hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Xiao Chen (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-12482) Provide a configuration to adjust the weight of EC recovery tasks to adjust the speed of recovery
Date Fri, 20 Oct 2017 02:54:01 GMT

    [ https://issues.apache.org/jira/browse/HDFS-12482?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16212087#comment-16212087

Xiao Chen commented on HDFS-12482:

Thanks [~eddyxu] for revving. Looks good in general, a few comments:
- let's add input validation for {{xmitWeight}}. Although {{Math.max}} will use 1, negative
values and 0 looks invalid. Any thoughts on whether we should have an upper bound limit?
- in the .md documentation, suggest to add an example so people can know what to set without
checking code. Maybe something like 'For example, if there are 2 read streams and 1 write
stream, xmits weight of 0.5 means recovery will be scheduled for 1 EC block and 1 replicated
block. xmits weight of 1.0 means xxx' (Or better words)
- Test looks great. Trivially suggest we bump the timeout to at least 3 minutes, since it
took 20 seconds on my local.

> Provide a configuration to adjust the weight of EC recovery tasks to adjust the speed
of recovery
> -------------------------------------------------------------------------------------------------
>                 Key: HDFS-12482
>                 URL: https://issues.apache.org/jira/browse/HDFS-12482
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>          Components: erasure-coding
>    Affects Versions: 3.0.0-alpha4
>            Reporter: Lei (Eddy) Xu
>            Assignee: Lei (Eddy) Xu
>            Priority: Minor
>              Labels: hdfs-ec-3.0-nice-to-have
>         Attachments: HDFS-12482.00.patch, HDFS-12482.01.patch
> The relative speed of EC recovery comparing to 3x replica recovery is a function of (EC
codec, number of sources, NIC speed, and CPU speed, and etc). 
> Currently the EC recovery has a fixed {{xmitsInProgress}} of {{max(# of sources, # of
targets)}} comparing to {{1}} for 3x replica recovery, and NN uses {{xmitsInProgress}} to
decide how much recovery tasks to schedule to the DataNode this we can add a coefficient for
user to tune the weight of EC recovery tasks.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: hdfs-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-help@hadoop.apache.org

View raw message