hadoop-common-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Apache Wiki <wikidi...@apache.org>
Subject [Hadoop Wiki] Update of "InvalidJiraIssues" by SteveLoughran
Date Thu, 24 Dec 2015 10:34:51 GMT
Dear Wiki user,

You have subscribed to a wiki page or wiki category on "Hadoop Wiki" for change notification.

The "InvalidJiraIssues" page has been changed by SteveLoughran:

more on stack traces

   * [[http://www.slideshare.net/steve_l/2014-0110-reporting-a-bug |Accompanying slides]]
  Please look at the video and understand why your JIRA was closed with a reference to this
page. Then follow some of the suggestions below to help debug your cluster.
+ == Don't assume a stack trace means a Hadoop bug ==
+ It's easy to see a stack trace with Hadoop classes at the top and think "Oh No! A Hadoop
bug!" and rush to file a JIRA. Don't rush to do this, it won't address your problem in a timely
manner. Even if it is a bug, you won't get a fix in your hand for months unless you are prepared
to build and run your own Hadoop release. And if you are prepared to do that, you need to
be able to track down stack traces yourself.
+ A stack trace with Hadoop classes in it can be caused for lots of reasons, primarily being
"something happened while hadoop was trying to do some work". All the stack trace shows is
``where the code was when the failure occurred``. If it's a network problem: you see a Hadoop
stack trace. If it's a Kerberos problem: you see a stack trace. If it's a classpath problem
trying to load a Hadoop-compatible filesystem —again, a stack trace.
+ Look at the stack and try to understand it. If there are nested stack traces, it's the one
at the bottom which counts. The other ones show the convoluted steps to get there. Be advised:
sometimes that bottom stack is being relayed from a server over IPC calls -it's a sign of
something happening remotely, rather than in the application doing the reporting.
  == Read and Understand the Logs ==
@@ -75, +83 @@

  == Keep your version of Hadoop current ==
- Finally: the development and testing goes on Hadoop 2.6+, with some maintenance of branch
1, with Hadoop 1.2.1 being the latest (as of December 2013). If you have a problem with an
older version of Hadoop: upgrade. If you aren't prepared to upgrade, you can't expect any
help at all.
+ Finally: the development and testing goes on Hadoop 2.7+, with important bug fixes backported
to branch 2.6. If you have a problem with an older version of Hadoop: upgrade. If you aren't
prepared to upgrade, you can't expect any help at all.
  Returning to JIRA, it may seem unfair for the developers not to care about your "critical"
issue and close it as invalid, despite the fact they are clearly the experts in Hadoop internals.
However they -we- are busy trying to build the future of Hadoop, ''the operating system for
data''. Most of the people working on this are being paid to do so, either from companies
whose business is built around selling supported Hadoop-based products, or from people who
use in production internally. None of these people have the time to help you -because if they
did help everyone with a problem, they'd never get anything done.
@@ -86, +94 @@

   1. Use a recent release of Hadoop. Older versions will have old bugs.
+  1. Look at any stack traces and try and understand them.
   1. Read the error message and try to understand what it means.
   1. Search on the web for the error message -and see what others did when they encountered
   1. Ask on the Hadoop User lists and any vendor-specific forums or other support options
they offer.
@@ -99, +108 @@

   1. File JIRA issues on problems you have seen on outdated versions of Hadoop -update and
try to replicate first.
   1. File JIRA issues on problems you have with Apache Hadoop based products provided by
third parties, unless these products are actually using the apache artifacts. Try to replicate
on the ASF versions first.
   1. Ask questions about using Hadoop on the developer lists. You will be deliberately ignored.
+  1. Send emails direct to developers. That's like emailing Linus because your Linux laptop
doesn't power on.
  That's why your JIRA issue was closed. It's not that the developers don't care that you
can't get Hadoop to work -it's that they aren't the right people to ask.

View raw message