This is really atypical. What about nodetool compactionstats?
crontab jobs in each node like nodetool repair, etc? Also security these 2 nodes have the same ports open? Same configuration, same JVM params? nodetool ring normal? Cheers. On 23-05-2017 20:11, Andrew Jorgensen wrote: > Hello, > > I will preface this and say that all of the nodes have been running > for about the same amount of time and were not restarted before > running nodetool tpstats. > > This is more for my understanding that anything else but I have a 20 > node cassandra cluster running cassandra 3.0.3. I have 0 read and 0 > writes going to the cluster and I see a strange banding of load > average for some of the nodes. > > 18 out of 20 of the nodes sit around an 15 min LA of 0.3 while 2 nodes > are at 0.08. Once I applied writes to the cluster all of the load > averages increased by the same proportion, so 18 out of 20 nodes > increased to ~0.4 and 2 nodes increased to 0.1 > > When I look at what is different between these nodes, all of which > have been running for the same amount of time, the only numerical > difference in nodetool tpstats is the InternalResponseStage. 18 out of > 20 of the nodes are in the range of 20,000 completed while 2 are only > at 300. Interestingly it also looks like the 2 nodes that are on the > low side are exhibiting symptoms > of https://issues.apache.org/jira/browse/CASSANDRA-11090. > > I restarted one of the low nodes and it immediately jumped up to match > the other 18 nodes in the cluster and settled around 0.4 15 minute LA. > > I am curious of two things, first why the InternalResponseStage is so > low on two of the nodes in the cluster and what that means. Second is > it atypical for the other 18 nodes in the cluster to have such a high > number of InternalResponseStage completed tasks or do these numbers > seems reasonable for a idle cluster? > > Thanks! > Andrew Jorgensen > @ajorgensen