> Andreas, do you have a lot of installations in your production env? I guess the EJB could potentially time out while sending the messages to push networks (depends on your configuration of the wildfly container).

This is a very likely cause of the problem. Looking at the timestamps in the logs I have attached previously, the sending was aborted after 5 minutes. I logged into the WildFly admin console, and found that in "Container" -> "Transactions" the default timeout was set to 300s (5 minutes). I have now increased this to 7200s, and hope that this will solve the problem the next time a push message is sent. This would also explain why sending worked correctly some months ago, and now the number of device installations has increased significantly. Thanks for the help everyone, I am optimistic now!