Skip to main content
Standalone mode is the most direct way to run Flink: the JobManager and TaskManagers are launched as plain JVM processes on operating system nodes. There is no external resource manager — you are responsible for starting, stopping, and replacing processes.

​
Prerequisites

  • Java 8 or higher installed on all nodes.
  • A recent Flink distribution downloaded and unpacked from the Apache Flink downloads page.
  • For multi-node clusters: passwordless SSH access between nodes sharing the same Flink directory structure.

​
Starting a local session cluster

The fastest way to get started is a single-machine session cluster:
1

Start the cluster

./bin/start-cluster.sh
This starts one JobManager JVM and one TaskManager JVM. The Flink Web UI is available at http://localhost:8081.
2

Submit a job

./bin/flink run ./examples/streaming/TopSpeedWindowing.jar
3

Stop the cluster

./bin/stop-cluster.sh

​
Deployment modes

​
Session mode

In Session mode the cluster runs independently from any individual job. You submit multiple jobs to the same running cluster: 
# Start cluster
./bin/start-cluster.sh

# Submit first job
./bin/flink run ./examples/streaming/TopSpeedWindowing.jar

# Submit second job to the same cluster
./bin/flink run ./examples/batch/WordCount.jar

​
Application mode

In Application mode, one JobManager is dedicated to a single application. The application’s main() method runs on the JobManager itself.
1

Place your JAR in the lib directory

cp ./examples/streaming/TopSpeedWindowing.jar lib/
2

Start the application-mode JobManager

./bin/standalone-job.sh start \
    --job-classname org.apache.flink.streaming.examples.windowing.TopSpeedWindowing
The Web UI is available at http://localhost:8081.
3

Start one or more TaskManagers

./bin/taskmanager.sh start
# Start additional TaskManagers if the job needs more parallelism
./bin/taskmanager.sh start
4

Stop when done

./bin/taskmanager.sh stop
./bin/standalone-job.sh stop
You can also use artifact fetching instead of copying JARs to lib/: 
./bin/standalone-job.sh start \
    -D user.artifacts.base-dir=/tmp/flink-artifacts \
    --jars local:///path/to/TopSpeedWindowing.jar

​
Distributed cluster setup

For a multi-machine cluster, configure which hosts run which components.

​
Configuring masters and workers

Edit conf/masters to list JobManager hosts: 
master1
Edit conf/workers to list TaskManager hosts: 
worker1
worker2
worker3
Set jobmanager.rpc.address in conf/config.yaml to the master hostname: 
jobmanager.rpc.address: master1
Then start the distributed cluster: 
./bin/start-cluster.sh
The script uses SSH to start Flink processes on each listed host.

​
Local multi-process example

To simulate a cluster on one machine (useful for testing), list localhost multiple times in conf/workers: 
localhost
localhost
This starts two TaskManagers on the same machine.

​
Managing components individually

You can start and stop individual components without using start-cluster.sh: 
# Start/stop JobManager
./bin/jobmanager.sh start
./bin/jobmanager.sh stop

# Start/stop TaskManager (call multiple times for multiple instances)
./bin/taskmanager.sh start
./bin/taskmanager.sh stop

# Stop all running TaskManager instances at once
./bin/taskmanager.sh stop-all
To run a component in the foreground (useful in Docker): 
./bin/jobmanager.sh start-foreground

​
Dynamic properties

The component scripts accept dynamic configuration overrides via -D: 
./bin/jobmanager.sh start \
    -D jobmanager.rpc.address=192.168.1.10 \
    -D rest.port=8082
Dynamic properties override values in conf/config.yaml.

​
User JARs and classpath

ModeUser JAR recognition
Session modeJAR file specified in the flink run command
Application modeJAR specified in the startup command, plus all JARs in $FLINK_HOME/usrlib/

​
Setting up high availability

Standalone HA requires ZooKeeper. Configure it in conf/config.yaml and list multiple JobManagers in conf/masters: 
high-availability.type: zookeeper
high-availability.zookeeper.quorum: localhost:2181
high-availability.zookeeper.path.root: /flink
high-availability.cluster-id: /cluster_one
high-availability.storageDir: hdfs:///flink/recovery
conf/masters: 
localhost:8081
localhost:8082
1

Start ZooKeeper

./bin/start-zookeeper-quorum.sh
2

Start the HA cluster

./bin/start-cluster.sh
# Output confirms two JobManagers started:
# Starting HA cluster with 2 masters and 1 peers in ZooKeeper quorum.
3

Stop the cluster and ZooKeeper

./bin/stop-cluster.sh
./bin/stop-zookeeper-quorum.sh

​
Logs and debugging

Log files are written to the logs/ directory. Each service writes a .log file. Log files rotate on each service restart; older runs have a numeric suffix. To enable DEBUG logging, edit conf/log4j.properties: 
rootLogger.level = DEBUG
Logs are also available through the Web UI for both the JobManager and individual TaskManagers.

Build docs developers (and LLMs) love

Get started for freeTalk to us