From 2e361823f90329bf2aeac0b86c6fc78600f866c0 Mon Sep 17 00:00:00 2001
From: dafeliton <dafeliton@ucsd.edu>
Date: Fri, 30 Aug 2024 15:13:42 -0700
Subject: [PATCH] add debug

---
 start-cluster.sh | 99 +++++++++++++++++++++++++++++-------------------
 1 file changed, 61 insertions(+), 38 deletions(-)

diff --git a/start-cluster.sh b/start-cluster.sh
index 953f1b9..3bb1911 100644
--- a/start-cluster.sh
+++ b/start-cluster.sh
@@ -1,58 +1,81 @@
 #!/bin/bash
 
-IFS=: read -r FILESYSTEM HOMEMOUNT <<< $(findmnt -n -o SOURCE --target /home/$USER)
+set -euxo pipefail
 
-WORKSPACE=$(dirname $HOMEMOUNT)
-WORKSPACE=$(dirname $WORKSPACE)
+LOG_FILE="/tmp/prestop_$(date +'%Y%m%d_%H%M%S').log"
+exec 2> >(tee -a "$LOG_FILE")
 
-helm install $SPARK_CHART_NAME /opt/spark \
-    --set image.registry=${SPARK_CLUSTER_IMAGE_REGISTRY:-ghcr.io} \
-    --set image.repository=${SPARK_CLUSTER_IMAGE_REPO:-ucsd-ets/spark-node} \
-    --set image.tag=${SPARK_CLUSTER_IMAGE_TAG:-fa22-3} \
+echo "Running preStop hook script" | tee -a "$LOG_FILE"
+
+IFS=: read -r FILESYSTEM HOMEMOUNT <<< $(findmnt -n -o SOURCE --target /home/$USER) || {
+    echo "Error: Failed to find the filesystem and home mount." >&2
+    exit 1
+}
+
+WORKSPACE=$(dirname "$HOMEMOUNT") || {
+    echo "Error: Failed to determine the workspace from home mount." >&2
+    exit 1
+}
+WORKSPACE=$(dirname "$WORKSPACE") || {
+    echo "Error: Failed to determine the workspace directory." >&2
+    exit 1
+}
+
+echo "Filesystem: $FILESYSTEM, Workspace: $WORKSPACE" | tee -a "$LOG_FILE"
+
+helm install "$SPARK_CHART_NAME" /opt/spark \
+    --debug \
+    --set image.registry="${SPARK_CLUSTER_IMAGE_REGISTRY:-ghcr.io}" \
+    --set image.repository="${SPARK_CLUSTER_IMAGE_REPO:-ucsd-ets/spark-node}" \
+    --set image.tag="${SPARK_CLUSTER_IMAGE_TAG:-fa22-3}" \
     --set image.pullPolicy=Always \
     --set serviceAccount.name=default \
     --set serviceAccount.create=false \
-    --set master.podSecurityContext.runAsUser=$UID \
-    --set master.containerSecurityContext.runAsUser=$UID \
-    --set worker.replicaCount=${SPARK_CLUSTER_REPLICAS:-3} \
-    --set worker.podSecurityContext.runAsUser=$UID \
-    --set worker.containerSecurityContext.runAsUser=$UID \
-    --set master.podSecurityContext.runAsGroup=${SPARK_CLUSTER_RUNASGROUP:-0} \
-    --set master.podSecurityContext.fsGroup=${SPARK_CLUSTER_FSGROUP:-0} \
-    --set worker.podSecurityContext.runAsGroup=${SPARK_CLUSTER_RUNASGROUP:-0} \
-    --set worker.podSecurityContext.fsGroup=${SPARK_CLUSTER_FSGROUP:-0} \
-    --set worker.resources.requests.memory=${SPARK_CLUSTER_WORKER_MEM:-20G} \
-    --set worker.resources.limits.memory=${SPARK_CLUSTER_WORKER_MEM:-20G} \
-    --set worker.coreLimit=${SPARK_CLUSTER_WORKER_CPU:-2} \
-    --set worker.resources.limits.cpu=${SPARK_CLUSTER_WORKER_CPU:-2} \
-    --set worker.resources.requests.cpu=${SPARK_CLUSTER_WORKER_CPU:-2} \
-    --set master.resources.limits.cpu=${SPARK_CLUSTER_MASTER_CPU:-2} \
-    --set master.resources.requests.cpu=${SPARK_CLUSTER_MASTER_CPU:-2} \
-    --set master.resources.limits.memory=${SPARK_CLUSTER_MASTER_MEM:-8G} \
-    --set master.resources.requests.memory=${SPARK_CLUSTER_MASTER_MEM:-8G} \
-    --set master.memoryLimit=${SPARK_CLUSTER_MASTER_MEM:-8G} \
-    --set worker.memoryLimit=${SPARK_CLUSTER_WORKER_APP_MEM:-18G} \
+    --set master.podSecurityContext.runAsUser="$UID" \
+    --set master.containerSecurityContext.runAsUser="$UID" \
+    --set worker.replicaCount="${SPARK_CLUSTER_REPLICAS:-3}" \
+    --set worker.podSecurityContext.runAsUser="$UID" \
+    --set worker.containerSecurityContext.runAsUser="$UID" \
+    --set master.podSecurityContext.runAsGroup="${SPARK_CLUSTER_RUNASGROUP:-0}" \
+    --set master.podSecurityContext.fsGroup="${SPARK_CLUSTER_FSGROUP:-0}" \
+    --set worker.podSecurityContext.runAsGroup="${SPARK_CLUSTER_RUNASGROUP:-0}" \
+    --set worker.podSecurityContext.fsGroup="${SPARK_CLUSTER_FSGROUP:-0}" \
+    --set worker.resources.requests.memory="${SPARK_CLUSTER_WORKER_MEM:-20G}" \
+    --set worker.resources.limits.memory="${SPARK_CLUSTER_WORKER_MEM:-20G}" \
+    --set worker.coreLimit="${SPARK_CLUSTER_WORKER_CPU:-2}" \
+    --set worker.resources.limits.cpu="${SPARK_CLUSTER_WORKER_CPU:-2}" \
+    --set worker.resources.requests.cpu="${SPARK_CLUSTER_WORKER_CPU:-2}" \
+    --set master.resources.limits.cpu="${SPARK_CLUSTER_MASTER_CPU:-2}" \
+    --set master.resources.requests.cpu="${SPARK_CLUSTER_MASTER_CPU:-2}" \
+    --set master.resources.limits.memory="${SPARK_CLUSTER_MASTER_MEM:-8G}" \
+    --set master.resources.requests.memory="${SPARK_CLUSTER_MASTER_MEM:-8G}" \
+    --set master.memoryLimit="${SPARK_CLUSTER_MASTER_MEM:-8G}" \
+    --set worker.memoryLimit="${SPARK_CLUSTER_WORKER_APP_MEM:-18G}" \
     --set-json="worker.extraVolumes[0]={\"name\":\"course-workspace\",\"nfs\":{\"server\":\"${FILESYSTEM}\",\"path\":\"${WORKSPACE}\"}}" \
     --set-json='worker.extraVolumes[1]={"name":"home","persistentVolumeClaim":{"claimName":"home"}}' \
     --set-json="worker.extraVolumes[2]={\"name\":\"datasets\",\"nfs\":{\"server\":\"its-dsmlp-fs01.ucsd.edu\",\"path\":\"/export/datasets\"}}" \
     --set-json='worker.extraVolumeMounts[0]={"name":"course-workspace","mountPath":"/home/${USER}"}' \
-    --set worker.extraVolumeMounts[0].mountPath=/home/$USER \
-    --set worker.extraVolumeMounts[0].subPath=home/$USER \
+    --set worker.extraVolumeMounts[0].mountPath="/home/$USER" \
+    --set worker.extraVolumeMounts[0].subPath="home/$USER" \
     --set-json='worker.extraVolumeMounts[1]={"name":"course-workspace","mountPath":"/home/${USER}/public"}' \
-    --set worker.extraVolumeMounts[1].mountPath=/home/$USER/public \
-    --set worker.extraVolumeMounts[1].subPath=public \
+    --set worker.extraVolumeMounts[1].mountPath="/home/$USER/public" \
+    --set worker.extraVolumeMounts[1].subPath="public" \
     --set-json='worker.extraVolumeMounts[2]={"name":"home","mountPath":"/home/${USER}/private"}' \
-    --set worker.extraVolumeMounts[2].mountPath=/home/$USER/private \
+    --set worker.extraVolumeMounts[2].mountPath="/home/$USER/private" \
     --set-json='worker.extraVolumeMounts[3]={"name":"datasets","mountPath":"/datasets"}' \
     --set-json="master.extraVolumes[0]={\"name\":\"course-workspace\",\"nfs\":{\"server\":\"${FILESYSTEM}\",\"path\":\"${WORKSPACE}\"}}" \
     --set-json='master.extraVolumes[1]={"name":"home","persistentVolumeClaim":{"claimName":"home"}}' \
     --set-json="master.extraVolumes[2]={\"name\":\"datasets\",\"nfs\":{\"server\":\"its-dsmlp-fs01.ucsd.edu\",\"path\":\"/export/datasets\"}}" \
     --set-json='master.extraVolumeMounts[0]={"name":"course-workspace","mountPath":"/home/${USER}"}' \
-    --set master.extraVolumeMounts[0].mountPath=/home/$USER \
-    --set master.extraVolumeMounts[0].subPath=home/$USER \
+    --set master.extraVolumeMounts[0].mountPath="/home/$USER" \
+    --set master.extraVolumeMounts[0].subPath="home/$USER" \
     --set-json='master.extraVolumeMounts[1]={"name":"course-workspace","mountPath":"/home/${USER}/public"}' \
-    --set master.extraVolumeMounts[1].mountPath=/home/$USER/public \
-    --set master.extraVolumeMounts[1].subPath=public \
+    --set master.extraVolumeMounts[1].mountPath="/home/$USER/public" \
+    --set master.extraVolumeMounts[1].subPath="public" \
     --set-json='master.extraVolumeMounts[2]={"name":"home","mountPath":"/home/${USER}/private"}' \
-    --set-json='master.extraVolumeMounts[3]={"name":"datasets","mountPath":"/datasets"}' \
-    --set master.extraVolumeMounts[2].mountPath=/home/$USER/private
+    --set master.extraVolumeMounts[2].mountPath="/home/$USER/private" || {
+        echo "Error: Helm installation failed." >&2
+        exit 1
+    }
+
+echo "preStop hook script completed successfully."