Cluster Deployment and Operations

Module Duration: 4-5 hours Focus: Production deployment, monitoring, and maintenance Prerequisites: Understanding of Spark architecture and cluster computing

Overview

This module covers deploying Spark applications in production environments, including cluster management, resource allocation, monitoring, troubleshooting, and best practices for reliable operation at scale.

Deployment Modes

Local Mode: Single JVM for development and testing. Standalone: Spark’s built-in cluster manager. YARN: Hadoop’s resource manager for multi-tenant clusters. Kubernetes: Container orchestration for cloud-native deployments. Mesos: General-purpose cluster manager (deprecated).

Spark Architecture Review

Cluster Components

Driver Program
  ├── SparkContext
  ├── DAG Scheduler
  ├── Task Scheduler
  └── Block Manager

Cluster Manager (YARN/K8s/Standalone)
  ├── Resource Allocation
  ├── Executor Lifecycle
  └── Application Monitoring

Executors (Worker Nodes)
  ├── Tasks
  ├── Block Manager
  ├── Cache
  └── Shuffle Data

Execution Flow

Application Submission: spark-submit sends application to cluster manager
Resource Allocation: Cluster manager allocates executors
Driver Initialization: SparkContext created, DAG scheduler initialized
Job Submission: Actions trigger job submission to DAG scheduler
Task Distribution: Tasks sent to executors for execution
Result Collection: Results aggregated and returned to driver

Standalone Cluster

Setting Up Standalone Cluster

# Start master
$SPARK_HOME/sbin/start-master.sh

# Master URL will be displayed: spark://master-host:7077

# Start workers
$SPARK_HOME/sbin/start-worker.sh spark://master-host:7077

# Start all (master + workers)
$SPARK_HOME/sbin/start-all.sh

# Stop all
$SPARK_HOME/sbin/stop-all.sh

Master Configuration

# conf/spark-env.sh
export SPARK_MASTER_HOST=master-node
export SPARK_MASTER_PORT=7077
export SPARK_MASTER_WEBUI_PORT=8080
export SPARK_MASTER_OPTS="-Dspark.deploy.defaultCores=8"

Worker Configuration

# conf/spark-env.sh on workers
export SPARK_WORKER_CORES=16
export SPARK_WORKER_MEMORY=64g
export SPARK_WORKER_INSTANCES=1
export SPARK_WORKER_DIR=/data/spark/work

Submitting to Standalone

spark-submit \
  --master spark://master-host:7077 \
  --deploy-mode cluster \
  --executor-memory 8g \
  --executor-cores 4 \
  --total-executor-cores 32 \
  --driver-memory 4g \
  --conf spark.executor.memoryOverhead=2g \
  --class com.example.MyApp \
  application.jar \
  arg1 arg2

High Availability

# Use ZooKeeper for master HA
export SPARK_DAEMON_JAVA_OPTS="-Dspark.deploy.recoveryMode=ZOOKEEPER \
  -Dspark.deploy.zookeeper.url=zk1:2181,zk2:2181,zk3:2181 \
  -Dspark.deploy.zookeeper.dir=/spark"

# Start multiple masters
./sbin/start-master.sh  # On master1
./sbin/start-master.sh  # On master2
./sbin/start-master.sh  # On master3

YARN Deployment

YARN Architecture

ResourceManager (RM)
  ├── Scheduler
  ├── ApplicationsManager
  └── Resource Tracker

NodeManager (NM) - on each worker
  ├── Container Manager
  ├── Node Health Checker
  └── Log Aggregation

ApplicationMaster (AM) - Spark Driver
  ├── Resource Negotiation
  ├── Task Scheduling
  └── Progress Monitoring

YARN Configuration

<!-- yarn-site.xml -->
<configuration>
  <property>
    <name>yarn.resourcemanager.hostname</name>
    <value>rm-host</value>
  </property>

  <property>
    <name>yarn.nodemanager.resource.memory-mb</name>
    <value>65536</value>
  </property>

  <property>
    <name>yarn.nodemanager.resource.cpu-vcores</name>
    <value>16</value>
  </property>

  <property>
    <name>yarn.scheduler.maximum-allocation-mb</name>
    <value>32768</value>
  </property>

  <property>
    <name>yarn.scheduler.minimum-allocation-mb</name>
    <value>1024</value>
  </property>
</configuration>

YARN Client Mode

# Driver runs on client machine
spark-submit \
  --master yarn \
  --deploy-mode client \
  --executor-memory 8g \
  --executor-cores 4 \
  --num-executors 10 \
  --driver-memory 4g \
  --conf spark.yarn.submit.waitAppCompletion=true \
  --conf spark.yarn.maxAppAttempts=3 \
  application.py

YARN Cluster Mode

# Driver runs on YARN cluster
spark-submit \
  --master yarn \
  --deploy-mode cluster \
  --executor-memory 8g \
  --executor-cores 4 \
  --num-executors 10 \
  --driver-memory 4g \
  --driver-cores 2 \
  --conf spark.yarn.submit.waitAppCompletion=false \
  --conf spark.yarn.maxAppAttempts=3 \
  --conf spark.yarn.am.memory=2g \
  --conf spark.yarn.am.cores=1 \
  application.py

YARN Resource Queues

# Submit to specific queue
spark-submit \
  --master yarn \
  --queue production \
  --executor-memory 8g \
  application.py

# Configure queue capacity (capacity-scheduler.xml)

<property>
  <name>yarn.scheduler.capacity.root.queues</name>
  <value>default,production,development</value>
</property>

<property>
  <name>yarn.scheduler.capacity.root.production.capacity</name>
  <value>70</value>
</property>

<property>
  <name>yarn.scheduler.capacity.root.development.capacity</name>
  <value>30</value>
</property>

Kubernetes Deployment

Prerequisites

# Install kubectl
curl -LO "https://dl.k8s.io/release/$(curl -L -s https://dl.k8s.io/release/stable.txt)/bin/linux/amd64/kubectl"
chmod +x kubectl
sudo mv kubectl /usr/local/bin/

# Configure kubeconfig
kubectl config use-context my-cluster

# Create namespace
kubectl create namespace spark-apps

# Create service account
kubectl create serviceaccount spark -n spark-apps

# Grant permissions
kubectl create clusterrolebinding spark-role \
  --clusterrole=edit \
  --serviceaccount=spark-apps:spark \
  --namespace=spark-apps

Building Docker Image

# Dockerfile
FROM apache/spark:3.5.0

# Add dependencies
COPY requirements.txt /opt/spark/work-dir/
RUN pip install -r /opt/spark/work-dir/requirements.txt

# Add application
COPY application.py /opt/spark/work-dir/
COPY config.yaml /opt/spark/work-dir/

WORKDIR /opt/spark/work-dir

ENTRYPOINT ["/opt/spark/bin/spark-submit"]

# Build and push image
docker build -t myregistry/spark-app:v1.0 .
docker push myregistry/spark-app:v1.0

Submitting to Kubernetes

spark-submit \
  --master k8s://https://k8s-master:6443 \
  --deploy-mode cluster \
  --name spark-app \
  --conf spark.kubernetes.container.image=myregistry/spark-app:v1.0 \
  --conf spark.kubernetes.namespace=spark-apps \
  --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \
  --conf spark.executor.instances=5 \
  --conf spark.executor.memory=8g \
  --conf spark.executor.cores=4 \
  --conf spark.driver.memory=4g \
  --conf spark.driver.cores=2 \
  --conf spark.kubernetes.driver.request.cores=2 \
  --conf spark.kubernetes.executor.request.cores=4 \
  --conf spark.kubernetes.driver.limit.cores=2 \
  --conf spark.kubernetes.executor.limit.cores=4 \
  local:///opt/spark/work-dir/application.py

Kubernetes ConfigMap

# spark-config.yaml
apiVersion: v1
kind: ConfigMap
metadata:
  name: spark-config
  namespace: spark-apps
data:
  spark-defaults.conf: |
    spark.executor.memory=8g
    spark.executor.cores=4
    spark.sql.shuffle.partitions=200
    spark.serializer=org.apache.spark.serializer.KryoSerializer

Kubernetes Persistent Volumes

# persistent-volume.yaml
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: spark-data-pvc
  namespace: spark-apps
spec:
  accessModes:
    - ReadWriteMany
  resources:
    requests:
      storage: 100Gi
  storageClassName: nfs-storage

# Mount in spark-submit
spark-submit \
  --master k8s://https://k8s-master:6443 \
  --conf spark.kubernetes.driver.volumes.persistentVolumeClaim.spark-data.mount.path=/data \
  --conf spark.kubernetes.driver.volumes.persistentVolumeClaim.spark-data.options.claimName=spark-data-pvc \
  --conf spark.kubernetes.executor.volumes.persistentVolumeClaim.spark-data.mount.path=/data \
  --conf spark.kubernetes.executor.volumes.persistentVolumeClaim.spark-data.options.claimName=spark-data-pvc \
  local:///opt/spark/work-dir/application.py

Spark Operator

# Install Spark Operator
helm repo add spark-operator https://googlecloudplatform.github.io/spark-on-k8s-operator
helm install spark-operator spark-operator/spark-operator --namespace spark-operator --create-namespace

# Create SparkApplication

# spark-application.yaml
apiVersion: sparkoperator.k8s.io/v1beta2
kind: SparkApplication
metadata:
  name: spark-pi
  namespace: spark-apps
spec:
  type: Python
  pythonVersion: "3"
  mode: cluster
  image: "apache/spark:3.5.0"
  imagePullPolicy: Always
  mainApplicationFile: local:///opt/spark/examples/src/main/python/pi.py
  sparkVersion: "3.5.0"
  restartPolicy:
    type: OnFailure
    onFailureRetries: 3
    onFailureRetryInterval: 10
    onSubmissionFailureRetries: 5
    onSubmissionFailureRetryInterval: 20
  driver:
    cores: 2
    memory: "4g"
    serviceAccount: spark
    labels:
      version: 3.5.0
  executor:
    cores: 4
    instances: 5
    memory: "8g"
    labels:
      version: 3.5.0

kubectl apply -f spark-application.yaml
kubectl get sparkapplication -n spark-apps
kubectl logs spark-pi-driver -n spark-apps

Resource Management

Calculating Resources

# Formula for executor configuration
"""
Given cluster resources:
- N nodes
- C cores per node
- M memory per node

Recommended:
- executor_cores = 4-5 (optimal for HDFS throughput)
- Leave 1 core and ~10% memory for OS/services
- Memory overhead = 10% of executor memory (min 384MB)

Example: 10 nodes, 32 cores, 128GB RAM each
- Usable: 31 cores, 115GB per node
- Executors per node: 31 / 5 = 6
- Total executors: 10 * 6 = 60
- Memory per executor: 115GB / 6 = 19GB
- Split: 15GB executor.memory + 4GB overhead
"""

def calculate_executor_config(nodes, cores_per_node, memory_per_node_gb):
    """Calculate optimal executor configuration"""
    # Reserve resources for OS
    usable_cores = cores_per_node - 1
    usable_memory = memory_per_node_gb * 0.9

    # Optimal cores per executor
    executor_cores = 5

    # Calculate executors per node
    executors_per_node = usable_cores // executor_cores
    total_executors = nodes * executors_per_node

    # Calculate memory per executor
    memory_per_executor = usable_memory / executors_per_node

    # Split memory and overhead (10%)
    executor_memory = int(memory_per_executor * 0.9)
    memory_overhead = int(memory_per_executor * 0.1)

    return {
        "num_executors": total_executors,
        "executor_cores": executor_cores,
        "executor_memory": f"{executor_memory}g",
        "memory_overhead": f"{memory_overhead}g",
        "total_cores": total_executors * executor_cores
    }

# Example
config = calculate_executor_config(
    nodes=10,
    cores_per_node=32,
    memory_per_node_gb=128
)
print(config)

Dynamic Resource Allocation

from pyspark.sql import SparkSession

spark = SparkSession.builder \
    .appName("DynamicAllocation") \
    .config("spark.dynamicAllocation.enabled", "true") \
    .config("spark.dynamicAllocation.minExecutors", "2") \
    .config("spark.dynamicAllocation.maxExecutors", "100") \
    .config("spark.dynamicAllocation.initialExecutors", "10") \
    .config("spark.dynamicAllocation.executorIdleTimeout", "60s") \
    .config("spark.dynamicAllocation.cachedExecutorIdleTimeout", "300s") \
    .config("spark.dynamicAllocation.schedulerBacklogTimeout", "1s") \
    .config("spark.shuffle.service.enabled", "true") \
    .getOrCreate()

External Shuffle Service

# Start external shuffle service (on each node)
$SPARK_HOME/sbin/start-shuffle-service.sh

# Configure in spark-defaults.conf
spark.shuffle.service.enabled=true
spark.shuffle.service.port=7337

Monitoring and Observability

Spark UI

# Access Spark UI
# During execution: http://driver-host:4040
# History server: http://history-server:18080

# Key tabs:
# - Jobs: Job execution timeline
# - Stages: Stage details and task metrics
# - Storage: Cached RDDs and memory usage
# - Environment: Spark configuration
# - Executors: Executor metrics and logs
# - SQL: Query execution plans

History Server

# Configure history server (spark-defaults.conf)
spark.eventLog.enabled=true
spark.eventLog.dir=hdfs:///spark-logs
spark.history.fs.logDirectory=hdfs:///spark-logs

# Start history server
$SPARK_HOME/sbin/start-history-server.sh

# Access at http://history-server:18080

Metrics System

# metrics.properties
*.sink.graphite.class=org.apache.spark.metrics.sink.GraphiteSink
*.sink.graphite.host=graphite-server
*.sink.graphite.port=2003
*.sink.graphite.period=10
*.sink.graphite.unit=seconds
*.sink.graphite.prefix=spark

# Prometheus metrics
*.sink.prometheusServlet.class=org.apache.spark.metrics.sink.PrometheusServlet
*.sink.prometheusServlet.path=/metrics
*.source.jvm.class=org.apache.spark.metrics.source.JvmSource

Prometheus Integration

# Enable Prometheus metrics
spark = SparkSession.builder \
    .config("spark.ui.prometheus.enabled", "true") \
    .config("spark.metrics.conf", "/path/to/metrics.properties") \
    .getOrCreate()

# prometheus.yml
scrape_configs:
  - job_name: 'spark'
    static_configs:
      - targets: ['spark-driver:4040']
    metrics_path: '/metrics/prometheus'

Grafana Dashboards

{
  "dashboard": {
    "title": "Spark Application Metrics",
    "panels": [
      {
        "title": "Executor Memory Usage",
        "targets": [
          {
            "expr": "spark_executor_memory_usedMemory"
          }
        ]
      },
      {
        "title": "Task Completion Rate",
        "targets": [
          {
            "expr": "rate(spark_executor_completedTasks[5m])"
          }
        ]
      },
      {
        "title": "GC Time",
        "targets": [
          {
            "expr": "spark_executor_jvmGCTime"
          }
        ]
      }
    ]
  }
}

Structured Streaming Monitoring

from pyspark.sql.streaming import StreamingQuery

query = df.writeStream \
    .format("console") \
    .start()

# Monitor query progress
progress = query.lastProgress
if progress:
    print(f"Input rate: {progress['inputRowsPerSecond']}")
    print(f"Process rate: {progress['processedRowsPerSecond']}")
    print(f"Batch duration: {progress['durationMs']['triggerExecution']}ms")
    print(f"State rows: {progress.get('stateOperators', [{}])[0].get('numRowsTotal', 0)}")

# Status information
status = query.status
print(f"Is active: {status['isDataAvailable']}")
print(f"Message: {status['message']}")

Logging

Log Configuration

# log4j.properties
log4j.rootCategory=WARN, console
log4j.appender.console=org.apache.log4j.ConsoleAppender
log4j.appender.console.target=System.err
log4j.appender.console.layout=org.apache.log4j.PatternLayout
log4j.appender.console.layout.ConversionPattern=%d{yy/MM/dd HH:mm:ss} %p %c{1}: %m%n

# Application logging
log4j.logger.com.mycompany=INFO
log4j.logger.org.apache.spark=WARN
log4j.logger.org.apache.hadoop=WARN

# Separate file appender
log4j.appender.file=org.apache.log4j.RollingFileAppender
log4j.appender.file.File=/var/log/spark/application.log
log4j.appender.file.MaxFileSize=100MB
log4j.appender.file.MaxBackupIndex=10
log4j.appender.file.layout=org.apache.log4j.PatternLayout
log4j.appender.file.layout.ConversionPattern=%d{yy/MM/dd HH:mm:ss} %p %c{1}: %m%n

Python Logging

import logging

# Configure logging
logging.basicConfig(
    level=logging.INFO,
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger(__name__)

# Use in application
logger.info("Starting Spark application")
logger.warning("Low memory warning")
logger.error("Task failed", exc_info=True)

# Set Spark log level
spark.sparkContext.setLogLevel("WARN")

Centralized Logging

# Configure log aggregation (log4j.properties)
log4j.appender.fluentd=org.fluentd.log4j.appender.FluentdAppender
log4j.appender.fluentd.host=fluentd-server
log4j.appender.fluentd.port=24224
log4j.appender.fluentd.tag=spark.application

Troubleshooting

Common Issues

Issue 1: OutOfMemoryError

# Symptoms: Tasks fail with OOM, executor lost

# Solutions:
# 1. Increase executor memory
spark.conf.set("spark.executor.memory", "16g")
spark.conf.set("spark.executor.memoryOverhead", "4g")

# 2. Increase partitions
df = df.repartition(1000)

# 3. Reduce batch size for streaming
spark.conf.set("spark.sql.shuffle.partitions", "500")

# 4. Disable broadcast for large tables
spark.conf.set("spark.sql.autoBroadcastJoinThreshold", "-1")

# 5. Use external shuffle service
spark.conf.set("spark.shuffle.service.enabled", "true")

Issue 2: Slow Performance

# Diagnose:
# - Check Spark UI for stage bottlenecks
# - Look for data skew in task duration
# - Monitor shuffle read/write sizes

# Solutions:
# 1. Optimize partition count
optimal_partitions = data_size_gb / 0.5  # 500MB per partition

# 2. Cache frequently used data
df.cache()

# 3. Use broadcast joins
from pyspark.sql.functions import broadcast
result = large_df.join(broadcast(small_df), "key")

# 4. Enable adaptive query execution
spark.conf.set("spark.sql.adaptive.enabled", "true")

# 5. Optimize file format
df.write.parquet("/output")  # Use columnar format

Issue 3: Executor Lost

# Symptoms: Executors crash or become unresponsive

# Solutions:
# 1. Increase executor memory overhead
spark.conf.set("spark.executor.memoryOverhead", "4g")

# 2. Enable speculation
spark.conf.set("spark.speculation", "true")

# 3. Configure heartbeat timeout
spark.conf.set("spark.network.timeout", "600s")
spark.conf.set("spark.executor.heartbeatInterval", "60s")

# 4. Check for GC issues
spark.conf.set("spark.executor.extraJavaOptions",
               "-XX:+UseG1GC -verbose:gc")

Issue 4: Data Skew

# Diagnose: Some tasks take much longer than others

# Solution 1: Salt keys
from pyspark.sql.functions import rand, concat, lit

df_salted = df.withColumn(
    "salted_key",
    concat(col("key"), lit("_"), (rand() * 10).cast("int"))
)

# Solution 2: Adaptive skew join
spark.conf.set("spark.sql.adaptive.skewJoin.enabled", "true")
spark.conf.set("spark.sql.adaptive.skewJoin.skewedPartitionThresholdInBytes", "256MB")

# Solution 3: Separate processing for hot keys
hot_keys = df.groupBy("key").count().filter(col("count") > 1000000)
hot_data = df.join(broadcast(hot_keys), "key")
normal_data = df.join(broadcast(hot_keys), "key", "left_anti")

result = process_hot(hot_data).union(process_normal(normal_data))

Debugging Tools

# 1. Explain query plan
df.explain(True)

# 2. Show partition distribution
df.withColumn("partition_id", spark_partition_id()) \
  .groupBy("partition_id").count() \
  .show()

# 3. Sample for testing
sample = df.sample(0.01, seed=42)

# 4. Collect metrics
start_time = time.time()
result = df.count()
duration = time.time() - start_time
print(f"Execution time: {duration}s")

# 5. Profile Python UDFs
spark.conf.set("spark.python.profile", "true")

Security

Authentication

# Enable Spark authentication
spark-submit \
  --conf spark.authenticate=true \
  --conf spark.authenticate.secret=my-secret-key \
  --conf spark.network.crypto.enabled=true \
  --conf spark.network.crypto.keyLength=128 \
  application.py

Kerberos Integration

# Kinit before submitting
kinit -kt /path/to/keytab user@REALM

# Submit with Kerberos
spark-submit \
  --master yarn \
  --deploy-mode cluster \
  --principal user@REALM \
  --keytab /path/to/keytab \
  --conf spark.yarn.principal=user@REALM \
  --conf spark.yarn.keytab=/path/to/keytab \
  application.py

SSL/TLS

# spark-defaults.conf
spark.ssl.enabled=true
spark.ssl.protocol=TLSv1.2
spark.ssl.keyStore=/path/to/keystore.jks
spark.ssl.keyStorePassword=password
spark.ssl.trustStore=/path/to/truststore.jks
spark.ssl.trustStorePassword=password

Access Control

# Enable SQL authorization
spark.conf.set("spark.sql.authorization.enabled", "true")
spark.conf.set("spark.sql.warehouse.dir", "hdfs:///warehouse")

# Configure Hive metastore security
spark.conf.set("hive.metastore.uris", "thrift://metastore:9083")
spark.conf.set("hive.metastore.sasl.enabled", "true")

Best Practices

Configuration Management

# Use configuration files
# spark-defaults.conf
"""
spark.executor.memory=8g
spark.executor.cores=4
spark.sql.shuffle.partitions=200
spark.serializer=org.apache.spark.serializer.KryoSerializer
"""

# Override in code when needed
spark = SparkSession.builder \
    .appName("MyApp") \
    .config("spark.sql.shuffle.partitions", "500") \
    .getOrCreate()

Deployment Checklist

"""
Pre-deployment checklist:

1. Resource Configuration
   - Executor memory sized correctly
   - Number of executors calculated
   - Dynamic allocation configured

2. Performance
   - Caching strategy defined
   - Shuffle partitions tuned
   - Broadcast joins identified

3. Reliability
   - Checkpointing enabled (streaming)
   - Speculation configured
   - Max attempts set

4. Monitoring
   - Metrics collection enabled
   - Logging configured
   - Alerting set up

5. Security
   - Authentication enabled
   - Encryption configured
   - Access controls verified

6. Testing
   - Tested with production data volume
   - Load tested
   - Failure scenarios tested
"""

Production Patterns

# Pattern 1: Graceful shutdown
import signal
import sys

def signal_handler(sig, frame):
    logger.info("Received shutdown signal")
    spark.stop()
    sys.exit(0)

signal.signal(signal.SIGTERM, signal_handler)
signal.signal(signal.SIGINT, signal_handler)

# Pattern 2: Retry logic
from time import sleep

def read_with_retry(path, max_retries=3):
    for attempt in range(max_retries):
        try:
            return spark.read.parquet(path)
        except Exception as e:
            if attempt == max_retries - 1:
                raise
            logger.warning(f"Retry {attempt + 1}/{max_retries}")
            sleep(2 ** attempt)

# Pattern 3: Health checks
def health_check():
    try:
        spark.sparkContext.getConf().get("spark.app.id")
        return True
    except:
        return False

Hands-On Exercises

Exercise 1: Deploy to YARN

# TODO: Deploy application to YARN cluster
# 1. Calculate optimal executor configuration
# 2. Submit application in cluster mode
# 3. Monitor via YARN UI
# 4. Collect application logs

# Your solution here

Exercise 2: Set Up Monitoring

# TODO: Configure comprehensive monitoring
# 1. Enable Prometheus metrics
# 2. Create Grafana dashboard
# 3. Set up alerting
# 4. Test with sample workload

# Your code here

Exercise 3: Kubernetes Deployment

# TODO: Deploy to Kubernetes
# 1. Build Docker image
# 2. Create Kubernetes manifests
# 3. Deploy with Spark Operator
# 4. Monitor and troubleshoot

# Your solution here

Summary

Successful Spark operations require:

Cluster Management: Choose appropriate deployment mode
Resource Allocation: Size executors correctly
Monitoring: Comprehensive observability
Troubleshooting: Quick issue resolution
Security: Authentication and encryption
Reliability: Fault tolerance and recovery

Key Takeaways

Choose deployment mode based on infrastructure
Calculate resources carefully for efficiency
Monitor continuously with proper tooling
Log comprehensively for troubleshooting
Secure all communication channels
Test thoroughly before production
Have runbooks for common issues

Continue to the next module for advanced Spark techniques and optimizations.

​Cluster Deployment and Operations

​Overview

​Deployment Modes

​Spark Architecture Review

​Cluster Components

​Execution Flow

​Standalone Cluster

​Setting Up Standalone Cluster

​Master Configuration

​Worker Configuration

​Submitting to Standalone

​High Availability

​YARN Deployment

​YARN Architecture

​YARN Configuration

​YARN Client Mode

​YARN Cluster Mode

​YARN Resource Queues

​Kubernetes Deployment

​Prerequisites

​Building Docker Image

​Submitting to Kubernetes

​Kubernetes ConfigMap

​Kubernetes Persistent Volumes

​Spark Operator

​Resource Management

​Calculating Resources

​Dynamic Resource Allocation

​External Shuffle Service

​Monitoring and Observability

​Spark UI

​History Server

​Metrics System

​Prometheus Integration

​Grafana Dashboards

​Structured Streaming Monitoring

​Logging

​Log Configuration

​Python Logging

​Centralized Logging

​Troubleshooting

​Common Issues

​Issue 1: OutOfMemoryError

​Issue 2: Slow Performance

​Issue 3: Executor Lost

​Issue 4: Data Skew

​Debugging Tools

​Security

​Authentication

​Kerberos Integration

​SSL/TLS

​Access Control

​Best Practices

​Configuration Management

​Deployment Checklist

​Production Patterns

​Hands-On Exercises

​Exercise 1: Deploy to YARN

​Exercise 2: Set Up Monitoring

​Exercise 3: Kubernetes Deployment

​Summary

​Key Takeaways

Cluster Deployment and Operations

Overview

Deployment Modes

Spark Architecture Review

Cluster Components

Execution Flow

Standalone Cluster

Setting Up Standalone Cluster

Master Configuration

Worker Configuration

Submitting to Standalone

High Availability

YARN Deployment

YARN Architecture

YARN Configuration

YARN Client Mode

YARN Cluster Mode

YARN Resource Queues

Kubernetes Deployment

Prerequisites

Building Docker Image

Submitting to Kubernetes

Kubernetes ConfigMap

Kubernetes Persistent Volumes

Spark Operator

Resource Management

Calculating Resources

Dynamic Resource Allocation

External Shuffle Service

Monitoring and Observability

Spark UI

History Server

Metrics System

Prometheus Integration

Grafana Dashboards

Structured Streaming Monitoring

Logging

Log Configuration

Python Logging

Centralized Logging

Troubleshooting

Common Issues

Issue 1: OutOfMemoryError

Issue 2: Slow Performance

Issue 3: Executor Lost

Issue 4: Data Skew

Debugging Tools

Security

Authentication

Kerberos Integration

SSL/TLS

Access Control

Best Practices

Configuration Management

Deployment Checklist

Production Patterns

Hands-On Exercises

Exercise 1: Deploy to YARN

Exercise 2: Set Up Monitoring

Exercise 3: Kubernetes Deployment

Summary

Key Takeaways