[jira] [Commented] (HADOOP-13767) Aliyun Connection broken when idle then 1 minutes or build than 3 hours

["Steve Loughran (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HADOOP-13767?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15615177#comment-15615177 ] 

Steve Loughran commented on HADOOP-13767:
-----------------------------------------

* Is there a stack trace here? 
* what was the operation? GET, PUT?

If you look at s3a we treat a failure on a read as something to retry, just re-open as if it was a seek(). Same for a PUT, where, although the AWS library does some retries, it doesn't redo them all. I went for a hard coded retry policy class for backoff retry (see {{S3ABlockOutputStream}}:

{code}
RetryPolicy retryPolicy =
      RetryPolicies.retryUpToMaximumCountWithProportionalSleep(
          5,
          2000,
          TimeUnit.MILLISECONDS);
{code}

Hard code to avoid more config options to document and test, while setting things up to make it configurable in future.

Consider also adding a counter of disconnect/reconnects to metrics, as it could be a sign of connectivity issues. And when fielding support calls "why is IO slow", an answer like "your network is unreliable" is a good answer

> Aliyun Connection broken when idle then 1 minutes or build than 3 hours
> -----------------------------------------------------------------------
>
>                 Key: HADOOP-13767
>                 URL: https://issues.apache.org/jira/browse/HADOOP-13767
>             Project: Hadoop Common
>          Issue Type: Sub-task
>          Components: fs
>    Affects Versions: 3.0.0-alpha2
>            Reporter: Genmao Yu
>            Assignee: Genmao Yu
>             Fix For: 3.0.0-alpha2
>
>




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

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