Just wanted to update you guys that we turned on DEBUG level logging on the decommissioned node and the node receiving the decommissioned node's range. We did this by editing <cassandra-home>/conf/log4j-server.properties and changing the log4j.rootLogger to DEBUG. 

We ran decommission again and saw the that the receiving node was running out of disk space. The 184GB file was not able to fully stream to the receiving node.

We simply added more disk space to the receiving node and then decommission ran successfully.

Thanks for your help Aaron and also thanks for all those Cassandra articles on your blog. We found them helpful.

- Sameer
Accenture Technology Labs

On Thu, May 5, 2011 at 3:59 AM, aaron morton <aaron@thelastpickle.com> wrote:
Yes that was what I was trying to say. 

Aaron Morton
Freelance Cassandra Developer

On 5 May 2011, at 18:52, Tyler Hobbs wrote:

On Thu, May 5, 2011 at 1:21 AM, Peter Schuller <peter.schuller@infidyne.com> wrote:
> It's no longer recommended to run nodetool compact regularly as it can mean
> that some tombstones do not get to be purged for a very long time.

I think this is a mis-typing; it used to be that major compactions
were necessary to remove tombstones, but this is no longer the case in
0.7 so that the need for major compactions is significantly lessened
or even eliminated. However, running major compactions won't cause
tombstones *not* to be removed; it's just not required *in order* for
them to be removed.

I think he was suggesting that any tombstones *left* in the large sstable generated by the major compaction won't be removed for a long time because that sstable itself will not participate in any minor compactions for a long time.  (In general, rows in that sstable will not be merged for a long time.)

Tyler Hobbs
Software Engineer, DataStax
Maintainer of the pycassa Cassandra Python client library