18 Apr
2024
18 Apr
'24
11:58 p.m.
Greeting,
On Thu, 18 Apr 2024 17:51:06 +0200, Kirill A. Korinsky wrote:
All of this allows to reconstruct the workflow which leads to an issue:
- mx2 had a new mail.
- mx2 registered that mail as 147699 inside virtual folder.
- by some reason this email wasn't replicated to mx1 for at least 30-40 seconds. Maybe it runs full-scan on all users, I have no idea.
- mx1 had a new mail as well.
- mx1 resitered that mail as 147699 inside virtual folder.
- mx1's mail was replicated to mx2 as 147700
- mx2's mail was replicated back to mx1 as 147701
I was wrong. I run a series of expirements and achived a strategy which allows to reproduce this issue.
I've used the following definition of virtual folder all:
- !Archive -Trash -Trash/* -Junk -Junk/* all
I also have two servers: mx1 and mx2.
So, to reproduce an issue with dsync msgid inside virtual folder I need:
- Connect to mx1 via imap
- Connect to mx2 via smtp
- Initialize sending a mail via smtp
- In parallel move some mail from Inbox to Virtual.All
With quite big probability I broke the order.
When I enable the debug log, it slow down things because dovecot and smtpd uses the same /var/log/maillog on my system.
Another way to avoid such condition, is switch off multi core on machine.
Without parallerism I can't reproduce the issue.
-- wbr, Kirill