The situation for which I hope to use MySQL cluster, we have 100+
webservers each with persistent connections to many separate databases,
one or more of which will hopefully become clusters. Currently, mysql is
running only as the database servers and we are using our own connection
pooling functions (tied to php) to maintain persistent connections from
each webserver to each database server.
Following the model that you're talking about here, if I understand it
correctly, for my situation I would need to have each webserver running
as an API node. Will the DB be able to handle connections from 200+ API
nodes? And will each API node be able to connect to multiple clusters?
Would this be done by having one NDB table on each API node for each
cluster that it needs to connect to?
Devananda van der Veen
(still trying to get it to compile on Mandrake, unsuccessfully)
Brancaleoni Matteo wrote:
>>Alternatively, assuming all my tables can be converted to NDB I could
>>make every single system that uses mysql into an API server, handling
>>requests only from the that server, which should mean that there is no
>>failover required for API nodes (and failover amongst the DB nodes is
>>handled automagically anyhow).
>>Does that make sense, or have I misunderstood the documentation?
>sure, that's the way it should work.
>for example, assuming you have 2 DB nodes with 4 web servers,
>let run mysqld server on each webserver, so you'll have 4 API
>connected to the DB nodes. so if one webserver dies,
>the others will go on without problems.
>In that way each client has its own api (mysqld) that
>simply interfaces a standard application (web in this case)
>to the cluster.