Multiple clusters capability. The api interface has changed because all apis now need an atom identifying the connection.

[cibingeorge]

No description provided.

[silviucpp]

I think will be much better to provide the config as follow:

{clusters, [
    {erlcass_conn, [
        {keyspace, <<"load_test_erlcass">>},
        {contact_points, <<"127.0.0.1">>},
        {contact_points, <<"127.0.0.1">>},
        {latency_aware_routing, true},
        {token_aware_routing, true},
        {number_threads_io, 8},
        {queue_size_io, 128000},
        {core_connections_host, 5},
        {max_connections_host, 5},
        {tcp_nodelay, true},
        {tcp_keepalive, {true, 60}},
        {connect_timeout, 5000},
        {request_timeout, 20000},
        {retry_policy, {default, true}}
    ]}
]} 

Basically:

• remove name and put it as tuple {name, OptionList}
• remove cluster_options and add all those on the same level with the keyspace. (Most probably keyspace should be removed when you send the option list to the NIF).

[silviucpp]

Why not putting this struct into the .cc file ? In case we need the CassCluster* outside the nif_cass_cluster.cc we can make a function that takes the ERL_NIF_TERM and returns the CassCluster

[silviucpp]

Another thing that I don't like and I think we can change is the fact that for every single query we are doing that get_existing_table_name which does concatenations and then from binary to atom.

I think we can do an improvement here like:

• Merge the functionality of erlcass_stm_cache and erlcass_stm_sessions into one single ETS table which will have the name of the cluster.

I mean when you add a key just put a tuple of : {stm_sessions, Identifier} or {stm_cache, Identifier}

What do you think ?

Silviu

[cibingeorge]

• Merge the functionality of erlcass_stm_cache and erlcass_stm_sessions into one single ETS table which will have the name of the cluster.

@silviucpp I thought about this. If we merge both tables, the behavior is a bit weird when the erlcass genserver crashes. Right now init creates a new cluster and session and creates a prepared statement for every cached statement in erlcass_stm_cache. Since erlcass_stm_sessions is a child of the erlcass, that dies as well. A solution could be to remember the cluster and session in the new ets table and use it if it exists when init-ing erlcass.

The only alternate solution i can think of is this binary to existing atom code.