Dan Christensen via dovecot wrote:

I should add that these problems arise frequently in practice, even if changes are only being made on a single machine, e.g. laptop1. Flag changes on laptop1 propagate to server, but then are reverted on server when laptop2 syncs with server, and then are reverted on laptop1 when it syncs again with server. If I redo the flag changes on laptop1, the process just repeats and they are removed again.

Suggestions for debugging are very welcome, as currently syncing is much too unreliable for me to continue using it.

Dan

I (and others) have the same issue.

A quote from Stephan Bosch on 20-1-2019:

...

We're working on it. We have reproduced the problem without difficulty and isolated the commit that caused it.

Tracking internally as DOP-842.

Regards,

Stephan.

Marc

On Feb 16, 2019, Dan Christensen via dovecot <dovecot@dovecot.org> wrote:

...

I'm running dovecot 2.3.4.1 from https://repo.dovecot.org/ on Ubuntu 18.04 on three machines that I'll call server, laptop1 and laptop2.

Both laptop1 and laptop2 run dovecot sync against server to keep local copies of my imap folders. Even when I initially had only two machines, laptop1 and server, I occasionally noticed that flags were lost, usually custom flags used by Gnus, but I couldn't reliably reproduce the problem.

Now that I have two laptops syncing against the server, the problem has gotten worse and I figured out a way to reproduce it:

• on server: create new IMAP folder test, and put two read messages in it
• on laptop1: doveadm sync -u user -l 10 -m test -f user@server
• on laptop2: doveadm sync -u user -l 10 -m test -f user@server

At this point, all three machines show the two messages M1 and M2 as being read.

• on laptop1: mark message M1 unread
• on laptop2: mark message M2 unread
• on laptop1: doveadm sync -u user -l 10 -m test -f user@server Both laptop1 and server have M1 unread, M2 read, as expected.
• on laptop2: doveadm sync -u user -l 10 -m test -f user@server Now laptop2 and server have M1 *read*, M2 unread.
• on laptop1: doveadm sync -u user -l 10 -m test -f user@server Now laptop1 and the server have both M1 and M2 *read*.
• on laptop2: doveadm sync -u user -l 10 -m test -f user@server Now laptop2 has both read as well.

The two lines that say "*read*" are wrong in my opinion. dsync propagated a read mark to an unread message, even though that message was marked unread more recently than it was marked read.

I usually use stateful sync, and get many related problems. I just did a test in which M1 and M2 started out read, and I started with empty files named dstate.test on laptop1 and laptop2. Then I did the above procedure, using the command

doveadm sync -u user -l 10 -m test -s "cat dstate.test" user@server > dstate.test

At the end, laptop2 and server had both messages unread (which is good), but laptop1 had only M1 unread, and repeated runs of the sync command did not correct this. So the stateful sync failed to detect a change.

Are these bugs in dovecot? Is there more information that I can provide? The output of doveconf -n on one machine is below, and the others are almost identical.

Thanks for any help!

Dan

2.3.4.1 (3c0b8769e): /etc/dovecot/dovecot.conf

OS: Linux 4.15.0-45-generic x86_64 Ubuntu 18.04.1 LTS

Hostname: laptop2

auth_mechanisms = plain login listen = 127.0.0.1 mail_index_log2_max_age = 10 days mail_index_log_rotate_min_age = 1 days mail_index_log_rotate_min_size = 300 k mail_location = maildir:~/Maildir namespace inbox { inbox = yes location = mailbox Drafts { special_use = \Drafts } mailbox Junk { special_use = \Junk } mailbox Sent { special_use = \Sent } mailbox "Sent Messages" { special_use = \Sent } mailbox Trash { special_use = \Trash } prefix = } passdb { args = scheme=CRYPT username_format=%u /etc/dovecot/users driver = passwd-file } protocols = imap service imap-login { inet_listener imap { address = * port = 143 } inet_listener imaps { address = * port = 943 ssl = yes } } service imap { process_limit = 25 } ssl_cert = </etc/dovecot/private/dovecot.pem ssl_client_ca_dir = /etc/ssl/certs ssl_dh = # hidden, use -P to show it ssl_key = # hidden, use -P to show it userdb { args = username_format=%u /etc/dovecot/users driver = passwd-file } protocol lda { postmaster_address = [elided] } protocol imap { mail_max_userip_connections = 20 }

In another thread, Timo wrote:

On Mar 12, 2019, Timo Sirainen via dovecot <dovecot@dovecot.org> wrote:

That bug is fixed with attached patch.

Thanks! I'm attaching the patch here, so it is in this thread as well.

A couple of questions before I test this:

• Do I need the patch on the remote end of the sync, or the local end, or both?

• Does it make sense to try it based on the master-2.3 branch from git? Following https://wiki.dovecot.org/CompilingSource ?

I'll report back once I've tested it.

Dan

On 13 Mar 2019, at 22.43, Dan Christensen via dovecot <dovecot@dovecot.org> wrote:

On Mar 12, 2019, Dan Christensen via dovecot <dovecot@dovecot.org> wrote:

...

In another thread, Timo wrote:

On Mar 12, 2019, Timo Sirainen via dovecot <dovecot@dovecot.org> wrote:

...

That bug is fixed with attached patch.

I'll report back once I've tested it.

I applied 2656.patch to version 2.3.5 as found at

https://repo.dovecot.org/ce-2.3-latest/ubuntu/bionic/pool/main/2.3.5-1_ce/

and rebuilt the ubuntu package. I installed the resulting package on all three machines. And still I can reproduce the bug involving unread messages getting marked as read.

Looks like you're also using Maildir, which has another bug of keywords not being copied correctly.

Over many months I've been getting frequent errors in synchronization using dsync. They are hard to reproduce, and aren't always of the same nature, but there's a particular case that I can reproduce 100% of the time. It involves read status being synced in the wrong direction when two machines are both synchronizing with a server.

I'm reposting below a message that gives the method for reproducing the problem. I can still reproduce it with Ubuntu's 2:2.3.7.2-1~bionic packages.

I also have problems with only one machine synchronizing with a server, but those I can't reproduce on demand.

I'm happy to try patches or test other things.

I'm using Maildir storage.

Thanks for any suggestions!

Dan

On Feb 16, 2019, Dan Christensen via dovecot <dovecot@dovecot.org> wrote:

I'm running dovecot 2.3.4.1 from https://repo.dovecot.org/ on Ubuntu 18.04 on three machines that I'll call server, laptop1 and laptop2.

Both laptop1 and laptop2 run dovecot sync against server to keep local copies of my imap folders. Even when I initially had only two machines, laptop1 and server, I occasionally noticed that flags were lost, usually custom flags used by Gnus, but I couldn't reliably reproduce the problem.

Now that I have two laptops syncing against the server, the problem has gotten worse and I figured out a way to reproduce it:

• on server: create new IMAP folder test, and put two read messages in it
• on laptop1: doveadm sync -u user -l 10 -m test -f user@server
• on laptop2: doveadm sync -u user -l 10 -m test -f user@server

At this point, all three machines show the two messages M1 and M2 as being read.

• on laptop1: mark message M1 unread
• on laptop2: mark message M2 unread
• on laptop1: doveadm sync -u user -l 10 -m test -f user@server Both laptop1 and server have M1 unread, M2 read, as expected.
• on laptop2: doveadm sync -u user -l 10 -m test -f user@server Now laptop2 and server have M1 *read*, M2 unread.
• on laptop1: doveadm sync -u user -l 10 -m test -f user@server Now laptop1 and the server have both M1 and M2 *read*.
• on laptop2: doveadm sync -u user -l 10 -m test -f user@server Now laptop2 has both read as well.

The two lines that say "*read*" are wrong in my opinion. dsync propagated a read mark to an unread message, even though that message was marked unread more recently than it was marked read.

I usually use stateful sync, and get many related problems. I just did a test in which M1 and M2 started out read, and I started with empty files named dstate.test on laptop1 and laptop2. Then I did the above procedure, using the command

doveadm sync -u user -l 10 -m test -s "cat dstate.test" user@server > dstate.test

At the end, laptop2 and server had both messages unread (which is good), but laptop1 had only M1 unread, and repeated runs of the sync command did not correct this. So the stateful sync failed to detect a change.

Are these bugs in dovecot? Is there more information that I can provide? The output of doveconf -n on one machine is below, and the others are almost identical.

Thanks for any help!

Dan

2.3.4.1 (3c0b8769e): /etc/dovecot/dovecot.conf

OS: Linux 4.15.0-45-generic x86_64 Ubuntu 18.04.1 LTS

Hostname: laptop2

auth_mechanisms = plain login listen = 127.0.0.1 mail_index_log2_max_age = 10 days mail_index_log_rotate_min_age = 1 days mail_index_log_rotate_min_size = 300 k mail_location = maildir:~/Maildir namespace inbox { inbox = yes location = mailbox Drafts { special_use = \Drafts } mailbox Junk { special_use = \Junk } mailbox Sent { special_use = \Sent } mailbox "Sent Messages" { special_use = \Sent } mailbox Trash { special_use = \Trash } prefix = } passdb { args = scheme=CRYPT username_format=%u /etc/dovecot/users driver = passwd-file } protocols = imap service imap-login { inet_listener imap { address = * port = 143 } inet_listener imaps { address = * port = 943 ssl = yes } } service imap { process_limit = 25 } ssl_cert = </etc/dovecot/private/dovecot.pem ssl_client_ca_dir = /etc/ssl/certs ssl_dh = # hidden, use -P to show it ssl_key = # hidden, use -P to show it userdb { args = username_format=%u /etc/dovecot/users driver = passwd-file } protocol lda { postmaster_address = [elided] } protocol imap { mail_max_userip_connections = 20 }