Hivemq high cpu usage

[lichvayne]

Hi everyone, im running Hivemq on amazon ec2 on container which has 1 cpu, 475 mb ram, 8.65 gb disk space. there i have 2 connections and total inbound publish messages are 319 and total outbound publish messages are 313. problem is that cpu is always on 100%, ram 150-200/475mb and dis space 3.70/8.65. can anyone help me identify why cpu always run on 100%?

im using my authentication extension

public class CustomAuthenticatorService implements SimpleAuthenticator {

private static final Logger logger = LoggerFactory.getLogger(CustomAuthenticatorService.class);
private final HttpClient httpClient = HttpClient.newHttpClient();

@Override
public void onConnect(@NotNull SimpleAuthInput simpleAuthInput, @NotNull SimpleAuthOutput simpleAuthOutput) {
    try {
        ConnectPacket connectPacket = simpleAuthInput.getConnectPacket();
        String clientId = connectPacket.getClientId();
        logger.info("Started authentication of device: " + clientId);

        if (connectPacket.getPassword().isEmpty()) {
            System.out.println("Authentication failed: invalid client credentials " + clientId);
            simpleAuthOutput.failAuthentication("Invalid Client credentials");
            return;
        }
        ByteBuffer readOnlyBuffer = connectPacket.getPassword().get().asReadOnlyBuffer();
        CharBuffer charBuffer = StandardCharsets.US_ASCII.decode(readOnlyBuffer);
        String text = charBuffer.toString();
        WashingDeviceAuthParam washingDeviceAuthParam = new WashingDeviceAuthParam(text, clientId);
        String jsonPayload = washingDeviceAuthParam.toJson();

        HttpRequest httpRequest = HttpRequest
                .newBuilder(URI.create("autentication endpoint"))
                .header("Content-Type", "application/json")
                .POST(HttpRequest.BodyPublishers.ofString(jsonPayload))
                .build();

        HttpResponse<String> httpResponse = httpClient.send(httpRequest, HttpResponse.BodyHandlers.ofString());

        if (httpResponse.statusCode() >= 200 && httpResponse.statusCode() <= 299 && httpResponse.body().equals("1")) {
            logger.info("Client successfully authenticated: " + clientId);
            simpleAuthOutput.authenticateSuccessfully();
        } else {
            logger.info("Client failed authentication: " + clientId);
            simpleAuthOutput.failAuthentication("Invalid credentials");
        }

    } catch (Exception exception) {
        logger.info("Client failed authentication: ");
        logger.info("Exception during authentication: " + exception);
        simpleAuthOutput.failAuthentication("Error during authentication");
    }
}

}

[Daria_H]

Hi @lichvayne

Your high CPU usage is likely caused by the synchronous HTTP request in your authentication extension. Here’s what’s happening:

1. Blocking HTTP Call in Authentication

   • The httpClient.send(httpRequest, HttpResponse.BodyHandlers.ofString()); call is blocking, meaning the authentication thread waits until the HTTP response is received.
   • If the authentication endpoint is slow, under high load, or has network issues, this causes delays and spikes in CPU usage.
   • Multiple clients trying to authenticate at the same time will worsen the issue since HiveMQ will use multiple threads waiting for these responses.

2. CPU Resource Constraints

   • Your instance only has 1 CPU and 475MB RAM, which is very limited for running HiveMQ + authentication logic.
   • The combination of HiveMQ handling MQTT messages and blocking HTTP requests leads to CPU saturation.

Solutions to Fix High CPU Usage

1. Use Asynchronous HTTP Requests

Instead of blocking the authentication thread, use sendAsync():

httpClient.sendAsync(httpRequest, HttpResponse.BodyHandlers.ofString())
    .thenAccept(httpResponse -> {
        if (httpResponse.statusCode() >= 200 && httpResponse.statusCode() <= 299 && httpResponse.body().equals("1")) {
            logger.info("Client successfully authenticated: " + clientId);
            simpleAuthOutput.authenticateSuccessfully();
        } else {
            logger.info("Client failed authentication: " + clientId);
            simpleAuthOutput.failAuthentication("Invalid credentials");
        }
    })
    .exceptionally(exception -> {
        logger.info("Exception during authentication: " + exception);
        simpleAuthOutput.failAuthentication("Error during authentication");
        return null;
    });

This prevents HiveMQ authentication threads from being blocked.

2. Use a Local Cache to Reduce HTTP Requests

If the authentication response doesn’t change often, cache successful authentications using a ConcurrentHashMap:

private final Map<String, Boolean> authCache = new ConcurrentHashMap<>();

@Override
public void onConnect(@NotNull SimpleAuthInput simpleAuthInput, @NotNull SimpleAuthOutput simpleAuthOutput) {
    ConnectPacket connectPacket = simpleAuthInput.getConnectPacket();
    String clientId = connectPacket.getClientId();

    if (authCache.containsKey(clientId) && authCache.get(clientId)) {
        logger.info("Client authenticated from cache: " + clientId);
        simpleAuthOutput.authenticateSuccessfully();
        return;
    }

    // Perform async authentication...
    httpClient.sendAsync(httpRequest, HttpResponse.BodyHandlers.ofString())
        .thenAccept(httpResponse -> {
            if (httpResponse.statusCode() >= 200 && httpResponse.body().equals("1")) {
                authCache.put(clientId, true);
                simpleAuthOutput.authenticateSuccessfully();
            } else {
                simpleAuthOutput.failAuthentication("Invalid credentials");
            }
        })
        .exceptionally(exception -> {
            simpleAuthOutput.failAuthentication("Error during authentication");
            return null;
        });
}

This avoids repeatedly hitting the authentication endpoint.

3. Upgrade Your EC2 Instance

• 1 CPU & 475MB RAM is too low for HiveMQ.
  Consider upgrading to at least:
  • 2 vCPUs
  • 2GB RAM
  • 20GB Disk
• If you have a high number of clients connecting simultaneously, increasing resources will prevent performance bottlenecks.

Final Recommendation

1. Change send() to sendAsync()
2. Implement a local cache to reduce HTTP calls
3. Upgrade your EC2 instance if possible

Let me know if you need further debugging!

Best,
Dasha from The HiveMQ Team