ReconnectionManager giving up after an error when SM enabled

Boris_Grozev · December 5, 2024, 8:50pm

We ran into an issue with ReconnectionManager and stream resumption. I can reproduce with both 4.4.8 and 4.4.6. The scenario is the following:

Enable ReconnectionManager with fixed delay 1 second
Connect and login successfully
Restart the XMPP server, removing the virtual domain
Smack attempts to re-connect and gets a host-unknown error

Expected behavior:
Smack should continue to attempt to re-connect at 1 second interval.

Actual behavior:
Smack stops the re-connect attempts.

The issue only occurs when Stream Management is enabled. If we disable it, the behavior is as expected. Also, if we stop the XMPP server instead of mis-configuring it, so the TCP socket fails to connect then the behavior is as expected.

XMPPTCPConnection.setUseStreamManagementResumptionDefault(false);
XMPPTCPConnection.setUseStreamManagementDefault(false);

I suspect it has to do with ReconnectionManager running after the XMPPTCPConnection is marked as connected here, but I don’t understand the details:

github.com

igniterealtime/Smack/blob/3bb07521bbc063d805eae91a6f405a2ae8a85657/smack-tcp/src/main/java/org/jivesoftware/smack/tcp/XMPPTCPConnection.java#L851


      
           * @throws SmackException if Smack detected an exceptional situation.
           * @throws IOException if an I/O error occurred.
           * @throws InterruptedException if the calling thread was interrupted.
           */
          @Override
          protected void connectInternal() throws SmackException, IOException, XMPPException, InterruptedException {
              // Establishes the TCP connection to the server and does setup the reader and writer. Throws an exception if
              // there is an error establishing the connection
              connectUsingConfiguration();
          
              connected = true;
          
              // We connected successfully to the servers TCP port
              initConnection();
          
              // TLS handled will be true either if TLS was established, or if it was not mandatory.
              waitForConditionOrThrowConnectionException(() -> tlsHandled, "establishing TLS");
          
              // Wait with SASL auth until the SASL mechanisms have been received
              waitForConditionOrThrowConnectionException(() -> saslFeatureReceived, "SASL mechanisms stream feature from server");
          }

This file can help repro the problem (you need the server setup as well):

gist.github.com

https://gist.github.com/bgrozev/b1c720446afdfa6ef12e78f577f6803e

gistfile1.txt

import org.jivesoftware.smack.tcp.*;

import javax.net.ssl.*;
import java.security.cert.*;
import org.jivesoftware.smack.parsing.ExceptionLoggingCallback;
import org.jivesoftware.smack.SmackConfiguration;
import org.jivesoftware.smack.ConnectionListener;
import org.jivesoftware.smack.ReconnectionManager;
import org.jivesoftware.smack.ReconnectionListener;
import org.jivesoftware.smack.XMPPConnection;

This file has been truncated. show original

Flow · December 7, 2024, 1:50pm

Thanks for the detailed issue report. I hope to find the time to look into it soon.