMETRY_ENDPOINT" \
         -H "Content-Type: application/json" \
         -d '{"user": "'"${SLURM_JOB_USER}"'", "job_id": "'"${SLURM_JOB_ID}"'", "slurm_nodes": '"${SLURM_NNODES}"', "slurm_job_name": "'"${SLURM_JOB_NAME}"'", "slurm_partition": "'"${SLURM_JOB_PARTITION}"'", "slurm_time": "'"${SML_TIME}"'", "slurm_account": "'"${SLURM_JOB_ACCOUNT}"'", "slurm_environment": "'"${SML_ENVIRONMENT}"'", "interactive": false, "serving_framework": "'"${FRAMEWORK}"'", "framework_args": "'"${FRAMEWORK_ARGS}"'", "pre_launch_cmds": "'"${PRE_LAUNCH_CMDS}"'", "model_name": "'"${SERVED_MODEL_NAME}"'", "workers": '"${WORKERS}"', "nodes_per_worker": '"${NODES_PER_WORKER}"', "worker_port": '"${WORKER_PORT}"', "use_router": '"${USE_ROUTER}"', "router_environment": "'"${ROUTER_ENVIRONMENT}"'", "router_port": 30000, "router_args": "'"${ROUTER_ARGS}"'", "ocf_enabled": '"${USE_OCF}"', "ocf_bootstrap_addr": "'"${OCF_BOOTSTRAP_ADDR}"'", "ocf_service_name": "llm", "ocf_service_port": 8080}' || true
@@ -52,7 +52,7 @@ _sglang_setup() {
 _sglang_worker_cmd() {
     local local_rank=$1 worker_host_ip=$2
     local dist_args=""
-    if [ "$NODES_PER_WORKER" -gt 1 ]; then
+    if [[ "$NODES_PER_WORKER" -gt 1 ]]; then
         dist_args="--dist-init-addr ${worker_host_ip}:5757 --nnodes ${NODES_PER_WORKER} --node-rank ${local_rank}"
     fi
     FRAMEWORK_CMD="$FRAMEWORK_LAUNCH $dist_args $FRAMEWORK_ARGS"
@@ -69,18 +69,18 @@ _vllm_worker_cmd() {
     local local_rank=$1 worker_host_ip=$2
     local ray_port=6379 num_gpus=4
 
-    if [ "$NODES_PER_WORKER" -gt 1 ]; then
+    if [[ "$NODES_PER_WORKER" -gt 1 ]]; then
         # For multi-node: only the head node runs the API server via Ray;
         # follower nodes join the Ray cluster and block.
-        if [ "$local_rank" -eq 0 ]; then
+        if [[ "$local_rank" -eq 0 ]]; then
             FRAMEWORK_CMD="ray start --head --port=${ray_port} --num-gpus=${num_gpus} --block &
 
 echo 'Waiting for all Ray worker nodes to connect...'
 EXPECTED_GPUS=\$((${NODES_PER_WORKER} * ${num_gpus}))
 while true; do
     AVAILABLE_GPUS=\$(python3 -c 'import ray; ray.init(address=\"auto\"); print(int(ray.available_resources().get(\"GPU\", 0)))' 2>/dev/null || echo 0)
     echo \"Available GPUs: \${AVAILABLE_GPUS} / \${EXPECTED_GPUS}\"
-    if [ \"\${AVAILABLE_GPUS}\" -ge \"\${EXPECTED_GPUS}\" ]; then
+    if [[ \"\${AVAILABLE_GPUS}\" -ge \"\${EXPECTED_GPUS}\" ]]; then
         echo 'All Ray workers connected!'
         break
     fi
@@ -104,7 +104,7 @@ case "$FRAMEWORK" in
 esac
 
 EXPECTED_NODES=$((WORKERS * NODES_PER_WORKER))
-if [ "$TOTAL_NODES" -ne "$EXPECTED_NODES" ]; then
+if [[ "$TOTAL_NODES" -ne "$EXPECTED_NODES" ]]; then
     echo "Warning: Total nodes ($TOTAL_NODES) doesn't match WORKERS($WORKERS) * NODES_PER_WORKER($NODES_PER_WORKER) = $EXPECTED_NODES"
     echo "Adjusting to use all available nodes with WORKERS workers"
     NODES_PER_WORKER=$((TOTAL_NODES / WORKERS))
@@ -118,7 +118,7 @@ for ((worker_id=0; worker_id<WORKERS; worker_id++)); do
     worker_host_node=${nodes[$start_node]}
     worker_host_ip=$(srun --nodes=1 --ntasks=1 -w "${worker_host_node}" hostname -i)
 
-    if [ -z "$worker_host_ip" ]; then
+    if [[ -z "$worker_host_ip" ]]; then
         echo "Error: Could not retrieve IP address for worker $worker_id host ${worker_host_node}"
         exit 1
     fi
@@ -143,7 +143,7 @@ for ((worker_id=0; worker_id<WORKERS; worker_id++)); do
             vllm)   _vllm_worker_cmd   "$local_rank" "$worker_host_ip" ;;
         esac
 
-        if [ "$USE_OCF" = "true" ] && [ "$local_rank" -eq 0 ]; then
+        if [[ "$USE_OCF" = "true" ]] && [[ "$local_rank" -eq 0 ]]; then
             FRAMEWORK_CMD="\$OCF_BIN start --bootstrap.addr \"$OCF_BOOTSTRAP_ADDR\" --service.name $OCF_SERVICE_NAME --service.port $OCF_SERVICE_PORT --subprocess \"$FRAMEWORK_CMD\""
         fi
 
@@ -153,7 +153,7 @@ for ((worker_id=0; worker_id<WORKERS; worker_id++)); do
             bash --norc --noprofile -c "\
 set -ex
 $FRAMEWORK_ENV_SETUP
-if [ -n \"$PRE_LAUNCH_CMDS\" ]; then
+if [[ -n \"$PRE_LAUNCH_CMDS\" ]]; then
     echo \"Running pre-launch commands...\"
     eval \"$PRE_LAUNCH_CMDS\"
 fi
@@ -164,7 +164,7 @@ done
 # vmagent runs on the batch node rather than inside a container: pyxis containers
 # share the host network namespace, so the framework's API server is reachable
 # at localhost:8080 from here without any extra networking.
-if [ -n "$METRICS_REMOTE_WRITE_URL" ] && [ -x "$METRICS_AGENT_BIN" ]; then
+if [[ -n "$METRICS_REMOTE_WRITE_URL" ]] && [[ -x "$METRICS_AGENT_BIN" ]]; then
     "$METRICS_AGENT_BIN" \
         -promscrape.config=/capstor/store/cscs/swissai/infra01/ocf-share/vmagent-scrape.yaml \
         -remoteWrite.url="${METRICS_REMOTE_WRITE_URL}" \
@@ -174,11 +174,11 @@ if [ -n "$METRICS_REMOTE_WRITE_URL" ] && [ -x "$METRICS_AGENT_BIN" ]; then
         -remoteWrite.label="user=${SLURM_JOB_USER}" \
         "-remoteWrite.tmpDataPath=/tmp/vmagent-data-${SLURM_JOB_ID}" \
         > "/tmp/vmagent-${SLURM_JOB_ID}.log" 2>&1 &
-elif [ -n "$METRICS_REMOTE_WRITE_URL" ]; then
+elif [[ -n "$METRICS_REMOTE_WRITE_URL" ]]; then
     echo "metrics: $METRICS_AGENT_BIN not found, skipping push" >&2
 fi
 
-if [ "$USE_ROUTER" = "true" ] && [ "$WORKERS" -gt 1 ]; then
+if [[ "$USE_ROUTER" = "true" ]] && [[ "$WORKERS" -gt 1 ]]; then
     router_host_node=${nodes[0]}
     router_host_ip=${worker_head_ips[0]}
     worker_urls_str="${worker_urls[*]}"
@@ -199,7 +199,7 @@ unset http_proxy https_proxy HTTP_PROXY HTTPS_PROXY all_proxy ALL_PROXY
 echo \"Waiting for all workers to fully initialize the GPU engine before starting router...\"
 for worker_ip in ${worker_head_ips[*]}; do
     echo \"Checking worker at \$worker_ip...\"
-    while [ \"\$(curl --noproxy \"*\" -s -o /dev/null -w '%{http_code}' http://\${worker_ip}:${WORKER_PORT}/health)\" != \"200\" ]; do # NOSONAR
+    while [[ \"\$(curl --noproxy \"*\" -s -o /dev/null -w '%{http_code}' http://\${worker_ip}:${WORKER_PORT}/health)\" != \"200\" ]]; do # NOSONAR
         sleep 10
     done
     echo \"Worker \$worker_ip is fully ready!\"