In the Spring Boot Document, they said that 'Each SpringApplication will register a shutdown hook with the JVM to ensure that the ApplicationContext is closed gracefully on exit.'
When I click ctrl+c on the shell command, the application can be shutdown gracefully. If I run the application in a production machine, I have to use the command
java -jar ProApplicaton.jar. But I can't close the shell terminal, otherwise it will close the process.
If I run command like nohup java -jar ProApplicaton.jar &, I can't use ctrl+c to shutdown it gracefully.
What is the correct way to start and stop a Spring Boot Application in the production environment?
This question is related to
linux
spring
spring-boot
All of the answers seem to be missing the fact that you may need to complete some portion of work in coordinated fashion during graceful shutdown (for example, in an enterprise application).
@PreDestroy allows you to execute shutdown code in the individual beans. Something more sophisticated would look like this:
@Component
public class ApplicationShutdown implements ApplicationListener<ContextClosedEvent> {
@Autowired ... //various components and services
@Override
public void onApplicationEvent(ContextClosedEvent event) {
service1.changeHeartBeatMessage(); // allows loadbalancers & clusters to prepare for the impending shutdown
service2.deregisterQueueListeners();
service3.finishProcessingTasksAtHand();
service2.reportFailedTasks();
service4.gracefullyShutdownNativeSystemProcessesThatMayHaveBeenLaunched();
service1.eventLogGracefulShutdownComplete();
}
}
As of Spring Boot 2.3 and later, there's a built-in graceful shutdown mechanism.
Pre-Spring Boot 2.3, there is no out-of-the box graceful shutdown mechanism. Some spring-boot starters provide this functionality:
I am the author of nr. 1. The starter is named "Hiatus for Spring Boot". It works on the load balancer level, i.e. simply marks the service as OUT_OF_SERVICE, not interfering with application context in any way. This allows to do a graceful shutdown and means that, if required, the service can be taken out of service for some time and then brought back to life. The downside is that it doesn't stop the JVM, you will have to do it with kill command. As I run everything in containers, this was no big deal for me, because I will have to stop and remove the container anyway.
Nos. 2 and 3 are more or less based on this post by Andy Wilkinson. They work one-way - once triggered, they eventually close the context.
You can make the springboot application to write the PID into file and you can use the pid file to stop or restart or get the status using a bash script. To write the PID to a file, register a listener to SpringApplication using ApplicationPidFileWriter as shown below :
SpringApplication application = new SpringApplication(Application.class);
application.addListeners(new ApplicationPidFileWriter("./bin/app.pid"));
application.run();
Then write a bash script to run the spring boot application . Reference.
Now you can use the script to start,stop or restart.
Here is another option that does not require you to change the code or exposing a shut-down endpoint. Create the following scripts and use them to start and stop your app.
start.sh
#!/bin/bash
java -jar myapp.jar & echo $! > ./pid.file &
Starts your app and saves the process id in a file
stop.sh
#!/bin/bash
kill $(cat ./pid.file)
Stops your app using the saved process id
start_silent.sh
#!/bin/bash
nohup ./start.sh > foo.out 2> foo.err < /dev/null &
If you need to start the app using ssh from a remote machine or a CI pipeline then use this script instead to start your app. Using start.sh directly can leave the shell to hang.
After eg. re/deploying your app you can restart it using:
sshpass -p password ssh -oStrictHostKeyChecking=no [email protected] 'cd /home/user/pathToApp; ./stop.sh; ./start_silent.sh'
I don't expose any endpoints and start (with nohup in background and without out files created through nohup) and stop with shell script(with KILL PID gracefully and force kill if app is still running after 3 mins). I just create executable jar and use PID file writer to write PID file and store Jar and Pid in folder with same name as of application name and shell scripts also have same name with start and stop in the end. I call these stop script and start script via jenkins pipeline also. No issues so far. Perfectly working for 8 applications(Very generic scripts and easy to apply for any app).
Main Class
@SpringBootApplication
public class MyApplication {
public static final void main(String[] args) {
SpringApplicationBuilder app = new SpringApplicationBuilder(MyApplication.class);
app.build().addListeners(new ApplicationPidFileWriter());
app.run();
}
}
YML FILE
spring.pid.fail-on-write-error: true
spring.pid.file: /server-path-with-folder-as-app-name-for-ID/appName/appName.pid
Here is the start script(start-appname.sh):
#Active Profile(YAML)
ACTIVE_PROFILE="preprod"
# JVM Parameters and Spring boot initialization parameters
JVM_PARAM="-Xms512m -Xmx1024m -Dspring.profiles.active=${ACTIVE_PROFILE} -Dcom.webmethods.jms.clientIDSharing=true"
# Base Folder Path like "/folder/packages"
CURRENT_DIR=$(readlink -f "$0")
BASE_PACKAGE="${CURRENT_DIR%/bin/*}"
# Shell Script file name after removing path like "start-yaml-validator.sh"
SHELL_SCRIPT_FILE_NAME=$(basename -- "$0")
# Shell Script file name after removing extension like "start-yaml-validator"
SHELL_SCRIPT_FILE_NAME_WITHOUT_EXT="${SHELL_SCRIPT_FILE_NAME%.sh}"
# App name after removing start/stop strings like "yaml-validator"
APP_NAME=${SHELL_SCRIPT_FILE_NAME_WITHOUT_EXT#start-}
PIDS=`ps aux |grep [j]ava.*-Dspring.profiles.active=$ACTIVE_PROFILE.*$APP_NAME.*jar | awk {'print $2'}`
if [ -z "$PIDS" ]; then
echo "No instances of $APP_NAME with profile:$ACTIVE_PROFILE is running..." 1>&2
else
for PROCESS_ID in $PIDS; do
echo "Please stop the process($PROCESS_ID) using the shell script: stop-$APP_NAME.sh"
done
exit 1
fi
# Preparing the java home path for execution
JAVA_EXEC='/usr/bin/java'
# Java Executable - Jar Path Obtained from latest file in directory
JAVA_APP=$(ls -t $BASE_PACKAGE/apps/$APP_NAME/$APP_NAME*.jar | head -n1)
# To execute the application.
FINAL_EXEC="$JAVA_EXEC $JVM_PARAM -jar $JAVA_APP"
# Making executable command using tilde symbol and running completely detached from terminal
`nohup $FINAL_EXEC </dev/null >/dev/null 2>&1 &`
echo "$APP_NAME start script is completed."
Here is the stop script(stop-appname.sh):
#Active Profile(YAML)
ACTIVE_PROFILE="preprod"
#Base Folder Path like "/folder/packages"
CURRENT_DIR=$(readlink -f "$0")
BASE_PACKAGE="${CURRENT_DIR%/bin/*}"
# Shell Script file name after removing path like "start-yaml-validator.sh"
SHELL_SCRIPT_FILE_NAME=$(basename -- "$0")
# Shell Script file name after removing extension like "start-yaml-validator"
SHELL_SCRIPT_FILE_NAME_WITHOUT_EXT="${SHELL_SCRIPT_FILE_NAME%.*}"
# App name after removing start/stop strings like "yaml-validator"
APP_NAME=${SHELL_SCRIPT_FILE_NAME_WITHOUT_EXT:5}
# Script to stop the application
PID_PATH="$BASE_PACKAGE/config/$APP_NAME/$APP_NAME.pid"
if [ ! -f "$PID_PATH" ]; then
echo "Process Id FilePath($PID_PATH) Not found"
else
PROCESS_ID=`cat $PID_PATH`
if [ ! -e /proc/$PROCESS_ID -a /proc/$PROCESS_ID/exe ]; then
echo "$APP_NAME was not running with PROCESS_ID:$PROCESS_ID.";
else
kill $PROCESS_ID;
echo "Gracefully stopping $APP_NAME with PROCESS_ID:$PROCESS_ID..."
sleep 5s
fi
fi
PIDS=`/bin/ps aux |/bin/grep [j]ava.*-Dspring.profiles.active=$ACTIVE_PROFILE.*$APP_NAME.*jar | /bin/awk {'print $2'}`
if [ -z "$PIDS" ]; then
echo "All instances of $APP_NAME with profile:$ACTIVE_PROFILE has has been successfully stopped now..." 1>&2
else
for PROCESS_ID in $PIDS; do
counter=1
until [ $counter -gt 150 ]
do
if ps -p $PROCESS_ID > /dev/null; then
echo "Waiting for the process($PROCESS_ID) to finish on it's own for $(( 300 - $(( $counter*5)) ))seconds..."
sleep 2s
((counter++))
else
echo "$APP_NAME with PROCESS_ID:$PROCESS_ID is stopped now.."
exit 0;
fi
done
echo "Forcefully Killing $APP_NAME with PROCESS_ID:$PROCESS_ID."
kill -9 $PROCESS_ID
done
fi
If you are using spring boot version 2.3 n up , There is an in-build way to shutdown app gracefully.Add below in application.properties
server.shutdown=graceful spring.lifecycle.timeout-per-shutdown-phase=20s
If you are using lower spring boot version, You can write a custom shutdown hook and handle different beans, how they should shutdown or in which order they should shutdown. Example code below.
@Component
public class AppShutdownHook implements ApplicationListener<ContextClosedEvent> {
private static final Logger logger = LoggerFactory.getLogger(AppShutdownHook.class);
@Override
public void onApplicationEvent(ContextClosedEvent event) {
logger.info("shutdown requested !!!");
try {
//TODO Add logic to shutdown, diff elements of your application
} catch (Exception e) {
logger.error("Exception occcured while shutting down Application:", e);
}
}
}
They are many ways to shutdown a spring application. One is to call close() on the ApplicationContext:
ApplicationContext ctx =
SpringApplication.run(HelloWorldApplication.class, args);
// ...
ctx.close()
Your question suggest you want to close your application by doing Ctrl+C, that is frequently used to terminate a command. In this case...
Use endpoints.shutdown.enabled=true is not the best recipe. It means you expose an end-point to terminate your application. So, depending on your use case and your environment, you will have to secure it...
Ctrl+C should work very well in your case. I assume your issue is caused by the ampersand (&) More explanation:
A Spring Application Context may have register a shutdown hook with the JVM runtime. See ApplicationContext documentation.
I don't know if Spring Boot configure this hook automatically as you said. I assume it is.
On Ctrl+C, your shell sends an INT signal to the foreground application. It means "please interrupt your execution". The application can trap this signal and do cleanup before its termination (the hook registered by Spring), or simply ignore it (bad).
nohup is command that execute the following program with a trap to ignore the HUP signal. HUP is used to terminate program when you hang up (close your ssh connexion for example). Moreover it redirects outputs to avoid that your program blocks on a vanished TTY. nohupdoes NOT ignore INT signal. So it does NOT prevent Ctrl+C to work.
I assume your issue is caused by the ampersand (&), not by nohup. Ctrl+C sends a signal to the foreground processes. The ampersand causes your application to be run in background. One solution: do
kill -INT pid
Use kill -9 or kill -KILL is bad because the application (here the JVM) cannot trap it to terminate gracefully.
Another solution is to bring back your application in foreground. Then Ctrl+C will work. Have a look on Bash Job control, more precisely on fg.
As to @Jean-Philippe Bond 's answer ,
here is a maven quick example for maven user to configure HTTP endpoint to shutdown a spring boot web app using spring-boot-starter-actuator so that you can copy and paste:
1.Maven pom.xml:
<dependency>
<groupId>org.springframework.boot</groupId>
<artifactId>spring-boot-starter-actuator</artifactId>
</dependency>
2.application.properties:
#No auth protected
endpoints.shutdown.sensitive=false
#Enable shutdown endpoint
endpoints.shutdown.enabled=true
All endpoints are listed here:
3.Send a post method to shutdown the app:
curl -X POST localhost:port/shutdown
if you need the shutdown method auth protected, you may also need
<dependency>
<groupId>org.springframework.boot</groupId>
<artifactId>spring-boot-starter-security</artifactId>
</dependency>
If you are using maven you could use the Maven App assembler plugin.
The daemon mojo (which embed JSW) will output a shell script with start/stop argument. The stop will shutdown/kill gracefully your Spring application.
The same script can be used to use your maven application as a linux service.
Spring Boot provided several application listener while try to create application context one of them is ApplicationFailedEvent. We can use to know weather the application context initialized or not.
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.springframework.boot.context.event.ApplicationFailedEvent;
import org.springframework.context.ApplicationListener;
public class ApplicationErrorListener implements
ApplicationListener<ApplicationFailedEvent> {
private static final Logger LOGGER =
LoggerFactory.getLogger(ApplicationErrorListener.class);
@Override
public void onApplicationEvent(ApplicationFailedEvent event) {
if (event.getException() != null) {
LOGGER.info("!!!!!!Looks like something not working as
expected so stoping application.!!!!!!");
event.getApplicationContext().close();
System.exit(-1);
}
}
}
Add to above listener class to SpringApplication.
new SpringApplicationBuilder(Application.class)
.listeners(new ApplicationErrorListener())
.run(args);
If you are using the actuator module, you can shutdown the application via JMX or HTTP if the endpoint is enabled.
add to application.properties:
endpoints.shutdown.enabled=true
Following URL will be available:
/actuator/shutdown - Allows the application to be gracefully shutdown (not enabled by default).
Depending on how an endpoint is exposed, the sensitive parameter may be used as a security hint.
For example, sensitive endpoints will require a username/password when they are accessed over HTTP (or simply disabled if web security is not enabled).
From the Spring boot documentation
For Spring boot web apps, Spring boot provides the out-of-box solution for graceful shutdown from version 2.3.0.RELEASE.
An excerpt from Spring doc
Refer this answer for the Code Snippet
Use the static exit() method in the SpringApplication class for closing your spring boot application gracefully.
public class SomeClass {
@Autowire
private ApplicationContext context
public void close() {
SpringApplication.exit(context);
}
}
Spring Boot now supports graceful shut down (currently in pre-release versions, 2.3.0.BUILD-SNAPSHOT)
When enabled, shutdown of the application will include a grace period of configurable duration. During this grace period, existing requests will be allowed to complete but no new requests will be permitted
You can enable it with:
server.shutdown.grace-period=30s
SpringApplication implicitly registers a shutdown hook with the JVM to ensure that ApplicationContext is closed gracefully on exit. That will also call all bean methods annotated with @PreDestroy. That means we don't have to explicitly use the registerShutdownHook() method of a ConfigurableApplicationContext in a boot application, like we have to do in spring core application.
@SpringBootConfiguration
public class ExampleMain {
@Bean
MyBean myBean() {
return new MyBean();
}
public static void main(String[] args) {
ApplicationContext context = SpringApplication.run(ExampleMain.class, args);
MyBean myBean = context.getBean(MyBean.class);
myBean.doSomething();
//no need to call context.registerShutdownHook();
}
private static class MyBean {
@PostConstruct
public void init() {
System.out.println("init");
}
public void doSomething() {
System.out.println("in doSomething()");
}
@PreDestroy
public void destroy() {
System.out.println("destroy");
}
}
}
If you are in a linux environment all you have to do is to create a symlink to your .jar file from inside /etc/init.d/
sudo ln -s /path/to/your/myboot-app.jar /etc/init.d/myboot-app
Then you can start the application like any other service
sudo /etc/init.d/myboot-app start
To close the application
sudo /etc/init.d/myboot-app stop
This way, application will not terminate when you exit the terminal. And application will shutdown gracefully with stop command.
Source: Stackoverflow.com