• SolvedFixed

  • [SOLVED] Mail delivery issues for POP/IMAP in latest Grommunio releases

Also G-Web performs better when this modification is in place.

My results: on two server, this modification works perfect and the connection errors are gone. On a third server, Postfix can't access the virtual lookup tables in DB. Currently I did not have time to debug this, so I removed the configuration file from this server. Will post the results within the next days.

When you encounter postfix lookup errors, there are 2 options:

  1. remove this line skip_name_resolve=ON from /etc/my.cnf.d/grommunio.cnfand restart mariadb and postfix.

  2. verify this 4 Postfix lookup files:

    ls -1 /etc/postfix/grommunio-*.cf | sort
/etc/postfix/grommunio-bcc-forwards.cf
/etc/postfix/grommunio-virtual-mailbox-alias-maps.cf
/etc/postfix/grommunio-virtual-mailbox-domains.cf
/etc/postfix/grommunio-virtual-mailbox-maps.cf

    cat /etc/postfix/grommunio-*.cf | grep hosts
    The hosts string must be localhost like: hosts = localhost, if you see 127.0.0.1 change ist to localhost and restart mariadb and postfix.
    Reason: skip_name_resolve=ON disables name resolution for Maria DB, so 127.0.0.1 do not get resolved as localhost any longer.

  • crpb replied to this.

    WalterH Reason: skip_name_resolve=ON disables name resolution for Maria DB, so 127.0.0.1 do not get resolved as localhost any longer.

    even localhost? oh... should have read a bit further i guess 😛.

    Fixed my config above.....

      So I would like to cautiously throw my results into the round that the mysql tuning options on my server Rocky Linux 9.4 - aarch64) do not lead to the desired success. The mails still arrive every 10 minutes late (see midb.cfg)

      I have created the following entry:
      /etc/my.cnf.d/grommunio.cnf

      [mariadb]
innodb_buffer_pool_size=1G
max_allowed_packet=256M
net_read_timeout=3600
net_write_timeout=3600
read_rnd_buffer_size=8M
skip_name_resolve=ON
sort_buffer_size=8M
table_cache=1000
thread_cache_size=512 #this value is not adopted from the configuration file in my case, I then set it myself in mariadb to be able to test it
wait_timeout=288000

      The settings in /etc/gromox/midb.cfg

      midb_cache_interval=5minutes
midb_reload_interval=10minutes

      are still in place.

      On a positive note, the mysql [Warning]s have actually disappeared.

      (sorry for writing in German in the first place ...)

      It isn't persistent for me aswell... laggs are happening again.. 🙁

        So my MySQL tuning above doesn't resolve entirely the problem, it just makes the POP/IMAP delays happen way more less, so there's definitly room for improvement.

        I still get a couple mysql errors:

        2024-08-26 10:07:45 7 [Warning] Aborted connection 7 to db: 'grommunio' user: 'grommunio' host: 'localhost' (Got timeout reading communication packets)
2024-08-26 10:07:45 9 [Warning] Aborted connection 9 to db: 'grommunio' user: 'grommunio' host: 'localhost' (Got timeout reading communication packets)
2024-08-26 11:07:45 54 [Warning] Aborted connection 54 to db: 'grommunio' user: 'grommunio' host: 'localhost' (Got timeout reading communication packets)
2024-08-26 12:07:45 89 [Warning] Aborted connection 89 to db: 'grommunio' user: 'grommunio' host: 'localhost' (Got timeout reading communication packets)
2024-08-26 13:07:45 105 [Warning] Aborted connection 105 to db: 'grommunio' user: 'grommunio' host: 'localhost' (Got timeout reading communication packets)
2024-08-26 14:07:45 132 [Warning] Aborted connection 132 to db: 'grommunio' user: 'grommunio' host: 'localhost' (Got timeout reading communication packets)

        I'll try a bit more tuning. Currently trying to raise connect_timeout.

        crpb
        What is not persistent for you?
        The "thread_cache_size" entry?

        @ozy for me it sadly does not change anything.
        The first mail after restart is instant, the others are delayed around 10 mins.

            GeneralProbe

            how the mails show up in imap. well, new mails seems to come in ok. read/unread status seems a bit fonky..

              today I see the "[Warning] Aborted connection" again for all three servers.

              @crpb Not persistent for me too. There must be some side effect somewhere which is triggered by restarting mysql.
              I've implemented the midb cache/reload settings too meanwhile.

              Still, server runs faster with the mysql settings.
              The final settings I use which don't show any other connections error since 2 hours:
              Note that I added connect_timeout which seems to handle the remaining errors.

              [mysqld]
table_cache             = 1000
innodb_buffer_pool_size = 1G
read_rnd_buffer_size    = 8M
sort_buffer_size        = 8M
max_allowed_packet      = 256M
innodb_log_file_size    = 256M
wait_timeout            = 288000
innodb-file-per-table   = 1
net_read_timeout        = 3600
net_write_timeout       = 3600
connect_timeout         = 3600

                So the above mysql tuning with connect_timout set does not produce any errors since 6 hours.

                Strange part is, POP/IMAP has no delay with these setups for a while, and then fades back to delays defined by midb reload/cache settings.
                There's definitly some relation, but I can't understand which one.

                Sorry if I suggested this completely solves the delay issues, but I guess it will help diagnose them somehow.

                Right,

                2024-08-26 11:02:45 0 [Note] Starting MariaDB 10.5.22-MariaDB source revision 7e650253dc488debcb0898ebe6d385bf6dfa3656 as process 992
2024-08-26 11:02:45 0 [Note] InnoDB: Uses event mutexes
2024-08-26 11:02:45 0 [Note] InnoDB: Compressed tables use zlib 1.2.11
2024-08-26 11:02:45 0 [Note] InnoDB: Number of pools: 1
2024-08-26 11:02:45 0 [Note] InnoDB: Using ARMv8 crc32 + pmull instructions
2024-08-26 11:02:45 0 [Note] InnoDB: Using Linux native AIO
2024-08-26 11:02:45 0 [Note] InnoDB: Initializing buffer pool, total size = 1073741824, chunk size = 134217728
2024-08-26 11:02:45 0 [Note] InnoDB: Completed initialization of buffer pool
2024-08-26 11:02:45 0 [Note] InnoDB: 128 rollback segments are active.
2024-08-26 11:02:45 0 [Note] InnoDB: Creating shared tablespace for temporary tables
2024-08-26 11:02:45 0 [Note] InnoDB: Setting file './ibtmp1' size to 12 MB. Physically writing the file full; Please wait ...
2024-08-26 11:02:45 0 [Note] InnoDB: File './ibtmp1' size is now 12 MB.
2024-08-26 11:02:45 0 [Note] InnoDB: 10.5.22 started; log sequence number 658220; transaction id 1920
2024-08-26 11:02:45 0 [Note] InnoDB: Loading buffer pool(s) from /var/lib/mysql/ib_buffer_pool
2024-08-26 11:02:45 0 [Note] Plugin 'FEEDBACK' is disabled.
2024-08-26 11:02:45 0 [Note] Server socket created on IP: '::1'.
2024-08-26 11:02:45 0 [Note] InnoDB: Buffer pool(s) load completed at 240826 11:02:45
2024-08-26 11:02:46 0 [Warning] 'proxies_priv' entry '@% root@mail.example.com' ignored in --skip-name-resolve mode.
2024-08-26 11:02:46 0 [Note] Reading of all Master_info entries succeeded
2024-08-26 11:02:46 0 [Note] Added new Master_info '' to hash table
2024-08-26 11:02:46 0 [Note] /usr/libexec/mariadbd: ready for connections.
Version: '10.5.22-MariaDB'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MariaDB Server

                that's my log.

                But I do get the delays just after receiving the first mail.

                ozy innodb-file-per-table = 1

                you schuld never set this parameters in a config file like this one, this parameter change the behavior of the database and my render the database unusable.

                Setting: innodb_buffer_pool_size = 1G is in most cases to big for a mail system.

                  ozy gromox keeps the mysql connections open, ie in TIME_WAIT state

                  Yeah no. TIME_WAIT is a duplex-closed state, cf. https://datatracker.ietf.org/doc/html/rfc793#page-23 . It's also a TCP state and does not exist in Unix Domain Sockets (which are used here for SQL communications).

                  and mysql drops the connection after net_read_timeout and net_write_timeout is reached, which is 5 sec by default.

                  Indeed Gromox holds connections, but that counts as idle, and the corresponding mysqld variable is wait_timeout.
                  net_read_timeout is the maximum wait time to receive the rest of a command/query once the client has started issuing one. Queries are on the order of 200 bytes and local machine transfer speeds in excess of 10 GBps—all said, with practical overheads that should take just 12 µs. Of course, with enough system loading, debugger tricks (or bugs, for that matter), enough latency could theoretically be induced into the system to delay transmission and make processes run into said timeouts.

                  No need to bother with skip_name_resolve. Lookup of localhost/127 has a fastpath thanks to the static entry in /etc/hosts. It goes without saying that, because UDS are used, there is not even something to resolve in the first place.

                    Unfortunately, with all of this settings, I see this warnings: [Warning] Aborted connection some days more and some days less.

                    © 2020-2024 grommunio GmbH. All rights reserved. | https://grommunio.com | Data Protection | Legal notice