Bug in replicating large messages

[timfox]

I've been tracking down a bug in failover of replicated large messages (in patch I sent to Clebert).

It seems this is because, on the backup server the message is always being created as durable=false (should be durable=true).

This means it gets given the wrong file extension (.tmp not .msg) so on failover the message can't be found.

It's an easy fix, but I can't work out how this code ever worked... Any ideas?

[timfox]

Another issue - LargeServerMessage map in replication end point - items aren't removed if the send was aborted - this could result in a memory leak.

[timfox]

No comments on these issues?

Also, one more: Why are you replicating non persistent large messages? Doesn't make sense.

[clebert.suconic]

"It's an easy fix, but I can't work out how this code ever worked... Any ideas?"

It seemed to be something else when I was debugging this. I would need to debug it.

Regarding the mapping.. I should send an endLargeMessage if the transmission is interrupted. (I.e. if I get another message on the ServerSession before completing the currentLargeMessage).

I would need to spend some time on it.