[
https://issues.apache.org/jira/browse/KUDU-2666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16753652#comment-16753652
]
wkhapy123 commented on KUDU-2666:
-
but even nodelocal read cost still slow than read parquet from hdfs
>
[
https://issues.apache.org/jira/browse/KUDU-2666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16750879#comment-16750879
]
wkhapy123 commented on KUDU-2666:
-
look at the 3.png rpc_address is ip
> kudu spark intergration taskRead
[
https://issues.apache.org/jira/browse/KUDU-2666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16750876#comment-16750876
]
wkhapy123 commented on KUDU-2666:
-
reason maybe find
TaskSchedulerImpl class
[
https://issues.apache.org/jira/browse/KUDU-2666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16750664#comment-16750664
]
wkhapy123 commented on KUDU-2666:
-
I will debug spark TaskSetManager hope to find why
> kudu spark
[
https://issues.apache.org/jira/browse/KUDU-2666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16750645#comment-16750645
]
wkhapy123 commented on KUDU-2666:
-
yes,kudu cluster and spark executor in same nodes 。and i change
[
https://issues.apache.org/jira/browse/KUDU-2666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16750240#comment-16750240
]
Grant Henke commented on KUDU-2666:
---
I looked at recent spark runs I have done and have seen NODE_LOCAL
