[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Rpc.ldmd processes don't die right away in 5.1.2beta1... (fwd)

This archive contains answers to questions sent to Unidata support through mid-2025. Note that the archive is no longer being updated. We provide the archive for reference; many of the answers presented here remain technically correct, even if somewhat outdated. For the most up-to-date information on the use of NSF Unidata software and data services, please consult the Software Documentation first.

  • Subject: Re: Rpc.ldmd processes don't die right away in 5.1.2beta1... (fwd)
  • Date: Wed, 9 Aug 2000 09:13:18 -0600 (MDT)

===============================================================================
Robb Kambic                                Unidata Program Center
Software Engineer III                      Univ. Corp for Atmospheric Research
address@hidden             WWW: http://www.unidata.ucar.edu/
===============================================================================

---------- Forwarded message ----------
Date: Tue, 08 Aug 2000 15:57:14 -0400
From: James D. Marco <address@hidden>
To: Gilbert Sebenste <address@hidden>
Subject: Re: Rpc.ldmd processes don't die right away in 5.1.2beta1...

Hi ho,
        One quick note on this.  Unix internal disk structures do not actually
allow a file with an open filehandle to be deleted.  Soooo, a product queue
that is deleted while an update is in progress is not a fatal "file not found"
error.  As soon as the active process releases the link, the system will do
the actual deletion.
        The consequences to the LDM stuff can be a profound loss of disk space
if it is started and stopped too quickly, or, an active LDM process is
killed in a critical region, sometimes resulting in a zombie process. (I 
haven't actually tried to intentionally damage it like this.) If you try to 
make a new queue, it may report a shortage of disk space even though the 
system reports enough room on the partition. If you have a large queue,
wait at least 10-15 seconds before starting up again to avoid this. If 
your system is busy, wait longer.
        Caution dictates a system shutdown to fix this if it reports a 
zombie in the system process table (though, I've managed to successfully 
kill and recover from some Linux zombies.)
                                                        jdm  

  
  
At 10:14 AM 8/8/00 -0500, Gilbert Sebenste wrote:
>Hello Dave,
>
>> Gilbert,
>> 
>> This is typical (and correct) behavior under Linux.  I believe that
>> these processes are finishing I/O before exiting, which is a good
>> thing for the integrity of your product queue file.
>> 
>> Dave Raymond
>
>I guess I hadn't noticed it before like this...however, I did stop the LDM
>while it was still "catching up" from a re-start where I had deleted the
>queue, so I guess it just must have been backlogged.
>
>With the new LDM and the pqsurf.conf entries suggested, I am now noticing
>a significant performance improvement on my machine. Load average is
>typically just over "1", now it runs at .8 or a little less. If there's
>more activity on the radar and lots of people logging in, then it
>skyrockets, but it's still below what I've seen it to be in the past.
>
>***************************************************************************
****
>Gilbert Sebenste                                                     ********
>Internet: address@hidden    (My opinions only!)                     ******
>Staff Meteorologist, Northern Illinois University                      ****
>E-mail: address@hidden                                 ***
>web: http://weather.admin.niu.edu                                      **
>Work phone: 815-753-5492                                                *
>***************************************************************************
****
>
>