Slow index performance after upgrade from 5.6.8 to 6.1.4

After upgrade, my cluster become very worst when ingest data.
With normal status (not ingest), everything ok(cpu, heap, io, network, load.. keep low)
But with some small ingest continously, cluster become poor perfomance. CPU and heap still keep in low, but load and iowait raise high and not release (iowait about 40, load avg 20, index speed less than 10k/s with 8 datanode cluster). Before upgrade, index speed about 200k/s.
I don't change any config or amount of data daily. So where I can look to deal with that?
Log in datanode have many line with 'starting segment upgrade/finished segment upgrade'. Is it hurt my index perfomance? And if yes, how to deal with that?
Thanks you.

Did you reindex your data from the original 5.6.8 index format to a new index created with ES 6.1.4? Running old version indexes in a newer ES version might be a problem. (I'm not sure on this, but I have to ask)

Yeah I already reindex all data to new format. But with some restart node, hot shards is not balances across my cluster, then some node are overworkload and I think it must swap current-process different shard continously. So, anyone has some heuristic/strategy/experience with manual/auto routing shards to deal with that?

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.