Hello,
Thanks everyone for the attention and advice. Following the documentation, I did implement "noreplicate". To do so, I found an unused attribute in my LDAP's schema, mailDeliveryOption. I probably could have made a new one, but I wanted to be done before the workweek started in earnest, so I just quickly used that one. In this attribute, I wrote a quick script that users with "mdbox" format specified in mailMessageStore got the value "no", and users without any mailMessageStore override (therefore, using the default of mbox storage) got the value "yes".
I then amended my /etc/dovecot/dovecot-ldap.conf.ext and assigned those values into the userdb (and in the pass db prefetch too):
user_attrs = homeDirectory=home,uidNumber=uid,gidNumber=gid,mailmessagestore=mail,mailDeliveryOption=noreplicate pass_attrs = uid=user,userPassword=password,homeDirectory=userdb_home,uidNumber=userdb_uid,gidNumber=userdb_gid,mailmessagestore=userdb_mail,mailDeliveryOption=userdb_noreplicate
I pushed the changes, restarted dovecot, and then ran 'doveadm replicator replicate "*"' to force a global resync. After waiting a while, all 78 user accounts on the old-style storage format dropped out of the "doveadm replicator status" table. I was surprised to find there are still 78 such accounts, but that exactly matches the number of "mailDeliveryOption: yes" values in LDAP. So now users with mbox storage on NFS mounts are not replicating, and the crashes / backtraces have disappeared from the logs.
Hopefully there are no other concerns with this setup, in case one of the 2 active hosts has a problem or is rebooted, such as index cache. But, dovecot so far seems to be smart enough to detect this sort of problem and fsck the indexes / reset the IMAP connection.
Thanks again everyone!
Ben
On 8/5/24 07:57, Markus Bach via dovecot wrote:
Since v2.3.1 you can disable replication for a user by providing |noreplicate| user database field https://doc.dovecot.org/configuration_manual/authentication/user_database_ex....
https://doc.dovecot.org/configuration_manual/replication/
On 8/5/24 13:14, Aki Tuomi via dovecot wrote:
I can't recall off the bat what version it was added but there is noreplicate key in userdb reply that can be used to stop replication for a user.
Aki
On 05/08/2024 14:09 EEST Benjamin Rose via dovecotdovecot@dovecot.org wrote:
Hello,
Disappointing, but understandable. It is not trivial to move mbox users who are used to direct access for many decades now.
As a workaround, is it possible to disable replication on a per-user basis? Since mbox files are stored on an NFS-mounted /u (vs mdbox which is stored locally in /home), I do not think it is really needed for this dozen or so users. These users are differentiated by the lack of a mailMessageStore LDAP value in their DN, vs users such as myself on the newer format which have a mailMessageStore value of "mdbox:/home/benrose/mail".
Thanks, Ben
On 8/5/24 03:38, Aki Tuomi via dovecot wrote:
I just realized that the bug is in mbox handling, which is nowadays frozen, so even if there is a bug, we won't fix it.
Have you considered suggesting to your mbox users that they could use pine with imap instead of direct access?
Aki
On 05/08/2024 09:35 EEST Benjamin Rose via dovecotdovecot@dovecot.org wrote:
Hello,
The requested information has been sent off-list.
Thanks, Ben
On 8/5/24 02:10, Aki Tuomi via dovecot wrote:
Hi!
We have seen this before but were unable to reproduce the issue.
Could you please send directly to me the core file processed with
https://raw.githubusercontent.com/dovecot/core/master/src/util/dovecot-sysre...
You should use
dovecot-sysreport --core /path/to/core /usr/lib/dovecot/imap
as parameters.
Also, if possible, index files that match the core from both source & destination server. If you can't provide these, we'll be happy to look just at the core.
Aki
> On 05/08/2024 06:59 EEST Benjamin Rose via > dovecotdovecot@dovecot.org wrote: > > Hello, > > I am running Dovecot 2.3.16 (7e2e900c1a) on RHEL 9.2. I attach my > "doveconf -n" configuration. I have replication enabled between 2 > servers, both very beefy with 16 cores of Intel(R) Xeon(R) Gold > 6244 CPU > @ 3.60GHz, 768gb memory, and 100-gig ethernet. The configs are > the same > except the replication part which points at each other via a puppet > template. It seems to work well for most users on modern email > clients > using the mdbox storage format. The issue I'm having is that > some of my > users are on the old-style mbox storage format so they can use > legacy > mail readers such as pine and mutt natively. For some of these > users, > syncing works just fine. For others, right now about a dozen > users, the > sync never completes, and I get the following error & backtrace in > /var/log/maillog: > > Aug 4 23:54:22 mail6 dovecot[1177530]: doveadm: Panic: file > dsync-mailbox-import.c: line 2163 > (dsync_mailbox_import_handle_mail): > assertion failed: (array_count(&wanted_uids) > 0) > Aug 4 23:54:22 mail6 dovecot[1177530]: doveadm: Error: Raw > backtrace: > /usr/lib64/dovecot/libdovecot.so.0(backtrace_append+0x46) > [0x7fa6ab05c486] -> > /usr/lib64/dovecot/libdovecot.so.0(backtrace_get+0x22) > [0x7fa6ab05c5a2] > -> /usr/lib64/dovecot/libdovecot.so.0(+0x10a41b) > [0x7fa6ab06b41b] -> > /usr/lib64/dovecot/libdovecot.so.0(+0x10a4b7) [0x7fa6ab06b4b7] -> > /usr/lib64/dovecot/libdovecot.so.0(+0x5d11a) [0x7fa6aafbe11a] -> > dovecot/doveadm-server(+0x24029) [0x56433f2f7029] -> > dovecot/doveadm-server(dsync_mailbox_import_changes_finish+0x29c) > [0x56433f32fb9c] -> > dovecot/doveadm-server(dsync_brain_sync_mails+0x8d5) > [0x56433f331f15] -> dovecot/doveadm-server(+0x52255) > [0x56433f325255] -> > dovecot/doveadm-server(+0x52639) [0x56433f325639] -> > dovecot/doveadm-server(+0x5fa53) [0x56433f332a53] -> > /usr/lib64/dovecot/libdovecot.so.0(io_loop_call_io+0x6d) > [0x7fa6ab081cbd] -> > /usr/lib64/dovecot/libdovecot.so.0(io_loop_handler_run_internal+0x13a) > > [0x7fa6ab083bba] -> > /usr/lib64/dovecot/libdovecot.so.0(io_loop_handler_run+0x54) > [0x7fa6ab083c64] -> > /usr/lib64/dovecot/libdovecot.so.0(io_loop_run+0x40) > [0x7fa6ab083e20] -> dovecot/doveadm-server(+0x37d7c) > [0x56433f30ad7c] -> > dovecot/doveadm-server(+0x3974d) [0x56433f30c74d] -> > dovecot/doveadm-server(+0x5020b) [0x56433f32320b] -> > /usr/lib64/dovecot/libdovecot.so.0(io_loop_call_io+0x6d) > [0x7fa6ab081cbd] -> > /usr/lib64/dovecot/libdovecot.so.0(io_loop_handler_run_internal+0x13a) > > [0x7fa6ab083bba] -> > /usr/lib64/dovecot/libdovecot.so.0(io_loop_handler_run+0x54) > [0x7fa6ab083c64] -> > /usr/lib64/dovecot/libdovecot.so.0(io_loop_run+0x40) > [0x7fa6ab083e20] -> > /usr/lib64/dovecot/libdovecot.so.0(master_service_run+0x17) > [0x7fa6aaff2337] -> dovecot/doveadm-server(main+0x102) > [0x56433f2f8f92] > -> /lib64/libc.so.6(+0x3feb0) [0x7fa6aac3feb0] -> > /lib64/libc.so.6(__libc_start_main+0x80) [0x7fa6aac3ff60] -> > dovecot/doveadm-server(_start+0x25) [0x56433f2f9015] > Aug 4 23:54:22 mail6 dovecot[1177530]: doveadm: Fatal: master: > service(doveadm): child 1117475 killed with signal 6 (core dumped) > > I'm not sure what might be causing this, and wanted to see if > anyone > here could offer any suggestions. Please let me know if any > additional > information is needed. > > Thanks, > Ben > _______________________________________________ > dovecot mailing list --dovecot@dovecot.org > To unsubscribe send an email todovecot-leave@dovecot.org
dovecot mailing list --dovecot@dovecot.org To unsubscribe send an email todovecot-leave@dovecot.org
dovecot mailing list --dovecot@dovecot.org To unsubscribe send an email todovecot-leave@dovecot.org
dovecot mailing list --dovecot@dovecot.org To unsubscribe send an email todovecot-leave@dovecot.org
dovecot mailing list --dovecot@dovecot.org To unsubscribe send an email todovecot-leave@dovecot.org
dovecot mailing list --dovecot@dovecot.org To unsubscribe send an email todovecot-leave@dovecot.org
dovecot mailing list -- dovecot@dovecot.org To unsubscribe send an email to dovecot-leave@dovecot.org