Brian Bockelman wrote:
On Dec 4, 2008, at 6:58 AM, Steve Loughran wrote:
Leeau wrote:
Dear,
I want to config a 4-site hadoop cluster. but failed. Who can help me to
know why? and how can i start it? thanks.
you need to learn to read stack traces
Ah, the common claim of the Java programmer to the Unix admin.
As the person at the end of the phone call when the message is
"Java.net.IOException "no route to host" ", I think it is the job of the
ops team to know that "No route to host" means DNS is playing up.
Whereas in fact they would stop at the word java and reach for the
developers, which would waste a lot of time before they would eventually
acknowledge the problem lay in the external DNS servers, not our code.
I believe the correct response is "you need to learn to write readable
logging messages".
I semi-agree. Semi because the ways an installation can fail is very
much greater than the ways the dev team can think of, and a lot of
those errors your code sees gets thrown up from below. when an
application tries to be clever and give you the likeliest error, it can
actually be misleading. As a case in point, when windows is playing up
the OS often says "memory is getting low" because in the days of 128MB
machines, low memory was the problem. Now with 3-4GB of RAM, the cause
is something else, yet the error messages are still
If you look at how Hadoop is evolving, the latest versions add things
like the actual port that is in use, because we were all fed up with
seeing in the logs that a port was in use, but not which one. Yet I
still think that everyone running a hadoop cluster should embrace java
stack traces. Not least because it lets you look into the code to see
what is playing up.
-steve
