autofs-ldap-nfs is slow because of mtab

autofs-ldap-nfs is slow because of mtab

[Gokdeniz Karadag]

Hello,

In my setup there are three automount mount points. They are all ldap maps. The 
nfs shares are on the same nfs server. nfs client and automount is on a debian 
system, nfs server is a solaris one.

The problem is, trying to mount a nonexisting directory fails immediately  with 
the message, "No such file or directory" in two of the shares. But the third 
share waits for 10-30 seconds before giving the error. The maps have 1, 9 and 
19 items respectively.

I have run strace on the different automount processes. The quick ones show a 
query to ldap server, then a reasonable amount of mtab stats & reads, then they 
  finish. The process for the slow mount point queries ldap, gets mount map, 
then for about 100000 times; opens mtab file, reads contents and then closes it.
the trace file is full of these;
open("/etc/mtab", O_RDONLY)             = 7
fstat64(7, {st_mode=S_IFREG|0644, st_size=1733, ...}) = 0
mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 
0xa7afb000
read(7, "/dev/sda1 / ext3 rw,errors=remou"..., 4096) = 1733
close(7)


Any pointers to where should I look ?


-- 
Gokdeniz Karadag

Re: autofs-ldap-nfs is slow because of mtab

[jmoyer]

==> Regarding [autofs] autofs-ldap-nfs is slow because of mtab; Gokdeniz Karadag <gokdeniz@cclub.metu.edu.tr> adds:

gokdeniz> Hello, In my setup there are three automount mount points. They
gokdeniz> are all ldap maps. The nfs shares are on the same nfs server. nfs
gokdeniz> client and automount is on a debian system, nfs server is a
gokdeniz> solaris one.

gokdeniz> The problem is, trying to mount a nonexisting directory fails
gokdeniz> immediately with the message, "No such file or directory" in two
gokdeniz> of the shares. But the third share waits for 10-30 seconds before
gokdeniz> giving the error. The maps have 1, 9 and 19 items respectively.

gokdeniz> I have run strace on the different automount processes. The quick
gokdeniz> ones show a query to ldap server, then a reasonable amount of
gokdeniz> mtab stats & reads, then they finish. The process for the slow
gokdeniz> mount point queries ldap, gets mount map, then for about 100000
gokdeniz> times; opens mtab file, reads contents and then closes it.  the
gokdeniz> trace file is full of these; open("/etc/mtab", O_RDONLY) = 7
gokdeniz> fstat64(7, {st_mode=S_IFREG|0644, st_size=1733, ...}) = 0
gokdeniz> mmap2(NULL, 4096, PROT_READ|PROT_WRITE,
gokdeniz> MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xa7afb000 read(7, "/dev/sda1
gokdeniz> / ext3 rw,errors=remou"..., 4096) = 1733 close(7)


gokdeniz> Any pointers to where should I look ?

Well, first we need to start with a version.  ;)  What version of autofs
are you running?  Next, you will want to gather a debug log.  My people
page has instructions on how to do that:
  http://people.redhat.com/jmoyer

That *should* make things fairly obvious.  If it doesn't (or even if it
does), post a follow-up message here.

Thanks!

Jeff

Re: autofs-ldap-nfs is slow because of mtab

[Gokdeniz Karadag]

jmoyer@redhat.com demis ki::
> ==> Regarding [autofs] autofs-ldap-nfs is slow because of mtab; Gokdeniz Karadag <gokdeniz@cclub.metu.edu.tr> adds:
> 
> gokdeniz> Hello, In my setup there are three automount mount points. They
> gokdeniz> are all ldap maps. The nfs shares are on the same nfs server. nfs
> gokdeniz> client and automount is on a debian system, nfs server is a
> gokdeniz> solaris one.
> 
> gokdeniz> The problem is, trying to mount a nonexisting directory fails
> gokdeniz> immediately with the message, "No such file or directory" in two
> gokdeniz> of the shares. But the third share waits for 10-30 seconds before
> gokdeniz> giving the error. The maps have 1, 9 and 19 items respectively.
> 
> gokdeniz> I have run strace on the different automount processes. The quick
> gokdeniz> ones show a query to ldap server, then a reasonable amount of
> gokdeniz> mtab stats & reads, then they finish. The process for the slow
> gokdeniz> mount point queries ldap, gets mount map, then for about 100000
> gokdeniz> times; opens mtab file, reads contents and then closes it.  the
> gokdeniz> trace file is full of these; open("/etc/mtab", O_RDONLY) = 7
> gokdeniz> fstat64(7, {st_mode=S_IFREG|0644, st_size=1733, ...}) = 0
> gokdeniz> mmap2(NULL, 4096, PROT_READ|PROT_WRITE,
> gokdeniz> MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xa7afb000 read(7, "/dev/sda1
> gokdeniz> / ext3 rw,errors=remou"..., 4096) = 1733 close(7)
> 
> 
> gokdeniz> Any pointers to where should I look ?
> 
> Well, first we need to start with a version.  ;)  What version of autofs
> are you running?  Next, you will want to gather a debug log.  My people
> page has instructions on how to do that:
>   http://people.redhat.com/jmoyer
> 
> That *should* make things fairly obvious.  If it doesn't (or even if it
> does), post a follow-up message here.
> 

Hello,

Restarting autofs fixed the problem. I added --debug and restarted autofs. And 
there is no latency now. I removed the --debug and restarted again, still good. 
Version is debian's 4.1.4-13, I'm stating it to provide clue if anyone 
encounters the same problem.

Thanks for your answer.

- Gokdeniz