inode leak 2.2.12+ why??

inode leak 2.2.12+ why??

[technews]

Hi,

The most puzzling thing is happeneing.  I have compiled a vanillat 2.2.18
kernel with scsi aic7xxx compiled in, 3com network support. (nothing fancy
no sound, no isdn, video, etc...)

I installed this kernel on a redhat 5.2 system, it boots in fine, but then
after some time I get messages saying fork: resource unavailable.  I
checked the inodes and I was chocked to see the number goingg up and up
without stopping.  I changed the value of inode-max to a high number, and
at the same time file-max, but to no avail.  Once it hits that max, it
will not allow me to do anything but shutdown the power to restart.

The server is a production email server, it uses redhat 5.2, sendmail, and
cyrus imap.  The filesystem is ext2 (i tried reiserfs, but I stopped since
I though the leaks are associated with resierfs, but I was wrong)  I tried
compiling the kernel with modules enabled, and also disabled.  Nothing
changes.  The server is single PIII 600E Mhz with 768Mb RAM, and 192G
external raid-5 box connecting via Adaptec UW2 SCSI.

I tried kernel 2.2.12 as well with similar problems, I tried the kernels
with other dual procesesor machines and the same thing happens.  The only
difference is that the inodes do not grow as fast, and when they arrive at
the set inode-max they stop there and they do not go in an endless loop.
I finally had to keep the 2.0.36 kernel on it, it seems very stable no
inode problems there..

Any ideas?? I have read the mailing lists, and people talk about it, and
assosictae it to nfs or squid, and we are not using either one.  ANy help
will be appretiated..

Best regards,
adonis

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

Re: inode leak 2.2.12+ why??

[Andrea Arcangeli]

On Thu, Jan 11, 2001 at 02:01:58PM -0500, technews@egsx.com wrote:
> Hi,
> 
> The most puzzling thing is happeneing.  I have compiled a vanillat 2.2.18
> kernel with scsi aic7xxx compiled in, 3com network support. (nothing fancy
> no sound, no isdn, video, etc...)
> 
> I installed this kernel on a redhat 5.2 system, it boots in fine, but then
> after some time I get messages saying fork: resource unavailable.  I
> checked the inodes and I was chocked to see the number goingg up and up
> without stopping.  I changed the value of inode-max to a high number, and
> at the same time file-max, but to no avail.  Once it hits that max, it
> will not allow me to do anything but shutdown the power to restart.

Yes, this is a known limitation of the icache (and 2.4.x still can't shrink the
filp pool, infact I'm surprised such pool is still there, I'd kill the filp
poll enterely and I'd use the slab cache for that that should be more efficient
in SMP and it's dynamic, should be just a few liner plus some code removal).

Your mistake is been to enlarge the inode-max, you shouldn't do that.
2.2.19pre7aa1 uses smarter calculations to setup the inode-max and icache hash
size (so you're more likely to get sane values for servers), I suggest to try
it without touching inode-max from its default value and see what happens.

Andrea
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

Re: inode leak 2.2.12+ why??

[Andrea Arcangeli]

On Thu, Jan 11, 2001 at 08:16:27PM +0100, Andrea Arcangeli wrote:
> On Thu, Jan 11, 2001 at 02:01:58PM -0500, technews@egsx.com wrote:
> > Hi,
> > 
> > The most puzzling thing is happeneing.  I have compiled a vanillat 2.2.18
> > kernel with scsi aic7xxx compiled in, 3com network support. (nothing fancy
> > no sound, no isdn, video, etc...)
> > 
> > I installed this kernel on a redhat 5.2 system, it boots in fine, but then
> > after some time I get messages saying fork: resource unavailable.  I
> > checked the inodes and I was chocked to see the number goingg up and up
> > without stopping.  I changed the value of inode-max to a high number, and
> > at the same time file-max, but to no avail.  Once it hits that max, it
> > will not allow me to do anything but shutdown the power to restart.
> 
> Yes, this is a known limitation of the icache (and 2.4.x still can't shrink the
> filp pool, infact I'm surprised such pool is still there, I'd kill the filp
> poll enterely and I'd use the slab cache for that that should be more efficient
> in SMP and it's dynamic, should be just a few liner plus some code removal).
> 
> Your mistake is been to enlarge the inode-max, you shouldn't do that.
> 2.2.19pre7aa1 uses smarter calculations to setup the inode-max and icache hash
> size (so you're more likely to get sane values for servers), I suggest to try
> it without touching inode-max from its default value and see what happens.

BTW, earlier 2.2.12 really had inode-leak bugs, but those are fixed in the
latest 2.2.1x (so make sure `uname -r` says 2.2.18 ;).

Andrea
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/