An odd problem between Solaris client and Linux server - NFS

This is a discussion on An odd problem between Solaris client and Linux server - NFS ; While we've had Solaris and Linux mixed in our shop for a while, this is the first time I can recall making extensive use of a Linux server and a Solaris client (outside of an "appliance" type server that we ...

+ Reply to Thread
Results 1 to 2 of 2

Thread: An odd problem between Solaris client and Linux server

  1. An odd problem between Solaris client and Linux server


    While we've had Solaris and Linux mixed in our shop for a while,
    this is the first time I can recall making extensive use of a
    Linux server and a Solaris client (outside of an "appliance" type
    server that we eventually returned).

    This was almost an accident. My home directory happens to
    reside on a Redhat Enterprise 3 machine. I was testing some
    software to be installed on a Solaris 8 web server, so I was
    using a Solaris 8 machine for testing. It's running Apache,
    and the DocumentRoot and log files for the particular virtual
    server are under my home directory (mounted from that Linux
    machine).

    I've fairly verbose debugging going on in the mod_perl
    components I'm testing. But what I noticed is that,
    after a very small amount of use, the error log - and
    the access log! - cease getting new data. It's acting
    as if the file system had gone full...except of course
    this isn't the case.

    While the volume is in this state, I cannot write to the
    file from that client from any software, even:

    echo "123\n" >> error_log

    I can write to it from the server (and other NFS clients).
    And the writes do eventually appear on the Solaris client
    in question.

    If I leave the Apache processes up, this problem doesn't
    disappear. If I shut down Apache, the problem persists
    for a few minutes (from five to ten, typically), and then
    clears up.

    Unmounting and then remounting also clears it up.

    The mount looks like:

    /home/andrew-peanut on taco:/export/home2/andrew-peanut remote/read/write/setuid/publoc/proto=tcp/port=2049/dev=3f80059 on Thu Apr 21 17:18:59 2005

    The filesystem on the Linux side is ext3 with ACLs active.

    I've tried to recreate this w/o Apache. For example, I wrote
    a program that would fork() off N children which would all open
    and write to the same file. I'd thought this a good replica
    of what Apache was doing, but this program never caused the
    problem. There may be some buffering difference, and I plan to
    look further at http_log.c to see if I can spot a difference.

    I've also some other things yet to try: udp instead of tcp, a different
    Linux server, etc. But I'm at something of a loss regarding
    this problem. Any suggestions would be appreciated.

    Thanks...

    Andrew


  2. Re: An odd problem between Solaris client and Linux server

    Andrew Gideon wrote:

    > I've also some other things yet to try: udp instead of tcp, a different
    > Linux server, etc. But I'm at something of a loss regarding
    > this problem. Any suggestions would be appreciated.


    FWIW, udp didn't make a difference. However, using an NFS server running a
    recent Fedora kernel (2.6.11) did make a difference: the problem
    disappeared.

    So it is something particular to RHE3.

    - Andrew


+ Reply to Thread