[ 
https://issues.apache.org/jira/browse/HDFS-10365?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15271433#comment-15271433
 ] 

Kihwal Lee commented on HDFS-10365:
-----------------------------------

bq. Are you suggesting to tune dfs.blockreport.split.threashold ...
Yes. Each FBR rpc will be smaller, so the impact of timeout-retransmit will be 
lower. Also NN will process individual report quicker.

NN creates lots of temporary objects and their life time can be longer than 
expected, especially when requests need to sit in the call queue for a long 
time. A larger CMS young gen has been shown to absorb them well and the young 
gen collection time does not become excessive as most objects get freed rather 
than copied. Of course it is a bit different during start-up when the data 
structures are being built up.  We thought a high rate of allocation and 
freeing will go well with what G1 is designed for, but so far we haven't found 
a magic recipe that works better than CMS for large namenodes.   I hear it does 
a better job for HBase region servers though.

> FullBlockReports retransmission delays NN startup time in large cluster.
> ------------------------------------------------------------------------
>
>                 Key: HDFS-10365
>                 URL: https://issues.apache.org/jira/browse/HDFS-10365
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: hdfs
>    Affects Versions: 2.6.0
>         Environment: version - hadoop-2.6.0 (hdp-2.2)
> DN - 1200 nodes
>            Reporter: Chackaravarthy
>            Priority: Critical
>
> Whenever NN is restarted, it takes huge time for NN to come back to stable 
> state. i.e. Last contact time remains more than 1 or 2 mins continuously for 
> around 3 to 4 hours. This is mainly because most of the DN's getting timeout 
> (60s) in blockReport (FBR) rpc call and then it keep sending FBR again.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to