Shared unix socket between ve

atuin · Jul 31, 2009

Hi,

We want Isolated MySQL Installation inside OpenVZ Cluster by shared mysql socket
view http://sudhaker.com/linux-unix/openvz/isolated-mysql-installation-openvz-cluster.html

Unfortunately, it seems there is a problem with openvz kernels, preventing this to work.

when we try to connect to mysql on other ve
=> ERROR 2002 (HY000): Can't connect to local MySQL server through socket '/shared/mysql/mysql.sock' (111)

view patch
http://git.openvz.org/?p=linux-2.6.26-openvz;a=commit;h=89451f9b7cf93d3c71cdb201c624244eb49f1223
http://git.openvz.org/?p=linux-2.6.24-openvz;a=commit;h=8caa941fe7be3a18d8b88b8f38ac9497e5fb2fb2

can you include this on your kernel

Many thanks
cyril

dietmar · Aug 3, 2009

Cant you simply use TCP/IP instead of a shared socket?

atuin · Aug 3, 2009

Hi,

I guess using TCP networking would be a non-optimal workaround.

MySQL Manual says :

"A MySQL client on Unix can connect to the mysqld server in two different ways: By using a Unix socket file to connect through a file in the filesystem (default /tmp/mysql.sock), or by using TCP/IP, which connects through a port number. A Unix socket file connection is faster than TCP/IP, but can be used only when connecting to a server on the same computer. A Unix socket file is used if you don't specify a hostname or if you specify the special hostname localhost."

For mysql configuration ; skip networking or localhost only => no iptables or tcp wrapper rules

And just for fun !!

We recompile the kernel starting from your sources
We changed in the Makefile:

Code:

${KERNEL_SRC}/README: ${KERNEL_SRC}.org/README
        rm -rf ${KERNEL_SRC}
        cp -av ${KERNEL_SRC}.org ${KERNEL_SRC}
        ./applypatches.pl ${KERNEL_SRC}
        sed -i ${KERNEL_SRC}/Makefile -e 's/^EXTRAVERSION.*$$/EXTRAVERSION=${EXTRAVERSION}/'
        cd ${KERNEL_SRC}; patch -p 1 <../${BOOTSPLASH_PATCH}
        cd ${KERNEL_SRC}; patch -p 1 <../bridge-patch.diff
        touch $@

by:

Code:

${KERNEL_SRC}/README: ${KERNEL_SRC}.org/README
        chattr -i ${KERNEL_SRC}/net/unix/af_unix.c
        rm -rf ${KERNEL_SRC}
        cp -av ${KERNEL_SRC}.org ${KERNEL_SRC}
        ./applypatches.pl ${KERNEL_SRC}
        sed -i ${KERNEL_SRC}/Makefile -e 's/^EXTRAVERSION.*$$/EXTRAVERSION=${EXTRAVERSION}/'
        cd ${KERNEL_SRC}; patch -p 1 <../${BOOTSPLASH_PATCH}
        cd ${KERNEL_SRC}; patch -p 1 <../bridge-patch.diff
        cd ${KERNEL_SRC}; patch -p 1 <../../cross_ve_unixsocket.diff
        cd ${KERNEL_SRC}; chattr +i net/unix/af_unix.c
        patch -p 1 <../../kvm_mv_qemu.1.patch
        touch $@

We had to make the file immutable af_unix.c as it was amended during the compilation ???

All runs very well

Sorry fo my bad english ...
Cyril

dietmar · Aug 4, 2009

atuin said:
I guess using TCP networking would be a non-optimal workaround.

MySQL Manual says :

"A MySQL client on Unix can connect to the mysqld server in two different ways: By using a Unix socket file to connect through a file in the filesystem (default /tmp/mysql.sock), or by using TCP/IP, which connects through a port number. A Unix socket file connection is faster than TCP/IP, but can be used only when connecting to a server on the same computer. A Unix socket file is used if you don't specify a hostname or if you specify the special hostname localhost."

So why is it non-optimal? Is performance really that worse? The shared socket is non-optimal because you loose connection each time the VM restarts?

atuin · Aug 4, 2009

dietmar said:
The shared socket is non-optimal because you loose connection each time the VM restarts?

Yes but on a production server, you does not restart mysql every day ...

for performance ...

check cpu usage
http://pix.academ.org/img/2008/10/01/c9ac14fecd7353f978b75ace4de26f21

check cache perf between tcp/ip vs socket
http://www.mysqlperformanceblog.com/2006/08/09/cache-performance-comparison/

And with unix socket:
- no tcp overload
- no iptables or tcp wrapper rule

dietmar · Aug 4, 2009

atuin said:
check cache perf between tcp/ip vs socket
http://www.mysqlperformanceblog.com/2006/08/09/cache-performance-comparison/

so it is about 40% faster? - wow.

atuin said:
And with unix socket:
- no tcp overload
- no iptables or tcp wrapper rule

OK - i will try to include that in the next release.

atuin · Aug 4, 2009

Ok many thanks !

dietmar · Aug 17, 2009

new release is online - please test

atuin · Aug 18, 2009

Hi,

Dont run

We had to make the file immutable af_unix.c as it was amended during the compilation ???

I do not know why, but works very well with our kernel compiled with chattr command

We had in makefile :

Code:

${KERNEL_SRC}/README: ${KERNEL_SRC}.org/README
        [B]chattr -i ${KERNEL_SRC}/net/unix/af_unix.c[/B]
        rm -rf ${KERNEL_SRC}
        cp -av ${KERNEL_SRC}.org ${KERNEL_SRC}
        ./applypatches.pl ${KERNEL_SRC}
        sed -i ${KERNEL_SRC}/Makefile -e 's/^EXTRAVERSION.*$$/EXTRAVERSION=${EXTRAVERSION}/'
        cd ${KERNEL_SRC}; patch -p 1 <../${BOOTSPLASH_PATCH}
        cd ${KERNEL_SRC}; patch -p 1 <../bridge-patch.diff
        cd ${KERNEL_SRC}; patch -p 1 <../../cross_ve_unixsocket.diff
       [B] cd ${KERNEL_SRC}; chattr +i net/unix/af_unix.c[/B]
        touch $@

dietmar · Aug 21, 2009

atuin said:
We had to make the file immutable af_unix.c as it was amended during the compilation ???

The file is not changed during compilation here. but the resulting kernel prevents ubuntu guest from running.

see http://www.proxmox.com/forum/showthread.php?t=1931

so I will remove that patch now from the kernel.

- Dietmar

dietmar · Aug 21, 2009

atuin said:
We had to make the file immutable af_unix.c as it was amended during the compilation ???

That is quite strange - you should really find out what happens (the file is not changed here).

atuin · Aug 21, 2009

Hi,

I do not know why but without chattr command,
once the compilation is complete

diff linux-2.6.24-openvz.org/net/unix/af_unix.c linux-2.6.24-openvz/net/unix/af_unix.c

are completely different but the patch cross_ve_unixsocket.diff is not applied.

So I think the bug in http://www.proxmox.com/forum/showthread.php?t=1931 has nothing to do because the patch is not applied

dietmar · Aug 21, 2009

atuin said:
So I think the bug in http://www.proxmox.com/forum/showthread.php?t=1931 has nothing to do because the patch is not applied

So why does it work again if I remove that patch?

atuin · Aug 21, 2009

dietmar said:
So why does it work again if I remove that patch?

Patch with chattr command or without ?

dietmar · Aug 21, 2009

atuin said:
diff linux-2.6.24-openvz.org/net/unix/af_unix.c linux-2.6.24-openvz/net/unix/af_unix.c

are completely different but the patch cross_ve_unixsocket.diff is not applied.

You compare different things!

linux-2.6.24-openvz.org/net/unix/af_unix.c (original)
linux-2.6.24-openvz/net/unix/af_unix.c (patched files)

The patch is applied to
linux-2.6.24-openvz/net/unix/af_unix.c, and that file does not change.

- Dietmar

dietmar · Aug 21, 2009

atuin said:
Patch with chattr command or without ?

Why should I use chattrib - that is nonsense. You just compare the wrong files anyways.

atuin · Aug 21, 2009

dietmar said:
You compare different things!

linux-2.6.24-openvz.org/net/unix/af_unix.c (original)
linux-2.6.24-openvz/net/unix/af_unix.c (patched files)

The patch is applied to
linux-2.6.24-openvz/net/unix/af_unix.c, and that file does not change.

- Dietmar

Yes so why in linux-2.6.24-openvz/net/unix/af_unix.c file the patch
are not applied ? and the file is completely different from original
linux-2.6.24-openvz.org/net/unix/af_unix.c after compilation

dietmar · Aug 21, 2009

atuin said:
Yes so why in linux-2.6.24-openvz/net/unix/af_unix.c file the patch
are not applied ?

Why do you ask me? Here, everything works a expected.

atuin said:
and the file is completely different from original
linux-2.6.24-openvz.org/net/unix/af_unix.c after compilation

because we apply about 200 openvz patches (see makefile)

- Dietmar

dietmar · Aug 21, 2009

atuin said:

Maybe the path to the patch is wrong ("../../cross_ve_unixsocket.diff" should be "../cross_ve_unixsocket.diff")?

atuin · Aug 21, 2009

So, I hope i found the error

The patch file 0001-2.6.24-ovz002.patch
(PVE_SRC/linux-2.6.24-openvz.org/debian/binary-custom.d/openvz/patchset/)

add this lines
if (s->sk_net != net)
continue;

in fonction static struct sock *__unix_find_socket_byname ()

Can you test if the bug (http://www.proxmox.com/forum/showthread.php?t=1931) is solved with this new patch ?

Cyril

Shared unix socket between ve

Renowned Member

Proxmox Staff Member

Renowned Member

Attachments

Proxmox Staff Member

Renowned Member

Proxmox Staff Member

Renowned Member

Proxmox Staff Member

Renowned Member

Proxmox Staff Member

Proxmox Staff Member

Renowned Member

Attachments

Proxmox Staff Member

Renowned Member

Proxmox Staff Member

Proxmox Staff Member

Renowned Member

Proxmox Staff Member

Proxmox Staff Member

Renowned Member

Attachments