Rebalancing Backends

Sometimes with long lasting IMAP connections you might end up in situation where you need to increase the amount of backends due to increased load.

Here’s a list of operations which should manually ease the situation and balance the load on the backend.

Start situation

doveadm director status

=========== ======= =======
backend     vhosts   users
=========== ======= =======
backend-01    100     80
backend-02    100     70
backend-03    100     60
backend-04    100     10
backend-05    100     10
============ ====== =======

(total users = 230, 230/5 = 46 per backend)

Update vhost count on node 1 and 2 to have vhost count = 0

# disable moving users to any backend that has too many of the users already:

doveadm director update backend-02 0

doveadm director update backend-03 0

Start flushing backends

doveadm director flush backend-01

=========== ======= =======
backend     vhosts   users
=========== ======= =======
backend-01    100     34
backend-02     0      70
backend-03     0      60
backend-04    100     33
backend-05    100     33
============ ====== =======

Adjust vhost count back and flush next backend

doveadm director update backend-02 100

doveadm director flush backend-02

=========== ======= =======
backend     vhosts   users
=========== ======= =======
backend-01    100     38
backend-02    100     55
backend-03     0      60
backend-04    100     38
backend-05    100     39
============ ====== =======

Adjust vhost count back and flush next backend

doveadm director update backend-03 100

doveadm director flush backend-03

=========== ======= =======
backend     vhosts   users
=========== ======= =======
backend-01    100     42
backend-02    100     58
backend-03    100     43
backend-04    100     43
backend-05    100     44
============ ====== =======

Flush next backend

doveadm director flush backend-04

=========== ======= =======
backend     vhosts   users
=========== ======= =======
backend-01    100     43
backend-02    100     59
backend-03    100     43
backend-04    100     40
backend-05    100     45
============ ====== =======

Flush next backend

doveadm director flush backend-05

=========== ======= =======
backend     vhosts   users
=========== ======= =======
backend-01    100     44
backend-02    100     60
backend-03    100     44
backend-04    100     41
backend-05    100     41
============ ====== =======

Flush next backend

# due to vhost changes, 02 still wasn’t fully balanced, so flush it again:

doveadm director flush backend-02

=========== ======= =======
backend     vhosts   users
=========== ======= =======
backend-01    100     46
backend-02    100     48
backend-03    100     46
backend-04    100     45
backend-05    100     45
============ ====== =======

Flush next backend

# still not entirely perfect. could flush all the other backends again to make sure they’re as balanced as hashing allows:

doveadm director flush backend-01

doveadm director flush backend-03

doveadm director flush backend-04

doveadm director flush backend-05