Reasons:
- Riak DB development is almost halted after Basho
- riak-erlang-client is abandoned and doesn't work
correctly with OTP22
- Riak is slow in comparison to other databases
- Missing key ordering makes it impossible to implement range
queries efficiently (e.g. MAM queries)
The commit introduces the following API incompatibilities:
In ejabberd_auth.erl:
* dirty_get_registered_users/0 is renamed to get_users/0
* get_vh_registered_users/1 is renamed to get_users/1
* get_vh_registered_users/2 is renamed to get_users/2
* get_vh_registered_users_number/1 is renamed to count_users/1
* get_vh_registered_users_number/2 is renamed to count_users/2
In ejabberd_auth callbacks
* plain_password_required/0 is replaced by plain_password_required/1
where the argument is a virtual host
* store_type/0 is replaced by store_type/1 where the argument is
a virtual host
* set_password/3 is now an optional callback
* remove_user/3 callback is no longer needed
* remove_user/2 now should return `ok | {error, atom()}`
* is_user_exists/2 now must only be implemented for backends
with `external` store type
* check_password/6 is no longer needed
* check_password/4 now must only be implemented for backends
with `external` store type
* try_register/3 is now an optional callback and should return
`ok | {error, atom()}`
* dirty_get_registered_users/0 is no longer needed
* get_vh_registered_users/1 is no longer needed
* get_vh_registered_users/2 is renamed to get_users/2
* get_vh_registered_users_number/1 is no longer needed
* get_vh_registered_users_number/2 is renamed to count_users/2
* get_password_s/2 is no longer needed
* get_password/2 now must only be implemented for backends with
`plain` or `scram` store type
Additionally, the commit introduces two new callbacks:
* use_cache/1 where the argument is a virtual host
* cache_nodes/1 where the argument is a virtual host
New options are also introduced: `auth_use_cache`, `auth_cache_missed`,
`auth_cache_life_time` and `auth_cache_size`.
It is now possible for client connections to login using PKIX certificates.
This is disabled by default, to enable it:
- either set 'tls_verify: true' and 'cafile: /path/to/CAfile'
in the corresponding listener's section
- or set equivalent per-vhost options 'c2s_tls_verify' and 'c2s_cafile'
We need to be able to run only a few test groups, even if we do not have all
database backends installed and configured locally.
ejabberd test suite configures a specific host per backend. I changed ejabberd
to allow ignoring some hosts from config file on start, by providing the exact
list of hosts we want to start.
This is done by setting an ejabberd app Erlang environment variable 'hosts' and
passing the list of hosts we want to actually define.
When doing so, the backend specific hosts defined in ejabberd test configuration file
are simply ignored. As a result, we do not try to connect to unavailable backends.
I linked that part to CT run test by defining the hosts list based on environment variable
CT_BACKENDS. This variable is expected to be a comma separated list of available backends.
When Erlang Common Tests are run with that environment variable set, only the host matching
the name of the backend will be set, plus the default "localhost", common to many tests.
This can be combined with rebar ct groups list.
Example commands to run tests:
CT_BACKENDS=riak,mnesia rebar ct suites=ejabberd
CT_BACKENDS=mnesia rebar ct suites=ejabberd groups=mnesia
