hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Haohui Mai (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HADOOP-10468) TestMetricsSystemImpl.testMultiThreadedPublish fails intermediately
Date Mon, 14 Jul 2014 18:54:05 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-10468?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14061022#comment-14061022
] 

Haohui Mai commented on HADOOP-10468:
-------------------------------------

Looking at the code for a while to refresh my memory. Here is a quick recap:

# Metric2 implements hierarchical configuration with {{SubsetConfiguration}}. The configuration
key has the format {{foo.bar.spam}}, where  each parts of the configuration is a hierarchy.
For example, {{spam}} is a child of {{bar}}, and {{bar}} is a children of {{foo}. Metric2
turns some parts of the hierarchy keys in lowercase: (MetricsConfig:87)

{code}
  MetricsConfig(Configuration c, String prefix) {
    super(c, prefix.toLowerCase(Locale.US), ".");
  }
{code}

All keys of immediate children will be stored in lowercase strings (MetricConfig:151)

{code}
  Map<String, MetricsConfig> getInstanceConfigs(String type) {
    Map<String, MetricsConfig> map = Maps.newHashMap();
    MetricsConfig sub = subset(type);

    for (String key : sub.keys()) {
      Matcher matcher = INSTANCE_REGEX.matcher(key);
      if (matcher.matches()) {
        String instance = matcher.group(1);
        if (!map.containsKey(instance)) {
          map.put(instance, sub.subset(instance));
        }
      }
    }
    return map;
  }
{code}

# To look up the value in the hierarchical configuration, {{MetricConfig}} first finds the
key in itself and then reconstructs the full key for its parent by concatenating the prefix,
the delimiter and the key itself (SubsetConfiguration:88): 

{code}
    protected String getParentKey(String key)
    {
        if ("".equals(key) || key == null)
        {
            return prefix;
        }
        else
        {
            return delimiter == null ? prefix + key : prefix + delimiter + key;
        }
    }
{code}

In this test case, the reconstructed key is {{test.sink.collector.queue.capacity}} instead
of {{test.sink.Collector.queue.capacity}}, which leads to the failure.

> TestMetricsSystemImpl.testMultiThreadedPublish fails intermediately
> -------------------------------------------------------------------
>
>                 Key: HADOOP-10468
>                 URL: https://issues.apache.org/jira/browse/HADOOP-10468
>             Project: Hadoop Common
>          Issue Type: Bug
>    Affects Versions: 2.5.0
>            Reporter: Haohui Mai
>            Assignee: Haohui Mai
>            Priority: Blocker
>         Attachments: HADOOP-10468.000.patch, HADOOP-10468.001.patch, HADOOP-10468.2.patch
>
>
> {{TestMetricsSystemImpl.testMultiThreadedPublish}} can fail intermediately due to the
insufficient size of the sink queue:
> {code}
> 2014-04-06 21:34:55,269 WARN  impl.MetricsSinkAdapter (MetricsSinkAdapter.java:putMetricsImmediate(107))
- Collector has a full queue and can't consume the given metrics.
> 2014-04-06 21:34:55,270 WARN  impl.MetricsSinkAdapter (MetricsSinkAdapter.java:putMetricsImmediate(107))
- Collector has a full queue and can't consume the given metrics.
> 2014-04-06 21:34:55,271 WARN  impl.MetricsSinkAdapter (MetricsSinkAdapter.java:putMetricsImmediate(107))
- Collector has a full queue and can't consume the given metrics.
> {code}
> The unit test should increase the default queue size to avoid intermediate failure.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message