hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Samarth Jain (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-11583) Parent spans not initialized to NullScope for every DFSPacket
Date Mon, 03 Apr 2017 18:45:41 GMT

    [ https://issues.apache.org/jira/browse/HDFS-11583?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15953984#comment-15953984
] 

Samarth Jain commented on HDFS-11583:
-------------------------------------

[~iwasakims]

What are your thoughts on what should be the trace id here:

{code}
if (parents.length > 0) {
                scope = Trace.startSpan("dataStreamer", new TraceInfo(0, parents[0]));
                // TODO: use setParents API once it's available from HTrace 3.2
//                scope = Trace.startSpan("dataStreamer", Sampler.ALWAYS);
//                scope.getSpan().setParents(parents);
              }

{code}

Generally, traceId establishes the correlation between all related spans. If there is already
an existing span in place, shouldn't we be using it's traceId instead? Probably calls for
a different JIRA I think.

> Parent spans not initialized to NullScope for every DFSPacket
> -------------------------------------------------------------
>
>                 Key: HDFS-11583
>                 URL: https://issues.apache.org/jira/browse/HDFS-11583
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: tracing
>    Affects Versions: 2.7.1
>            Reporter: Karan Mehta
>            Assignee: Masatake Iwasaki
>         Attachments: HDFS-11583-branch-2.7.001.patch, HDFS-11583-branch-2.7.002.patch,
HDFS-11583-branch-2.7.003.patch
>
>
> The issue was found while working with PHOENIX-3752.
> Each packet received by the {{run()}} method of {{DataStreamer}} class, uses the {{parents}}
field of the {{DFSPacket}} to create a new {{dataStreamer}} span, which in turn creates a
{{writeTo}} span as its child span. The parents field is initialized when the packet is added
to the {{dataQueue}} and the value is initialized from the {{ThreadLocal}}. This is how HTrace
handles spans. 
> A {{TraceScope}} is created and initialized to {{NullScope}} before the loop which runs
till the point when the stream is closed. 
> Consider the following scenario, when the {{dataQueue}} contains multiple packets, only
the first of which has a tracing enabled. The scope is initialized to the {{dataStreamer}}
scope and a {{writeTo}} span is created as its child, which gets closed once the packet is
sent out to a remote datanode. Before {{writeTo}} span is started, the {{dataStreamer}} scope
is detached. So calling the close method on it doesn't do anything at the end of loop. 
> The second iteration will be using the stale value of the {{scope}} variable with a DFSPacket
on which tracing is not enabled. This results in generation of an orphan {{writeTo}} spans
which are being delivered to the {{SpanReceiver}} as registered in the TraceFramework. This
may result in unlimited number of spans being generated and sent out to receiver. 



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

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


Mime
View raw message