incubator-cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Hiller, Dean" <Dean.Hil...@nrel.gov>
Subject found bottleneck but can we do these steps?
Date Fri, 22 Feb 2013 20:56:50 GMT
So, it turns out we don't have enough I/o going on for our upgradesstables but it is really
hitting the upper bounds of memory(8G) and our cpu is pretty low as well.

At any rate, we are trying to remove a 2 gig bloomfilter on a columnfamily.  Can we do the
following

 1.  Disable thrift/gossip (per previous emails)
 2.  Restart the node?  (any way to restart it without reading in that bloomfilter to lesson
the memory……should I temporarily turn up the node without the key cache maybe)
 3.  Run nodetool upgradesstables databus5 nreldata;

1.When I restart the node, will gossip/thrift stay off ??? Or do I change the seeds, change
9160 to xxxx and I don't see where I can change 7199 to something?  (how to do this safely).
2.Hmmmm, is there any way to run upgradesstables when cassandra is not running AND crank up
the memory of nodetool to 8G or does nodetool always  just tell cassandra to do it????

  I feel like I have a chicken and egg problem here.  I want to clean up this bloomfilter
which requires upgradesstables(from what I read), but I need the bloomfilter to not be there
so I am not bottlenecked by the memory.  At this rate, I will have to do each node each day
for 6 days before I can recover (and I would prefer to speed it up just a little).

Thanks,
Dean

Mime
View raw message