I have Axon 3.2.2 application running on two nodes (autoconfigured JPA token store). Single threaded tracking event processor sometimes “migrates” between nodes without any traces in logs.
After such a migration current index of event stream gets into past causing events processed twice!
there was a bug that made this possible at some point, when handling of events took a long time. This has been resolved, but can’t check in which version right now.
There is a 3.2.3 version and 3.3.5 as well. Consider upgrading as it may resolve your issue. If not, do let us know!
As of last week, we’ve also got version 3.3.6 available for you to use, with a couple of additional bug fixes.
If time allows it, you might even be able to wait for 3.4, which I assume should com quick as well (as 4.0 is coming the 18th of October).
So, just a heads up, please let us know if the version upgrade fixed your issue though!
I’ve made simple application with one tracking processor (four segments) and inserts event identifier into table with unique key on this column. Running 5 instances of this app against Axon Server (periodically restarting instance that owns tracking token) didn’t lead to the problem. Will look further.