There are a couple of solutions for big number of hosts (above 1500 in my case). By default mnesia doesn't support such a big number of tables and crashes when being launched. It can be started with -env ERL_MAX_ETS_TABLES 20000 for example but being started this way is works unstable.
If try to use MySQL as a DB there will no other way then creating separate db and hold a connection to each host (also it can be used a tricky solution wich is disable periodical requests for connect).
Is there any solution of this problem? Can we push everything in RAM? For authorization module is our own and we doesn't need to store users and roters, cause everyone has shared @online@ roster.
If I understand right, when starting nodes with default config, mnesia will crash anyway due to a big number of tables.
P.S. And one more, how can we solve a problem with server reboot when adding host. It is not a good idea to kick off all users when adding new host.
ejabberd 3.0.0-alpha3 works
ejabberd 3.0.0-alpha3 works better than ejabberd 2 when there are many vhosts. It also allows to add vhosts to a running ejabberd, using WebAdmin.
But ejabberd 3.0.0-alpha3 is not yet recommended for a production server, because maybe the tables definitions change.