I assumed with the synchronized sender this would not have been possible.

that’s correct. if you correctly synchronize the access then it should not be possible. are you sure your synchronized all places where you access the sender? the snippet you shared - is this the only place touching the Sender?

There are 2 places, both wrapped exactly like this. I'm puzzled as well and it's hard to debug as it happens infrequently. As you've confirmed I'm not doing anything too foolish I will keep investigating. Thanks.

your code looks OK, I don’t see anything incorrect. the whole interaction with the Sender is guarded by a mutex. that’s akin to a single threaded access.

Thanks for looking, I found the problem, it was my bug. I thought the trade.gets() were doing a simple native value fetch that couldn't go wrong. Turns out those gets can fail and when they do they leave the previous send incomplete. Thanks again.

oh, that explains it! good one!