hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Kihwal Lee (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-7217) Better batching of IBRs
Date Thu, 09 Oct 2014 17:03:35 GMT

    [ https://issues.apache.org/jira/browse/HDFS-7217?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14165372#comment-14165372

Kihwal Lee commented on HDFS-7217:

bq.  If possible, it would be nice for a test to ensure the receiving block IBRs are batched
although it's probably very difficult and may warrant a separate jira.
I've manually verified it.  Timing and batching aside, the correctness is covered by TestPipelinesFailover.
 E.g. if a receiving IBR is queued and delayed like deleted blocks or simply get dropped,
this test case fails.

> Better batching of IBRs
> -----------------------
>                 Key: HDFS-7217
>                 URL: https://issues.apache.org/jira/browse/HDFS-7217
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>            Reporter: Kihwal Lee
>         Attachments: HDFS-7217.patch
> After HDFS-2691 (pipeline recovery in HA), the number of IBR(incremental block report)s
have doubled.  Since processing of IBR requires exclusive FSNamesystem write lock, this can
be a source of significant overhead on clusters with high write load.  
> On one of the busy clusters, we have observed 60 to 70 percent of available handlers
being constantly occupied by IBRs. This degrades throughput greatly when compared to 0.23.

This message was sent by Atlassian JIRA

View raw message