On Wed, Dec 31, 2014 at 12:01 AM, Mikhail Strebkov <streb...@gmail.com> wrote:
> I set compaction_throughput_mb_per_sec to 0 and restarted Cassandra. > You can also set this online w/ nodetool, fyi. > It looks to me that compaction thread is busy doing something that > produces quite a lot of garbage for GC to collect. Compaction performance > is bounded by CPU, which is a surprise for me, I would expect disk IO to be > a bottleneck. > > If you have SSD-esque devices, it is common to be CPU bound on compaction. And yes, compaction creates huge amounts of heap garbage pressure. > I don't think increasing the heap size if going to help in this case, what > do you think? Should I try JMX call userDefinedCompaction? > > I don't really understand the scenario where unthrottling reduces GC pre-fail. It's somewhat unclear to me from the stats you pasted how effective the actual GCs are. In general, increasing heap is a workaround to having a working set that peaks into GC pre-fail when compaction or write traffic spikes garbage creation. How effective are compactions in this CF? "grep % /var/log/cassandra/system.log"? =Rob
