Hi all,
Still running into this issue. I have removed the [tcp default] section
that set "the 6M hard limit" for Send and Receive buffers and I have set
in the [ndbd default] section TotalSendBufferMemory = 256M and I'm still
getting "Error 1297 Got temporary error 1218 'Send Buffers overloaded in
NDB kernel' from NDBCLUSTER" when executing large numbers of queries
like this:
UPDATE sbtestndb set
c='947444285-1041194067-484405640-185870187-417051949-265839100-72785242
-363450386-566476655-628024932' where id=666518
and this:
UPDATE sbtestndb set k=k+1 where id=835636
I find it difficult to understand how 256M of buffer can be used up when
all 3 machines (2 ndbd, 1 mysqld) are on the same gigE switch.
What's the best way to diagnose where the real issue is?
-Matthew
> -----Original Message-----
> From: Andrew.Hutchings@stripped [mailto:Andrew.Hutchings@stripped]
> Sent: Wednesday, November 04, 2009 3:28 AM
> To: Boehm, Matthew
> Cc: Pekka.Nousiainen@stripped; cluster@stripped
> Subject: RE: Send Buffers overloaded in NDB kernel
>
> Hello Matthew,
>
> On Tue, 2009-11-03 at 10:03 -0600, Boehm, Matthew wrote:
> > A benchmarking application wouldn't produce good results if it had
to
> > "do again". Basically this shows that cluster cannot handle 256
> threads
> > of "update non-index" style SQL.
>
> Not quite, it shows your configuration cannot handle this. Cluster is
> not designed to perform well doing this but it can be done. I suggest
> increasing the 6M hard limit you have set on the Send and Receive
> buffers (or removing them completely so the dynamic
> TotalSendBufferMemory can be used). Especially if you are using
> ndbmtd.
>
> Kind Regards
> --
> Andrew Hutchings, MySQL Support Engineer, Americas
> Sun Microsystems, United Kingdom
> http://www.sun.com/mysql/
