Martin Sköld Wrote:
-------------------------------------------------------
> How can you determine a bottleneck in one of the
> ndbd, has it high CPU load?
By monitoring the different loads during testing, from different IOs to CPU usage, RAM usage, etc.
> Also, is it configured to use more cores, but you
> only see it using one?
> That can be changed with manual configuration,
> see
>
https://dev.mysql.com/doc/refman/8.0/en/mysql-cluster-ndbd-definition.html#ndbparam-ndbmtd-threadconfig
> If you have AutomaticThreadConfig set then you
> should use available cores.
> Assuming of course you are using ndbmtd
>
https://dev.mysql.com/doc/refman/8.0/en/mysql-cluster-programs-ndbmtd.html
>
> Another thing, failing to connect to a mysql
> server usually has nothing to do
> with the load of the Ndb Cluster. If the MySQL
> server is heavily loaded due to
> certain queries (joins) that cannot be pushed to
> the Ndb nodes, then that could be a factor.
> You should be able to connect, but have
> transactions time out due to
> heavy load on the cluster. The failing to connect
> is probably something else.
Thanks, I changed the startup to ndbmtd and configured the threads on the control node and it work. I sorted the unable to connect to database error after realising that for every mysql node, users have to be added to each one manually so this is something that isn't synced. After this the amount of errors significantly reduced and i was able to load balance.
Right now i'm trying to find other bottlenecks in the system to scale the test setup as far as possible as i find issues creep up with high throughput over time. At least the database issues has been solved.