[OpenAFS-devel] token/pag cleanup on i386_linux*

[Neulinger, Nathan]

This message is in MIME format. Since your mail reader does not understand
this format, some or all of this message may not be legible.

------_=_NextPart_000_01C0F299.03EF5AF0
Content-Type: text/plain;
	charset="ISO-8859-1"

I've attached the script I use on linux... We run this every 15-30 minutes
(could run less frequently w/o any negative effect) and it does a excellent
job keeping the pags under control. Might not be perfect, but it beats
rebooting machine all the time.

-- Nathan

> -----Original Message-----
> From: David Thompson [mailto:thomas@cs.wisc.edu]
> Sent: Monday, June 11, 2001 12:02 PM
> To: Neulinger, Nathan
> Cc: 'openafs-devel@openafs.org'
> Subject: Re: [OpenAFS-devel] token/pag cleanup on i386_linux* 
> 
> 
> 
> Yeah I'd be interested.  We are pondering an 
> afs-authenticated web server for 
> some of our needs.  The "1 pag per second" rule and the performance 
> degredation under thousands of pags make recycling of pags 
> necessary, but I 
> didn't think you could scan the whole process table 
> effeciently enough to be 
> useful.  I'm thinking of a list of "known pags" that could be 
> allocated, used, 
> and later reused when all the processes using them are known 
> to be gone.  
> Anyone gotten further into this?
> 
> I would really like to see the GCPAGS code enabled if 
> possible for other 
> things, but it won't help us for this.  We need the *web 
> server* (not the afs 
> client) to verify that no processes are using a given pag anymore.
> 
> "Neulinger, Nathan" wrote:
> >On my heavy-authentication-activity clients, I have to run a 
> token cleanup
> >script that gets rid of tokens for processes that have gone 
> away/etc. It's
> >my understanding that the GCPAGS code does this 
> automatically. What is
> >preventing this from being enabled on linux?
> >
> >(If anyone wants the script, I'll be happy to send it to 
> you, it uses a
> >combination of processing kdump output and examining 
> auxgroup membershup
> >from /proc.)
> >
> >If you don't run this, and you have a server that does a lot of token
> >activity, you get into a situation where the afs_users 
> structure in the
> >kernel is HUGE and very very very slow, which impacts 
> everything on the
> >machine. I'm running scripts to handle this on hpux (poorly) 
> and linux. 
> >
> >-- Nathan
> >
> >------------------------------------------------------------
> >Nathan Neulinger                       EMail:  nneul@umr.edu
> >University of Missouri - Rolla         Phone: (573) 341-4841
> >Computing Services                       Fax: (573) 341-4216
> >_______________________________________________
> >OpenAFS-devel mailing list
> >OpenAFS-devel@openafs.org
> >https://lists.openafs.org/mailman/listinfo/openafs-devel
> 
> 


------_=_NextPart_000_01C0F299.03EF5AF0
Content-Type: application/octet-stream;
	name="clean-tokens-redhat.pl"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: attachment;
	filename="clean-tokens-redhat.pl"

#!/umr/bin/perl=0A=
$| =3D 1;=0A=
=0A=
# =0A=
# Input sources=0A=
#=0A=
chomp($ver =3D `uname -r`);=0A=
$in_kdump =3D "/usr/afsws/etc/kdump-$ver -users|";=0A=
$in_ps =3D "/bin/ps -ef|";=0A=
=0A=
#=0A=
# Read in the kdump data=0A=
#=0A=
open(IN, $in_kdump);=0A=
print "Processing KDump: ";=0A=
$count =3D 0;=0A=
%pags =3D ();=0A=
%pags_per_uid =3D ();=0A=
while ( $line =3D <IN> )=0A=
{=0A=
	next if ( $line !~ /:/o );=0A=
=0A=
	if ( $line =3D~ /uid=3Dx(41[0-9a-fA-F]+),.*vid=3D([0-9]+)/o )=0A=
	{=0A=
		$pag =3D $1;=0A=
		$uid =3D $2;=0A=
=0A=
		$pagnum =3D hex($pag) - hex("41000000");=0A=
		$pag2 =3D $pagnum % 16384;=0A=
		$pag1 =3D ($pagnum-$pag2)/16384;=0A=
=0A=
		$pag1 +=3D 33536;=0A=
		$pag2 +=3D 32512;=0A=
=0A=
		if ( $uid > 0 )=0A=
		{=0A=
			$pags{"$pag1:$pag2"} =3D $uid;=0A=
			print "\t$pag1,$pag2\n";=0A=
		}=0A=
=0A=
#		if ( $count % 250 =3D=3D 0 ) { print "." };=0A=
		$count++;=0A=
	}=0A=
}=0A=
print "$count active tokens.\n";=0A=
close(IN);=0A=
=0A=
#=0A=
# Read in the info about active processes, delete from=0A=
# %pags for each active=0A=
#=0A=
print "Scanning processes:\n";=0A=
opendir(DIR, "/proc");=0A=
while ( $file =3D readdir(DIR) )=0A=
{=0A=
	next if ( $file !~ /^[0-9]+$/o );=0A=
	print "\t$file: ";=0A=
=0A=
	$groups =3D "";=0A=
	=0A=
	open(IN, "/proc/$file/status");=0A=
	while ( chomp($line =3D <IN>) )=0A=
	{=0A=
		if ( $line =3D~ /^Groups:(.*)$/o )=0A=
		{=0A=
			$groups =3D $1;=0A=
			last;=0A=
		}=0A=
	}=0A=
	close(IN);=0A=
	=0A=
	$groups =3D~ s/[\r\n\t ]+/ /gio;=0A=
	$groups =3D~ s/^ +//gio;=0A=
	$groups =3D~ s/ +$//gio;=0A=
	@groups =3D split(' ', $groups);=0A=
=0A=
	$pag1 =3D $groups[0];=0A=
	$pag2 =3D $groups[1];=0A=
	=0A=
	if ( $pag1 >=3D 32512 && $pag2 >=3D 33536 )=0A=
	{=0A=
		print "$pag1,$pag2 ";=0A=
		if ( $pags{"$pag1:$pag2"} )=0A=
		{=0A=
			print "skipping pag\n";=0A=
			delete $pags{"$pag1:$pag2"};		=0A=
		}=0A=
		else=0A=
		{=0A=
			print "no active token\n";=0A=
		}=0A=
	}=0A=
	else=0A=
	{=0A=
		print "not active pag\n";=0A=
	}=0A=
}=0A=
=0A=
closedir(DIR);=0A=
=0A=
#=0A=
# Clean pags=0A=
#=0A=
$cleaned =3D 0;=0A=
$skipped =3D 0;=0A=
$count =3D 0;=0A=
while ( ($pag,$uid) =3D each(%pags) )=0A=
{=0A=
	if ( $count % 100 =3D=3D 0 )=0A=
	{=0A=
		print "Processed Tokens: $count\n";=0A=
	}=0A=
	$count++;=0A=
=0A=
	=0A=
	($pag1, $pag2) =3D split(/:/, $pag);=0A=
	$) =3D "0 $pag1 $pag2";=0A=
	system("/usr/afsws/bin/unlog");=0A=
	$cleaned++;=0A=
=0A=
	print "\t$pag unlogged\n";=0A=
}=0A=
print "Cleaning Tokens: $cleaned tokens unlogged.\n";=0A=

------_=_NextPart_000_01C0F299.03EF5AF0--