Yarn running at 95%+ but hive jobs are running real slow

[#1]

we have a 4 data node cluster with the following RAM configuration:

Master Node - 32 cores - 251.7 GB

data node - 32 cores - 125.71 GB

data node - 32 cores - 62.71 GB

data node - 32 cores - 62.71 GB

edge node - 4 cores -47.01 GB

The memory/cpu/load is marginal as you can see in the images.

However the yarn seems to be running at 95+% but the hive jobs take a longtime.
any suggestions on how to improve the performance?

[#2]

Hello @Cody kamat!
Could you run an explain in one of your queries and share with us the result?
btw, a couple of questions related your queries:

- Are you using TEZ? Or LLAP?
- Are you using mapJoin? Vertex enable?
- Is CBO enable as well?
- Your tables has partitions/buckets?
- Are you using any fileformat like (parquet, avro, orc)?
Thanks

[#3]

Hi @Vinicius Higa Murakami,

- We are running TEZ but no llap as we have large data sets to crunch and we observed that the queries would just hang.

- We are definitely using mapJoin, however I am not sure about Vertex Enable -- However if we look at the DAG details for TEZ queries, it does create multiple Vertices.

- We have turned on CBO and we do run stats on our tables

- We have a single partition on all our tables

- All our HIVE tables are stored in the Parquet fileformat

(Additionally- If you see our configuration, our edge node has only 4 cores and has 47GB RAM and also has the YARN client services. Could this be a possible bottleneck ?. Plz advise. )

Thanks for responding _/\_

[#4]

Hi @Vinicius Higa Murakami,

I am enclosing the explain plan for the long running query. We have turned CBO on, however the explain plan is not using CBO , not sure why?

explain-plan.txt

[#5]

@Vinicius Higa Murakami, any help bud?