quarkus-jgroups-chat

Chat demo using the quarkus-jgroups extension

Running the demo

mvn package creates a runner JAR in the target directory. This can be run via the java -jar JAR command (in a first shell):

cd ./target
java -jar -Dquarkus.http.port=8080 quarkus-jgroups-chat-1.0.0-SNAPSHOT-runner.jar 

In a second shell, run

cd ./target
java -jar -Dquarkus.http.port=8081 quarkus-jgroups-chat-1.0.0-SNAPSHOT-runner.jar 

If this is on the same host, quarkus.http.port needs to be different for each instance. It does not need to be specified if each instance is run on a different host.

The 2 instances should find each other and form a cluster. This is seen in the output:

\-------------------------------------------------------------------
GMS: address=belasmac-39930, cluster=quarkus-jgroups-chat, physical address=127.0.0.1:7800
-------------------------------------------------------------------
-- view: [belasmac-29858|9] (2) [belasmac-29858, belasmac-39930]

As can be seen view belasmac-29858|9 has 2 members.

The chat instances can now be accessed via HTTP to send and receive messages. This can be done in a number of ways. The first one is to subscribe to messages:

http://localhost:8080/streaming.html

This will use Javascript to listen for server-side events on /chat/subscribe.

To send messages, URL /chat/send/message can be used, e.g. via curl (curl http://localhost:8080/chat/send/hello) or a web browser:

http://localhost:8082/chat/send/hello%20world

This accesses the second instance. Note that a message sent to one of the 2 instances will be sent to all instances, and each instance will print it to stdout. The above web client listening on /chat/subscribe/ will also print the message.

Creating the native demo

mvn package -Pnative creates the native demo. The rest of the steps is the same as for the non-native demo above.

Multiple instances can be run by changing the HTTP port, e.g. -Dquarkus.http.port=8001

Misc

Note that in order to generate a native image, the following properties need to be enabled on the quarkus-maven-plugin (in the POM):

<reportErrorsAtRuntime>true</reportErrorsAtRuntime>: the native build analyzes code like this:

@Property(description="Use InetAddress.isReachable() to verify suspected member instead of regular messages")
protected boolean use_icmp;
...
if(use_icmp)
    s.forEach(this::verifySuspectWithICMP);
else
    verifySuspect(s);

Since use_icmp is not final, the analysis needs to look at both branches, including verifySuspectWithICMP().

This method, however, calls a method (NetworkInterface.isReachable()) that's not implemented by the GraalVM, and therefore terminates the build with an error.

The default for use_icmp is false and unless a user changes this, this branch of the code will never be visited. However, this branch cannot be optimized away, as use_icmp might be set to true (even at runtime).

We have a couple of occurrences of this, so currently we decided to report issues caused by this at run time rather than build time.

<enableJni>true</enableJni>: don't know why this is needed, but if it isn't enabled, the native build fails with a non-sensical error message. Will need to revisit with later Quarkus builds...