If I remember correctly there have been rare cases when the number of
open files of 40 is not enough. I don't remember exactly under what
circumstances but I seem to remember that it was fixed by setting the
limit to 60 instead.
Adam D wrote:
> 8dn, 2sql, 2replica, RHEL4 64bit, 5.0.13rc.
> The cluster I have currently was doing next to
> nothing. Merely running my backup queries on it
> over night and I come back after a 4 day break and
> find that one of the 8 datanodes was not
> connected.
>
> Error log shows this;
> Current byte-offset of file-pointer is: 468
>
>
> Date/Time: Tuesday 18 October 2005 - 03:25:40
> Type of error: error
> Message: Max number of open files exceeded
> Fault ID: 2806
> Problem data:
> Object of reference: Ndbfs::createAsyncFile
> ProgramName: ndbd
> ProcessID: 11704
> TraceFile:
> /data/mysql-cluster/ndbd/ndb_17_trace.log.1
> Version 5.0.13 (rc)
> ***EOM***
>
> I have the trace if any mysql guys are
> interested.
>
> Did some list searching and ulimit -n shows 1024
> file limit. Which is a fair amount I would think.
>
> I currently do not have the MaxNoOfOpenFiles
> parameter set in my mgm config. This seems to
> default to 40. What should a busy-loaded cluster
> be set to etc? Are their any drawbacks to this
> setting.
Mikael Ronstrom
Senior Software Architect, MySQL AB
My blog:
http://mikaelronstrom.blogspot.com