Hello Doug,
On Sun, 2009-11-01 at 08:04 -0600, Doug Phillips wrote:
> I'm trying to set up a cluster and getting the 410 REDO logs overloaded message.
> I've tried various things, and seems like I'm not getting anywhere.
>
> I'm trying to load in a dump of approx. 2 GB of data to a cluster with disk storage.
>
> I've scoured the lists and the available information on the internet, and I'm sure
> I'm missing something here. My restore only gets ~200 lines before it dumps out with the
> error.
>
> I'm getting panicked here because I have a drop-dead date of mid next week to get
> this loaded up.
>
> The configuration is as follows:
> NDB 7.0.8a
>
> 2 data nodes, 2 sql nodes running on the same machines
> 6 GB of RAM on the servers.
Did you do a --reload on the mgm node and --initial start the data nodes
after changing NoOfFragmentLogFiles? If not this change probably hasn't
happened (and in the case of the data nodes could be very bad).
Basically your problem is you are hammering the disks and the REDO logs
are filling up too quickly. Your configuration requires quite a bit
more than 6GB of RAM just for the ndb nodes and the ndb nodes allocate
all the RAM they need for the life of the process at startup. So with
mysqld and other system usage it could well be swapping (very bad for
cluster).
Also unless your disks really can do 256MB/sec on top of swapping and
disk data I would reduce DiskCheckpointSpeed quite significantly.
I would recommend having checkpointing on separate disks to your disk
data if you are not doing this already so you have more I/O for this
task.
Finally importing the data at a slower rate and/or increasing
NoOfFragmentLogFiles/FragmentLogFileSize should help.
Kind Regards
--
Andrew Hutchings, MySQL Support Engineer, Americas
Sun Microsystems, United Kingdom
http://www.sun.com/mysql/
