hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Arpit Agarwal (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-11789) Maintain Short-Circuit Read Statistics
Date Thu, 01 Jun 2017 22:07:04 GMT

    [ https://issues.apache.org/jira/browse/HDFS-11789?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16033796#comment-16033796

Arpit Agarwal commented on HDFS-11789:

Thanks for contributing this improvement [~hanishakoneru]. A few comments:

# For maintaining stats one more option is RollingAverages which uses MutableRatesWithAggregation
and is optimized for multithreaded updates.
# Instead of floating point arithmetic here:
sampleRangeMax = (int) ((double) conf.getScrMetricsSamplingPercentage()
        / 100 * Integer.MAX_VALUE);
sampleRangeMax = (Integer.MAX_VALUE / 100) * conf.getScrMetricsSamplingPercentage();
Also we should limit getScrMetricsSamplingPercentage to \[0, 100\] if the administrator misconfigures
# We should add isolated test cases for BlockReaderIoProvider and BlockReaderLocalMetrics
if possible.

> Maintain Short-Circuit Read Statistics
> --------------------------------------
>                 Key: HDFS-11789
>                 URL: https://issues.apache.org/jira/browse/HDFS-11789
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>            Reporter: Hanisha Koneru
>            Assignee: Hanisha Koneru
>         Attachments: HDFS-11789.001.patch
> If a disk or controller hardware is faulty then short-circuit read requests can stall
indefinitely while reading from the file descriptor. Currently there is no way to detect when
short-circuit read requests are slow or blocked. 
> This Jira proposes that each BlockReaderLocal maintain read statistics while it is active
by measuring the time taken for a pre-determined fraction of read requests. These per-reader
stats can be aggregated into global stats when the reader is closed. The aggregate statistics
can be exposed via JMX.

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