Can you paste the output of cassandra compactionstats? What you’re describing should not happen. There’s a check that drops sstables out of a compaction task if there isn’t enough available disk space, see https://issues.apache.org/jira/browse/CASSANDRA-12979 <https://issues.apache.org/jira/browse/CASSANDRA-12979> for some details.
> On Oct 13, 2017, at 4:24 PM, Bruce Tietjen <bruce.tiet...@imatsolutions.com> > wrote: > > > We are new to Cassandra and have built a test cluster and loaded some data > into the cluster. > > We are seeing compaction behavior that seems to violate what we read about > it's behavior. > > Our cluster is configured with JBOD with 3 3.6T disks. Those disks currently > respectively have the following used/available space: > Disk Used Available > sdb1 1.8T 1.7T > sdc1 1.8T 1.6T > sdd1 1.5T 2.0T > > nodetool compactionstats -H reports that the compaction system is attempting > to do a compaction that has a total of 6.83T > > The system hasn't had that much free space since sometime after we started > loading data and there has never been that much free space on a single disk, > so why would it ever attempt such a compaction? > > What have we done wrong, or am I reading this wrong? > > We have seen the same behavior on most of our 8 nodes. > > Can anyone tell us what is happening or what we have done wrong? > > Thanks
