[Nss-mysql-users] Re: [Nss-mysql-devel] (no subject)

[Guillaume Morin]

Hi Larry,

I've posted to -users, since this is more an users question.

Dans un message du 22 avr à 22:10, Larry Weidig écrivait :
>       We are attempting to use the NSS MySQL program for authentication on
> our servers.  The data has been loaded and we can login using the
> information.  However, something is not quite right.  A simple perl program
> like the one below takes 20+ seconds on a PIII-550 system with 768MB RAM to
> read a 7000 user table that is located on a server which is very idle.
> $| = 1;
> while(my $username = getpwent()) {
>   print $username;
> }

There could be a lot of explanations for this. First, getpwent is very
slow. You should check if there is an index corresponding to your where
clause. You might want to tweak your MySQL server too (check show status
like 'Key%' and the related documentation to see if your server behaves
well).

Anyway, this may not be a problem since (afaik) no apps relies on this
function.  getgrent uses the same schema but we cache its resut.
Getpwent is not a good candidate for a benchmark :-)

Do you really need good performance for that function or have you used
this script just for testing/benchmarking purposes ?


>       The syslog receives one of these messages for each of the users in
> the database:
> 
> nss-mysql[7725]: _nss_mysql_strtol: string is empty.
> nss-mysql[7725]: seten(0)t: Can't convert count  into integer. Reverting to
> 1. Check select unixAccount.uid from unixAccount where unixAccount.server
> IN('*ALL','myserver') order by unixAccount.uid statement.

hmm, this is a bug. It is harmless, this warning is just wrong.
nss-mysql will behave correctly.
I've fixed it in the CVS. Loic, could you confirm the fix is okay ?

You might want to get the CVS version...

Reagards,

-- 
Guillaume Morin <address@hidden>

                              I wish you were here