Please keep dbi-users on Cc. The NLS_LANG issue posted by Sven Miller looks relevant.

On 08.04.2010 22:30, Perl Diety wrote:

To give you some more detail about what works and what does- we've run Perl CGI on this server for years. We access a MYSQL database every day, and all of that works 100% fine. We access library functions, system libs, etc again with no issues.
So, DBI is ok, and most of Apache is also ok.
The lone problem is ORACLE 10 access. So I think this is a good indicator that this is specifically related to Oracle ENV vars, or something to do with the DBD or DBI.
Can't be DBI, or else MySQL access would also fail.

Most likely are DBD::Oracle (especially the compiled part), Oracle client libraries, environment variables used by Oracle, and filesystem permission issues.

Our system admn did update the DBI and DBD to the latest version
Shouldn't hurt -- but HOW did he update? RPM install or compile on that machine?

RPMs are precompiled, and thus need a very specific version of the Oracle client libraries -- those they were compiled against.

A compile on the machine should link against the Oracle client libraries that are installed and you want to use, so there should be no problem unless the machine is really f....ed up, full of different Oracle client versions, Perl versions, and DBI versions. Sure, you can install multiple versions of Oracle, Perl and DBI on the same machine, but in that case, you must know very exactly what you are doing, and it is very likely that you end with a broken installation.

and rebuilt the server.
Which server? Oracle? In that case, you need to re-compile DBD::Oracle after updating Oracle.


Also, the script accesses Oracle fine from a command line. Which sort of suggests that the DBD is OK?
Right. If the simple-and-stupid script from my last mail works on the command line, it must be an issue with filesystem permissions or environment variables.

As far as your step-by-step, I did that, and when I add in the CONNECT is when this error is thrown and it dies. I tried to get more info like fatalsToBrowser, inspecting $! and ORa Error codes etc.
Comment out all code added by the connect step and the following steps, and compare the values of $<, $>, $(, $), and %ENV for CGI mode and command line mode. With Data::Dumper, the Useqq and Sortkeys options are really useful.

$< and $> are user IDs, both values should be equal for a single run, but they should differ between CGI and command line mode. Typically, they are less than 100 (often 80) for CGI mode, and at least 1000 (100 on old systems) for command line mode.

$( and $) are space-separated lists of group IDs, the first value is the primary group ID, the following values are additional group IDs. Again, both values should be equal for a single run, and they are very likely to differ between CGI and command line mode. Especially the first number should be different.

Look up the IDs in the output of getent group (or the file /etc/group), and find out which groups are not available in CGI mode. Compare with the owning group(s) of the Oracle client library files and the Oracle configuration files, especially those files that aren't world readable (i.e. ls -l output showing -rwxr-x--- or -rw-r----- instead of -rwxr-xr-x or -rw-r--r--). If the non-world-readable files are owned by a group available in command line mode, but not in CGI mode, you have a permission problem. Either put the user running Apache into the relevant group(s), or change the filesystem permissions (preferably though the Oracle provided unharden script).

Compare the entire %ENV between CGI and command line mode. All variables available in command line mode, but not set or set to a different value in CGI mode, are suspect. Some obvious ones aren't relevant, like BASH and BASH_*, HIST*, PS1 to PS4, PWD, OLDPWD, SHELL, DISPLAY. <http://www.orafaq.com/faq/managing_unix_environment_variables> lists some variables that may be relevant. Try adding PassEnv or SetEnv for those variables to the Apache configuration. As a quick hack, you could instead set the environment variables in a BEGIN block *BEFORE* loading the Oracle client libraries, i.e.:

#!/usr/bin/perl -Tw

use strict;
BEGIN {
   # Hack to test Oracle setup
# Must execute BEFORE Oracle client libraries are loaded, i.e. before "use DBI", before any implicit or explicit "use DBD::Oracle".
   $ENV{'ORACLE_HOME'}='/some/where';
   $ENV{'NLS_LANG'}='AL32UTF8';
}
use DBI;
# ...
my $dbh=DBI->connect('dbi:Oracle:....


(I really don't know if that trick still works, but with Oracle 8 and 9, it worked fine, and I don't see why it shouldn't work with 10 or 11.)

Alexander



Im off to the doctor have a great evening A!


MP



On Thu, Apr 8, 2010 at 3:06 PM, Alexander Foken <[email protected] <mailto:[email protected]>> wrote:

    On 08.04.2010 19:09, Perl Diety wrote:

        I wrote the script to dump %ENV (in a MUCH more Perlish way
        than the apache version!) but yes, you're correct; I filtered
        out a lot of unrelated stuff to make the list more friendly.

    ok


        I don't think a linux apache server cares about *PATH* does
        it? Not sure.

    Apache itself shouldn't care, but the programs invoked by Apache
    do. If $ENV{PATH} is not set, programs either fall back to a (more
    or less) sane default, like /bin:/usr/bin, or simply assume it to
    be empty. In the first case, only external programs in /bin and
    /usr/bin are found, unless the absolute path is passed to the exec
    library functions (execvp, execlp, ...). In the latter case, no
    external program is found unless its absolute path is passed to
    the exec functions. That can make a huge difference for many
    programs that don't contain hardcoded paths.


        The message seemed to imply PATH was only used for WINDOWS
        servers?

    On Unix, $ENV{PATH} is only used to find executables. On Windows,
    it is also used to find DLLs required by executables. So yes, some
    tricks with $ENV{PATH} are only required on Windows, to make sure
    all DLLs are found. On Unix, you would use $ENV{LD_LIBRARY_PATH}
    or something like that, or entries in /etc/ld.so.conf.



        I'm *pretty sure* our ENV VARS for the LIBs and server
        correctly- because before we set them we were getting a
        DIFFERENT error (cant load Oracle.so). Now we're past that
        error onto the next one. So I think the ORACLE_HOME and LD
        paths are OK now.

    Hmmm, try beginning with a simple and stupid script, then start to
    extend it.

    I would start with something like

    #!/usr/bin/perl -w
    use strict;
    print "Content-Type: text/plain\r\n\r\nHello World\n";

    Then I would add "use CGI::Carp qw(fatalsToBrowser);", and invoke
    the script again to see if it has survived the script.

    Next, modify print to dump the entire %ENV after the "Hello World"
    line, as readable and as precise as possible. Data::Dumper could
    help. Try again.

    Next, add "use DBI;" and try again.

    Then, add "my
    $dbh=DBI->connect("dbi:Oracle:whereever","scott","tiger",{
    RaiseError => 1 });" before "print", and try again.

    Looking at your last postings, this step should fail. However, the
    next step would be to add a simple select-and-fetch, like this:

    my $sth=$dbh->prepare("select 42 as result from dual");
    $sth->execute();
    my $data=$sth->fetchrow_hashref();
    $sth->finish();
    print Dumper($data);

    If you get to this point, your problem should be gone. ;-)



        Its a preplexing problem with LOTS and LOTS of difffernt
        proposed fixes in the forums. I suspect that's because there
        are so many possible problems.


    Yes. I think Oracle has grown too much.


        As a programmer, I'm not dazzled by the error message. I would
        never present my users with a message like "it might be A, or
        possibly B, or something to do with C, or possibly SOMETHING
        ELSE"..

    Right, I also don't like message like "oops, something went
    wrong". At least, I would expect some extension like "... while
    attempting to read the configuration file". Perl makes
    do_something() or die "Couldn't ...: $!" very easy.


        Instead I would CHECK A, report an A error if found,. Check B,
        report a B error if found. And so on. Its unreasonable to
        expect a user to react to a message that basically says IT
        COULD BE ANYTHING!

        In this case, some users say check the locale, language
        settings, others say check LIB PATH.


    DBI connect(...) failed: ERROR OCIEnvNlsCreate. Check ORACLE_HOME
    (Linux)
    env var or PATH (Windows) and or NLS settings, permissions, etc. at
    /cgi-bin/oratest.cgi line 32


    Something in your %ENV is so broken that Oracle becomes confused.

    What happens when you try to run the CGI script (preferably the
    one I just "invented")  from your user account, simply by invoking
    it from a shell prompt? (When the CGI module is used, some
    additional tricks are needed, like passing Query parameters as
    command line parameters or $ENV{QUERY_STRING}.)

    When it runs without trouble, compare %ENV of the command line and
    %ENV from the script running as CGI, and watch for differences.
    Also look at the user and group IDs ($<, $>, $(, $) ).

    One "new" feature of Apache is the fact that it does not pass its
    original environment to its child processes, like it used to do in
    ancient versions. The newer apache versions only create a minimal
    environment, to which PassEnv and SetEnv add new environment
    variables. So, "PassEnv ORACLE_HOME" is often needed.


        Others say its a DBI or DBD / Oracle version mismatch.

    Did you compile DBI and DBD::Oracle on that machine or did you use
    some precompiled stuff (RPMs, some tar copied over from a
    different machine)?

    If you compiled it yourself and DID NOT update Oracle or Perl
    later, everything should be fine.

    Mismatching DBI and DBD::Oracle should cause linker errors, and
    the same is also true for a mismatch between the Oracle libraries
    and DBD::Oracle.

        SOme say its an Oracle 10 security issue and I need to run an
        Oracle script to un-harden it?

    From a very quick google search for OCIEnvNlsCreate, it seems that
    the Oracle client library has trouble finding its configuration
    files, like TNSNAMES.ORA, or some libraries. And that happens
    because it does not know ORACLE_HOME, or because it does not have
    sufficient privileges (file permissions -rwxr-x--- and other ugly
    stuff, this is the "hardening").

    Alexander


        I've encountered very few times in my 20 years as a programmer
        where ONE error could be caused by so many possibilities.
        After like 4 man-weeks, we're still fumbling with it.

        Thanks!
        MP




        On Wed, Apr 7, 2010 at 4:53 PM, Alexander Foken
        <[email protected] <mailto:[email protected]>
        <mailto:[email protected] <mailto:[email protected]>>> wrote:

           On 07.04.2010 10:45, Martin Evans wrote:

               Perl Diety wrote:
ENV VARS

                   DOCUMENT_ROOT = /var2/www/html
                   GATEWAY_INTERFACE = CGI/1.1
                   HTTP_ACCEPT = */*
                   HTTP_ACCEPT_ENCODING = gzip, deflate
                   HTTP_ACCEPT_LANGUAGE = en-us
                   HTTP_HOST = ournode.com <http://ournode.com>
        <http://ournode.com>

                   HTTP_UA_CPU = x86
                   HTTP_USER_AGENT = Mozilla/4.0
                   LD_LIBRARY_PATH = /export/apps/oracle/product/10201/lib
                   LD_RUN_PATH = /export/apps/oracle/product/10201/lib
                   NLS_LANG = en_GB.UTF-8
                   ORACLE_BASE = /export/apps/oracle
                   ORACLE_HOME = /export/apps/oracle/product/10201
                   ORA_NLS10 = /export/apps/oracle/product/10201/nls/data/
                   REQUEST_METHOD = GET
                   REQUEST_URI = /cgi-bin/oratest.cgi
                   SERVER_PORT = 80
                   SERVER_PROTOCOL = HTTP/1.1
                   SERVER_SIGNATURE =

                   Apache/2.0.52 (Red Hat) Server at ournode.com
        <http://ournode.com>
                   <http://ournode.com> Port 80


                   SERVER_SOFTWARE = Apache/2.0.52 (Red Hat)
                   DBI connect(...) failed: ERROR OCIEnvNlsCreate. Check
                   ORACLE_HOME (Linux)
                   env var or PATH (Windows) and or NLS settings,
                   permissions, etc. at
                   /cgi-bin/oratest.cgi line 32
           This looks roughly like a CGI enviromnent, and the Oracle
           Variables seem to be set. But lots of variables seem to be
           missing, like PATH and some variabes starting with HTTP_,
        SERVER_,
           SCRIPT_, and REQUEST_. And the LD_xxx variables shouldn't
        be there.

           If this is the *complete* environment provided to the CGI,
           something is *very* wrong with the Apache.

           If not, post a complete dump of %ENV in CGI context, e.g.
        with the
           printenv script that came with Apache:

           #!/usr/local/bin/perl
           ##
           ##  printenv -- demo CGI program which just prints its
        environment
           ##

           print "Content-type: text/plain; charset=iso-8859-1\n\n";
           foreach $var (sort(keys(%ENV))) {
             $val = $ENV{$var};
             $val =~ s|\n|\\n|g;
             $val =~ s|"|\\"|g;
             print "${var}=\"${val}\"\n";

           }

               Did you really set AND export ORACLE_HOME or
        LD_LIBRARY_PATH
               such that
               Apache children see them? I can't remember what the
        syntax in the
               httpd.conf file is now (SetEnv perhaps) as I no longer use
               Apache but
               you used to have to explicitly tell Apache which env
        vars to
               allow.

               Martin
Two ways:

           1. PassEnv VariableName [...] -- see
           <http://httpd.apache.org/docs/2.2/mod/mod_env.html#passenv>
           2. SetEnv VariableName Value -- see
           <http://httpd.apache.org/docs/2.2/mod/mod_env.html#setenv>

           I prefer PassEnv over SetEnv when the variables already
        exist in
           the environment of the process invoking the Apache server. That
           way, I don't have to change the Apache configuration when an
           environment variable changes. I use SetEnv only to set
        additional
           variables that must not appear in the environment of the
        process
           invoking the Apache server.

           See also
<http://alexander-foken.de/Censored%20copy%20of%20Oracle%20Troubleshooter%20HOWTO.html>

           Alexander

           --     Alexander Foken
           mailto:[email protected] <mailto:[email protected]>
        <mailto:[email protected] <mailto:[email protected]>>

            http://www.foken.de/alexander/




-- Alexander Foken
    mailto:[email protected] <mailto:[email protected]>
     http://www.foken.de/alexander/




--
Alexander Foken
mailto:[email protected]  http://www.foken.de/alexander/

Reply via email to