Large number of mount request.

Large number of mount request.

[Latrell]

[-- Attachment #1.1: Type: text/plain, Size: 497 bytes --]

Hi, all:

I have a problem with the mount and umount packet.
My problem is when I have a large number of mount requests, some mount request packets are missing. Thus, when I "cd share1", I will get "fail to change directory" because share1 is not mounted. Fewer mount requests work normally. 
Another question, who sends the packets? I'm tracing source code, and found the mount and umount begain with handle_packet() function. But I don't know how it receive the packets.

Thanks,
Latrell.

[-- Attachment #1.2: Type: text/html, Size: 1193 bytes --]

[-- Attachment #2: Type: text/plain, Size: 140 bytes --]

_______________________________________________
autofs mailing list
autofs@linux.kernel.org
http://linux.kernel.org/mailman/listinfo/autofs

Re: Large number of mount request.

[Jeff Moyer]

==> Regarding [autofs] Large number of mount request.; "Latrell" <is85022@cis.nctu.edu.tw> adds:

is85022> Hi, all:

is85022> I have a problem with the mount and umount packet.  My problem is
is85022> when I have a large number of mount requests, some mount request
is85022> packets are missing. Thus, when I "cd share1", I will get "fail to
is85022> change directory" because share1 is not mounted. Fewer mount
is85022> requests work normally. Another question, who sends the packets? 
is85022> I'm tracing source code, and found the mount and umount begain
is85022> with handle_packet() function. But I don't know how it receive the
is85022> packets.

How many mounts are we talking about?  What leads you to believe that the
kernel isn't passing mount requests to the daemon?

Please take a look at the section entitled, "Filing bug reports" on my
people page:  http://people.redhat.com/jmoyer/.

Please provide the debug logs as detailed on that page and we will see if
we can't figure out what the problem is.

Thanks,

Jeff

Re: Large number of mount request.

[ramana]

--- Latrell <is85022@cis.nctu.edu.tw> wrote:

> Hi, all:
> 
> I have a problem with the mount and umount packet.
> My problem is when I have a large number of mount requests, some
> mount request packets are missing. Thus, when I "cd share1", I will
> get "fail to change directory" because share1 is not mounted. Fewer
> mount requests work normally. 

I am afraid, this is ENOENT bug.

Thanks in advance.

Regards
ramana

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

Re: Large number of mount request.

[Latrell]

Do you mean my ghost option didn't make it in time thus "cd share1" failed?
I got the following message whenever there's "Failed to change directory" 
error:

automount[2306]: aquire_lock: can't lock lock file interrupted: 
/var/lock/autofs
automount[2306]: failed to mount autofs path /tmp/users/shares/1034

automount[2306]: /ramdisk/mnt/var1/tmp/users/shares/1034: mount failed!

I also checked the ghost did create directory share1 under the home 
directory. Could the lock problem cause cd share1 fail?

Thanks for your comment,
Latrell.

----- Original Message ----- 
From: "ramana" <intraperson@yahoo.com>
To: "Latrell" <is85022@cis.nctu.edu.tw>; <autofs@linux.kernel.org>
Sent: Friday, June 23, 2006 7:27 PM
Subject: Re: [autofs] Large number of mount request.


> --- Latrell <is85022@cis.nctu.edu.tw> wrote:
>
>> Hi, all:
>>
>> I have a problem with the mount and umount packet.
>> My problem is when I have a large number of mount requests, some
>> mount request packets are missing. Thus, when I "cd share1", I will
>> get "fail to change directory" because share1 is not mounted. Fewer
>> mount requests work normally.
>
> I am afraid, this is ENOENT bug.
>
> Thanks in advance.
>
> Regards
> ramana
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
> 

Re: Large number of mount request.

[Ian Kent]

On Thu, 2006-07-06 at 11:52 +0800, Latrell wrote:
> Do you mean my ghost option didn't make it in time thus "cd share1" failed?
> I got the following message whenever there's "Failed to change directory" 
> error:
> 
> automount[2306]: aquire_lock: can't lock lock file interrupted: 
> /var/lock/autofs
> automount[2306]: failed to mount autofs path /tmp/users/shares/1034
> 
> automount[2306]: /ramdisk/mnt/var1/tmp/users/shares/1034: mount failed!
> 
> I also checked the ghost did create directory share1 under the home 
> directory. Could the lock problem cause cd share1 fail?
> 
> Thanks for your comment,

At least some information would be in order here!
autofs version?
kernel version?
what patches if any have you or the vendor applied to autofs?

> Latrell.
> 
> ----- Original Message ----- 
> From: "ramana" <intraperson@yahoo.com>
> To: "Latrell" <is85022@cis.nctu.edu.tw>; <autofs@linux.kernel.org>
> Sent: Friday, June 23, 2006 7:27 PM
> Subject: Re: [autofs] Large number of mount request.
> 
> 
> > --- Latrell <is85022@cis.nctu.edu.tw> wrote:
> >
> >> Hi, all:
> >>
> >> I have a problem with the mount and umount packet.
> >> My problem is when I have a large number of mount requests, some
> >> mount request packets are missing. Thus, when I "cd share1", I will
> >> get "fail to change directory" because share1 is not mounted. Fewer
> >> mount requests work normally.
> >
> > I am afraid, this is ENOENT bug.
> >
> > Thanks in advance.
> >
> > Regards
> > ramana
> >
> > __________________________________________________
> > Do You Yahoo!?
> > Tired of spam?  Yahoo! Mail has the best spam protection around
> > http://mail.yahoo.com
> >
> > __________________________________________________
> > Do You Yahoo!?
> > Tired of spam?  Yahoo! Mail has the best spam protection around
> > http://mail.yahoo.com
> > 
> 
> _______________________________________________
> autofs mailing list
> autofs@linux.kernel.org
> http://linux.kernel.org/mailman/listinfo/autofs

Re: Large number of mount request.

[Ian Kent]

On Thu, 2006-07-06 at 11:52 +0800, Latrell wrote:
> Do you mean my ghost option didn't make it in time thus "cd share1" failed?
> I got the following message whenever there's "Failed to change directory" 
> error:
> 
> automount[2306]: aquire_lock: can't lock lock file interrupted: 
> /var/lock/autofs

And what is sending the SIGQUIT, SIGTERM or SIGINT to autofs to cause it
to stop waiting and return a interrupted fail?

> automount[2306]: failed to mount autofs path /tmp/users/shares/1034
> 
> automount[2306]: /ramdisk/mnt/var1/tmp/users/shares/1034: mount failed!
> 
> I also checked the ghost did create directory share1 under the home 
> directory. Could the lock problem cause cd share1 fail?
> 
> Thanks for your comment,
> Latrell.
> 
> ----- Original Message ----- 
> From: "ramana" <intraperson@yahoo.com>
> To: "Latrell" <is85022@cis.nctu.edu.tw>; <autofs@linux.kernel.org>
> Sent: Friday, June 23, 2006 7:27 PM
> Subject: Re: [autofs] Large number of mount request.
> 
> 
> > --- Latrell <is85022@cis.nctu.edu.tw> wrote:
> >
> >> Hi, all:
> >>
> >> I have a problem with the mount and umount packet.
> >> My problem is when I have a large number of mount requests, some
> >> mount request packets are missing. Thus, when I "cd share1", I will
> >> get "fail to change directory" because share1 is not mounted. Fewer
> >> mount requests work normally.
> >
> > I am afraid, this is ENOENT bug.
> >
> > Thanks in advance.
> >
> > Regards
> > ramana
> >
> > __________________________________________________
> > Do You Yahoo!?
> > Tired of spam?  Yahoo! Mail has the best spam protection around
> > http://mail.yahoo.com
> >
> > __________________________________________________
> > Do You Yahoo!?
> > Tired of spam?  Yahoo! Mail has the best spam protection around
> > http://mail.yahoo.com
> > 
> 
> _______________________________________________
> autofs mailing list
> autofs@linux.kernel.org
> http://linux.kernel.org/mailman/listinfo/autofs

Re: Large number of mount request.

[Latrell]

There will be an automount daemon for 1 ftp session (used to mount shares 
under home directory).
I sent SIGTERM to automount when my FTP session close (because the 
corresponding automount daemod need to be killed).
In my scenario, there should be multi-sessions need to be closed. Thus, 
SIGTERM will not be unique.

My autofs version is 4.1.4 with patch autofs4-2.4.29-20050404.patch.
kernel is 2.4.31 (no autofs patch for 2.4.31)

Thanks, for your comment,
Latrell.
----- Original Message ----- 
From: "Ian Kent" <raven@themaw.net>
To: "Latrell" <is85022@cis.nctu.edu.tw>
Cc: <ramana@intraperson.com>; <autofs@linux.kernel.org>
Sent: Thursday, July 06, 2006 7:58 PM
Subject: Re: [autofs] Large number of mount request.


> On Thu, 2006-07-06 at 11:52 +0800, Latrell wrote:
>> Do you mean my ghost option didn't make it in time thus "cd share1" 
>> failed?
>> I got the following message whenever there's "Failed to change directory"
>> error:
>>
>> automount[2306]: aquire_lock: can't lock lock file interrupted:
>> /var/lock/autofs
>
> And what is sending the SIGQUIT, SIGTERM or SIGINT to autofs to cause it
> to stop waiting and return a interrupted fail?
>
>> automount[2306]: failed to mount autofs path /tmp/users/shares/1034
>>
>> automount[2306]: /ramdisk/mnt/var1/tmp/users/shares/1034: mount failed!
>>
>> I also checked the ghost did create directory share1 under the home
>> directory. Could the lock problem cause cd share1 fail?
>>
>> Thanks for your comment,
>> Latrell.
>>
>> ----- Original Message ----- 
>> From: "ramana" <intraperson@yahoo.com>
>> To: "Latrell" <is85022@cis.nctu.edu.tw>; <autofs@linux.kernel.org>
>> Sent: Friday, June 23, 2006 7:27 PM
>> Subject: Re: [autofs] Large number of mount request.
>>
>>
>> > --- Latrell <is85022@cis.nctu.edu.tw> wrote:
>> >
>> >> Hi, all:
>> >>
>> >> I have a problem with the mount and umount packet.
>> >> My problem is when I have a large number of mount requests, some
>> >> mount request packets are missing. Thus, when I "cd share1", I will
>> >> get "fail to change directory" because share1 is not mounted. Fewer
>> >> mount requests work normally.
>> >
>> > I am afraid, this is ENOENT bug.
>> >
>> > Thanks in advance.
>> >
>> > Regards
>> > ramana
>> >
>> > __________________________________________________
>> > Do You Yahoo!?
>> > Tired of spam?  Yahoo! Mail has the best spam protection around
>> > http://mail.yahoo.com
>> >
>> > __________________________________________________
>> > Do You Yahoo!?
>> > Tired of spam?  Yahoo! Mail has the best spam protection around
>> > http://mail.yahoo.com
>> >
>>
>> _______________________________________________
>> autofs mailing list
>> autofs@linux.kernel.org
>> http://linux.kernel.org/mailman/listinfo/autofs
> 

Re: Large number of mount request.

[Ian Kent]

On Fri, 7 Jul 2006, Latrell wrote:

> There will be an automount daemon for 1 ftp session (used to mount shares
> under home directory).
> I sent SIGTERM to automount when my FTP session close (because the
> corresponding automount daemod need to be killed).
> In my scenario, there should be multi-sessions need to be closed. Thus,
> SIGTERM will not be unique.

I'm a little confused.
What I was suggesting is that according to the log entry below automount 
received one of the above signals which caused the mount to fail. That was 
why I asked about where the signals might have come from. If you send a 
termination signal to version 4 and a mount is busy it usually won't exit 
but I would expect any mounts in progress to fail in this manner as well.

> 
> My autofs version is 4.1.4 with patch autofs4-2.4.29-20050404.patch.
> kernel is 2.4.31 (no autofs patch for 2.4.31)

You should consider appling the patches for 4.1.4 available on kernel.org.
At least apply autofs-4.1.4-locking-fix-1.patch.
However, the locking problem that this addresses results in lock timeout 
log entries not lock interrupted messages. The reason bieng that automount 
dosn't wiat for the lock.

> 
> Thanks, for your comment,
> Latrell.
> ----- Original Message ----- From: "Ian Kent" <raven@themaw.net>
> To: "Latrell" <is85022@cis.nctu.edu.tw>
> Cc: <ramana@intraperson.com>; <autofs@linux.kernel.org>
> Sent: Thursday, July 06, 2006 7:58 PM
> Subject: Re: [autofs] Large number of mount request.
> 
> 
> > On Thu, 2006-07-06 at 11:52 +0800, Latrell wrote:
> > > Do you mean my ghost option didn't make it in time thus "cd share1"
> > > failed?
> > > I got the following message whenever there's "Failed to change directory"
> > > error:
> > > 
> > > automount[2306]: aquire_lock: can't lock lock file interrupted:
> > > /var/lock/autofs
> > 
> > And what is sending the SIGQUIT, SIGTERM or SIGINT to autofs to cause it
> > to stop waiting and return a interrupted fail?
> > 
> > > automount[2306]: failed to mount autofs path /tmp/users/shares/1034
> > > 
> > > automount[2306]: /ramdisk/mnt/var1/tmp/users/shares/1034: mount failed!
> > > 
> > > I also checked the ghost did create directory share1 under the home
> > > directory. Could the lock problem cause cd share1 fail?
> > > 
> > > Thanks for your comment,
> > > Latrell.
> > > 
> > > ----- Original Message ----- From: "ramana" <intraperson@yahoo.com>
> > > To: "Latrell" <is85022@cis.nctu.edu.tw>; <autofs@linux.kernel.org>
> > > Sent: Friday, June 23, 2006 7:27 PM
> > > Subject: Re: [autofs] Large number of mount request.
> > > 
> > > 
> > > > --- Latrell <is85022@cis.nctu.edu.tw> wrote:
> > > >
> > > >> Hi, all:
> > > >>
> > > >> I have a problem with the mount and umount packet.
> > > >> My problem is when I have a large number of mount requests, some
> > > >> mount request packets are missing. Thus, when I "cd share1", I will
> > > >> get "fail to change directory" because share1 is not mounted. Fewer
> > > >> mount requests work normally.
> > > >
> > > > I am afraid, this is ENOENT bug.
> > > >
> > > > Thanks in advance.
> > > >
> > > > Regards
> > > > ramana
> > > >
> > > > __________________________________________________
> > > > Do You Yahoo!?
> > > > Tired of spam?  Yahoo! Mail has the best spam protection around
> > > > http://mail.yahoo.com
> > > >
> > > > __________________________________________________
> > > > Do You Yahoo!?
> > > > Tired of spam?  Yahoo! Mail has the best spam protection around
> > > > http://mail.yahoo.com
> > > >
> > > 
> > > _______________________________________________
> > > autofs mailing list
> > > autofs@linux.kernel.org
> > > http://linux.kernel.org/mailman/listinfo/autofs
> > 
> 
> _______________________________________________
> autofs mailing list
> autofs@linux.kernel.org
> http://linux.kernel.org/mailman/listinfo/autofs
> 

Re: Large number of mount request.

[Latrell]

Sorry for misunderstanding what you mean. I think I should describe my 
scenario in detail.
I use autofs to mount shares under user's home directory. It functions 
normally in normal cases.
However, in stress test,  the user continuously login->change directory 
(autofs mount in this time) -> logout (a SIGERM sent to the automount 
daemon) --> login.......
What I understand from your comment is the mounts failed because some signal 
(should be SIGTERM I sent) interruptted. Am I right?
It's possible that automount receive the SIGTERM but can't terminate in 
time! And the new session opened and issue a 'cd' command, and the mount 
interruptted. Do you think it's possible?
What I encounter is:
1 user continuous loop -> "aquire_lock: can't lock lock file interrupted" 
never happen.
3 user continuous loop  -> "aquire_lock: can't lock lock file interrupted" 
sometimes happen.
3 more user continuous loop -> "aquire_lock: can't lock lock file 
interrupted"  happened frequently

Thanks for your comment, I'm really appreciated with your help.
Latrell.

----- Original Message ----- 
From: "Ian Kent" <raven@themaw.net>
To: "Latrell" <is85022@cis.nctu.edu.tw>
Cc: <autofs@linux.kernel.org>; <ramana@intraperson.com>
Sent: Friday, July 07, 2006 5:03 PM
Subject: Re: [autofs] Large number of mount request.


> On Fri, 7 Jul 2006, Latrell wrote:
>
>> There will be an automount daemon for 1 ftp session (used to mount shares
>> under home directory).
>> I sent SIGTERM to automount when my FTP session close (because the
>> corresponding automount daemod need to be killed).
>> In my scenario, there should be multi-sessions need to be closed. Thus,
>> SIGTERM will not be unique.
>
> I'm a little confused.
> What I was suggesting is that according to the log entry below automount
> received one of the above signals which caused the mount to fail. That was
> why I asked about where the signals might have come from. If you send a
> termination signal to version 4 and a mount is busy it usually won't exit
> but I would expect any mounts in progress to fail in this manner as well.
>
>>
>> My autofs version is 4.1.4 with patch autofs4-2.4.29-20050404.patch.
>> kernel is 2.4.31 (no autofs patch for 2.4.31)
>
> You should consider appling the patches for 4.1.4 available on kernel.org.
> At least apply autofs-4.1.4-locking-fix-1.patch.
> However, the locking problem that this addresses results in lock timeout
> log entries not lock interrupted messages. The reason bieng that automount
> dosn't wiat for the lock.
>
>>
>> Thanks, for your comment,
>> Latrell.
>> ----- Original Message ----- From: "Ian Kent" <raven@themaw.net>
>> To: "Latrell" <is85022@cis.nctu.edu.tw>
>> Cc: <ramana@intraperson.com>; <autofs@linux.kernel.org>
>> Sent: Thursday, July 06, 2006 7:58 PM
>> Subject: Re: [autofs] Large number of mount request.
>>
>>
>> > On Thu, 2006-07-06 at 11:52 +0800, Latrell wrote:
>> > > Do you mean my ghost option didn't make it in time thus "cd share1"
>> > > failed?
>> > > I got the following message whenever there's "Failed to change 
>> > > directory"
>> > > error:
>> > >
>> > > automount[2306]: aquire_lock: can't lock lock file interrupted:
>> > > /var/lock/autofs
>> >
>> > And what is sending the SIGQUIT, SIGTERM or SIGINT to autofs to cause 
>> > it
>> > to stop waiting and return a interrupted fail?
>> >
>> > > automount[2306]: failed to mount autofs path /tmp/users/shares/1034
>> > >
>> > > automount[2306]: /ramdisk/mnt/var1/tmp/users/shares/1034: mount 
>> > > failed!
>> > >
>> > > I also checked the ghost did create directory share1 under the home
>> > > directory. Could the lock problem cause cd share1 fail?
>> > >
>> > > Thanks for your comment,
>> > > Latrell.
>> > >
>> > > ----- Original Message ----- From: "ramana" <intraperson@yahoo.com>
>> > > To: "Latrell" <is85022@cis.nctu.edu.tw>; <autofs@linux.kernel.org>
>> > > Sent: Friday, June 23, 2006 7:27 PM
>> > > Subject: Re: [autofs] Large number of mount request.
>> > >
>> > >
>> > > > --- Latrell <is85022@cis.nctu.edu.tw> wrote:
>> > > >
>> > > >> Hi, all:
>> > > >>
>> > > >> I have a problem with the mount and umount packet.
>> > > >> My problem is when I have a large number of mount requests, some
>> > > >> mount request packets are missing. Thus, when I "cd share1", I 
>> > > >> will
>> > > >> get "fail to change directory" because share1 is not mounted. 
>> > > >> Fewer
>> > > >> mount requests work normally.
>> > > >
>> > > > I am afraid, this is ENOENT bug.
>> > > >
>> > > > Thanks in advance.
>> > > >
>> > > > Regards
>> > > > ramana
>> > > >
>> > > > __________________________________________________
>> > > > Do You Yahoo!?
>> > > > Tired of spam?  Yahoo! Mail has the best spam protection around
>> > > > http://mail.yahoo.com
>> > > >
>> > > > __________________________________________________
>> > > > Do You Yahoo!?
>> > > > Tired of spam?  Yahoo! Mail has the best spam protection around
>> > > > http://mail.yahoo.com
>> > > >
>> > >
>> > > _______________________________________________
>> > > autofs mailing list
>> > > autofs@linux.kernel.org
>> > > http://linux.kernel.org/mailman/listinfo/autofs
>> >
>>
>> _______________________________________________
>> autofs mailing list
>> autofs@linux.kernel.org
>> http://linux.kernel.org/mailman/listinfo/autofs
>>
> 

Re: Large number of mount request.

[Jeff Moyer]

==> Regarding Re: [autofs] Large number of mount request.; "Latrell" <is85022@cis.nctu.edu.tw> adds:

is85022> Sorry for misunderstanding what you mean. I think I should describe my
is85022> scenario in detail.
is85022> I use autofs to mount shares under user's home directory. It functions
is85022> normally in normal cases.
is85022> However, in stress test,  the user continuously login->change directory
is85022> (autofs mount in this time) -> logout (a SIGERM sent to the automount
is85022> daemon) --> login.......

Why is a SIGTERM sent to automount when you logout?  It shouldn't be.

-Jeff

Re: Large number of mount request.

[Ian Kent]

On Fri, 2006-07-07 at 10:40 -0400, Jeff Moyer wrote:
> ==> Regarding Re: [autofs] Large number of mount request.; "Latrell" <is85022@cis.nctu.edu.tw> adds:
> 
> is85022> Sorry for misunderstanding what you mean. I think I should describe my
> is85022> scenario in detail.
> is85022> I use autofs to mount shares under user's home directory. It functions
> is85022> normally in normal cases.
> is85022> However, in stress test,  the user continuously login->change directory
> is85022> (autofs mount in this time) -> logout (a SIGERM sent to the automount
> is85022> daemon) --> login.......
> 
> Why is a SIGTERM sent to automount when you logout?  It shouldn't be.

I think Jeff has pretty much summed it up.

The SIGTERM terminates autofs and you then have unnecessary startup and
shutdown overhead giving a window where mounts will not work as there is
no daemon to manage them.

Don't forget that there was also a bug in the lock routines that cause
it not to wait sometimes. Apply, at least the locking patch to 4.1.4.

Ian

Re: Large number of mount request.

[ramana]

--- Ian Kent <raven@themaw.net> wrote:

> 
> I think Jeff has pretty much summed it up.
> 
> The SIGTERM terminates autofs and you then have unnecessary startup
> and
> shutdown overhead giving a window where mounts will not work as there
> is
> no daemon to manage them.
> 
> Don't forget that there was also a bug in the lock routines that
> cause
> it not to wait sometimes. Apply, at least the locking patch to 4.1.4.

I do not think so.

His explanation about what actually he is doing is not very clear.

Little more explanation on how he is doing strss test would help,
'rather than trying to include his own conclusions'

Thanks in advance.

Regards
ramana

--

Autodir

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

Re: Large number of mount request.

[Latrell]

Hi, All:

I use autofs to mount shares of users under home directory with. Each user 
login, a map file will be created and an automount daemon will be invoked 
with options pid-file, ghost, timeout and file.
When user logout, I need the automount daemon killed. Thus, I send SIGTERM 
to the automount daemon. It's probably there are several automount daemons 
existing in the same time.
I found a case if many users logout at the same time, many SIGTERMs will be 
sent to corresponding automount daemons. In such cirsumstance, some 
automount daemons will not be killed and will exist in system forever. Thus 
I check the pid under /proc and send SIGTERM again when the automonut daemon 
not killed.

Did I use autofs in the wrong way?
Ms stress test is to keep login(invoke automount), mount(cd share1),access, 
logout(umount, kill automount) for 4 users sessions.

Many thanks,
Latrell.

ps. I patched the lock patch, and the condition not changed.

----- Original Message ----- 
From: "ramana" <intraperson@yahoo.com>
To: "Ian Kent" <raven@themaw.net>; "Latrell" <is85022@cis.nctu.edu.tw>; 
"Jeff Moyer" <jmoyer@redhat.com>
Cc: <autofs@linux.kernel.org>
Sent: Saturday, July 08, 2006 1:53 PM
Subject: Re: [autofs] Large number of mount request.


>
>
> --- Ian Kent <raven@themaw.net> wrote:
>
>>
>> I think Jeff has pretty much summed it up.
>>
>> The SIGTERM terminates autofs and you then have unnecessary startup
>> and
>> shutdown overhead giving a window where mounts will not work as there
>> is
>> no daemon to manage them.
>>
>> Don't forget that there was also a bug in the lock routines that
>> cause
>> it not to wait sometimes. Apply, at least the locking patch to 4.1.4.
>
> I do not think so.
>
> His explanation about what actually he is doing is not very clear.
>
> Little more explanation on how he is doing strss test would help,
> 'rather than trying to include his own conclusions'
>
> Thanks in advance.
>
> Regards
> ramana
>
> --
>
> Autodir
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
> 

Re: Large number of mount request.

[Ian Kent]

On Mon, 2006-07-10 at 09:40 +0800, Latrell wrote:
> Hi, All:
> 
> I use autofs to mount shares of users under home directory with. Each user 
> login, a map file will be created and an automount daemon will be invoked 
> with options pid-file, ghost, timeout and file.
> When user logout, I need the automount daemon killed. Thus, I send SIGTERM 
> to the automount daemon. It's probably there are several automount daemons 
> existing in the same time.
> I found a case if many users logout at the same time, many SIGTERMs will be 
> sent to corresponding automount daemons. In such cirsumstance, some 
> automount daemons will not be killed and will exist in system forever. Thus 
> I check the pid under /proc and send SIGTERM again when the automonut daemon 
> not killed.
> 
> Did I use autofs in the wrong way?

Think so.
It sounds like you should have one autofs daemon instance managing the
mountpoint and let it mount and umount as needed.

You won't be able to do it this way with autofs version 5 if you upgrade
at some point.

> Ms stress test is to keep login(invoke automount), mount(cd share1),access, 
> logout(umount, kill automount) for 4 users sessions.
> 
> Many thanks,
> Latrell.
> 
> ps. I patched the lock patch, and the condition not changed.

Yes. Those signals will mess things up.

> 
> ----- Original Message ----- 
> From: "ramana" <intraperson@yahoo.com>
> To: "Ian Kent" <raven@themaw.net>; "Latrell" <is85022@cis.nctu.edu.tw>; 
> "Jeff Moyer" <jmoyer@redhat.com>
> Cc: <autofs@linux.kernel.org>
> Sent: Saturday, July 08, 2006 1:53 PM
> Subject: Re: [autofs] Large number of mount request.
> 
> 
> >
> >
> > --- Ian Kent <raven@themaw.net> wrote:
> >
> >>
> >> I think Jeff has pretty much summed it up.
> >>
> >> The SIGTERM terminates autofs and you then have unnecessary startup
> >> and
> >> shutdown overhead giving a window where mounts will not work as there
> >> is
> >> no daemon to manage them.
> >>
> >> Don't forget that there was also a bug in the lock routines that
> >> cause
> >> it not to wait sometimes. Apply, at least the locking patch to 4.1.4.
> >
> > I do not think so.
> >
> > His explanation about what actually he is doing is not very clear.
> >
> > Little more explanation on how he is doing strss test would help,
> > 'rather than trying to include his own conclusions'
> >
> > Thanks in advance.
> >
> > Regards
> > ramana
> >
> > --
> >
> > Autodir
> >
> > __________________________________________________
> > Do You Yahoo!?
> > Tired of spam?  Yahoo! Mail has the best spam protection around
> > http://mail.yahoo.com
> > 

Re: Large number of mount request.

[Latrell]

The original design of autofs is using one (or few) daemons to control mount 
and umount as needed.
It seems there can be mutiple automount daemons existing in system, but the 
lock file (/var/lock/autofs) could behave abnormally if the daemons invoke 
and terminate frequently.
If I need to use autofs in the way, do you have comments how to relieve the 
probelm?
Or how I use autofs is beyond the design concept and change it might have 
potential problem?
Or using one automount to control the mount and umount is the best way to 
solve my problem?

Many thanks,
Latrell.
----- Original Message ----- 
From: "Ian Kent" <raven@themaw.net>
To: "Latrell" <is85022@cis.nctu.edu.tw>
Cc: <ramana@intraperson.com>; "Jeff Moyer" <jmoyer@redhat.com>; 
<autofs@linux.kernel.org>
Sent: Monday, July 10, 2006 1:41 PM
Subject: Re: [autofs] Large number of mount request.


> On Mon, 2006-07-10 at 09:40 +0800, Latrell wrote:
>> Hi, All:
>>
>> I use autofs to mount shares of users under home directory with. Each 
>> user
>> login, a map file will be created and an automount daemon will be invoked
>> with options pid-file, ghost, timeout and file.
>> When user logout, I need the automount daemon killed. Thus, I send 
>> SIGTERM
>> to the automount daemon. It's probably there are several automount 
>> daemons
>> existing in the same time.
>> I found a case if many users logout at the same time, many SIGTERMs will 
>> be
>> sent to corresponding automount daemons. In such cirsumstance, some
>> automount daemons will not be killed and will exist in system forever. 
>> Thus
>> I check the pid under /proc and send SIGTERM again when the automonut 
>> daemon
>> not killed.
>>
>> Did I use autofs in the wrong way?
>
> Think so.
> It sounds like you should have one autofs daemon instance managing the
> mountpoint and let it mount and umount as needed.
>
> You won't be able to do it this way with autofs version 5 if you upgrade
> at some point.
>
>> Ms stress test is to keep login(invoke automount), mount(cd 
>> share1),access,
>> logout(umount, kill automount) for 4 users sessions.
>>
>> Many thanks,
>> Latrell.
>>
>> ps. I patched the lock patch, and the condition not changed.
>
> Yes. Those signals will mess things up.
>
>>
>> ----- Original Message ----- 
>> From: "ramana" <intraperson@yahoo.com>
>> To: "Ian Kent" <raven@themaw.net>; "Latrell" <is85022@cis.nctu.edu.tw>;
>> "Jeff Moyer" <jmoyer@redhat.com>
>> Cc: <autofs@linux.kernel.org>
>> Sent: Saturday, July 08, 2006 1:53 PM
>> Subject: Re: [autofs] Large number of mount request.
>>
>>
>> >
>> >
>> > --- Ian Kent <raven@themaw.net> wrote:
>> >
>> >>
>> >> I think Jeff has pretty much summed it up.
>> >>
>> >> The SIGTERM terminates autofs and you then have unnecessary startup
>> >> and
>> >> shutdown overhead giving a window where mounts will not work as there
>> >> is
>> >> no daemon to manage them.
>> >>
>> >> Don't forget that there was also a bug in the lock routines that
>> >> cause
>> >> it not to wait sometimes. Apply, at least the locking patch to 4.1.4.
>> >
>> > I do not think so.
>> >
>> > His explanation about what actually he is doing is not very clear.
>> >
>> > Little more explanation on how he is doing strss test would help,
>> > 'rather than trying to include his own conclusions'
>> >
>> > Thanks in advance.
>> >
>> > Regards
>> > ramana
>> >
>> > --
>> >
>> > Autodir
>> >
>> > __________________________________________________
>> > Do You Yahoo!?
>> > Tired of spam?  Yahoo! Mail has the best spam protection around
>> > http://mail.yahoo.com
>> >
>