On Wed, 9 Jun 2021, at 3:30 PM, glad@xxxxx wrote:
Jun 8 14:33:51 node3 imap[1096067]: requested partition directory for unknown partition 'user/xxx'
That's very strange. It seems to be trying to treat "user/xxx" as a partition name, when it's clearly an account name.
So you can move from node0 (2.5) to anywhere, but you can't move from node1 or node3 (3.2) to node 2 (3.4)... and based on that log, the problem looks to be that the source side gets confused and gives up. Which makes me wonder: can you xfer from any other node back to node0? Can you xfer from node2 to nodes 1 or 3? Can you xfer between nodes 1 and 3?
My current gut feeling is that there's some bug in 3.2 that's preventing moving mailboxes out (which is where we see the weird "requested partition directory..." log. If so, I expect the same bug exists in 3.4, because I don't think we've fixed anything particularly XFER-related recently. So, I'm expecting you won't be able to XFER back to node0 from any of the other nodes, with the same symptoms. Is that correct?
Another question: have you ever run 3.0 on any of these nodes? If you did, and it was able to XFER fine, then that suggests that whatever the problem is was introduced in 3.2. If you skipped 3.0, then it might have been affected too and you just never noticed.
And I guess another question is: what happens if you specify a destination partition (as a third xfermailbox argument) rather than leaving it to default?
And what about: if you're able to invoke XFER over IMAP directly (rather than using the cyradm "xfermailbox" command), does it work like that? This might suggest cyradm is the real issue...
Sorry, I don't know much about XFER specifically, so I'm kinda throwing out every question I can think of to narrow down the shape of the problem!
Cheers,
ellie
