linux-nfs.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* Randomly inaccessible files through NFS
@ 2012-08-15  7:54 Denis V. Nagorny
  2012-08-17 10:26 ` Denis V. Nagorny
  0 siblings, 1 reply; 4+ messages in thread
From: Denis V. Nagorny @ 2012-08-15  7:54 UTC (permalink / raw)
  To: linux-nfs

Hello,

Using Scientific Linux 6.1 (I think it's equal to RH EL 6.1) we met the 
strange issue.  Several last months we have problem. After one or two 
days of successful work, files on nfs server begins to be randomly 
unacessible.
I doesn't mean that files becames hidden or something like this. It 
means that attempts to open some random files may be unsuccessful. 
Usually restart of nfs server makes situation better but for several 
days only. There are no any messages about errors in logs on server and 
clients machines. Can anybody point me how can I try to understand what 
happens at least. Sorry for my english.

Denis.

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: Randomly inaccessible files through NFS
  2012-08-15  7:54 Randomly inaccessible files through NFS Denis V. Nagorny
@ 2012-08-17 10:26 ` Denis V. Nagorny
  2012-08-17 10:50   ` Adrien Kunysz
  0 siblings, 1 reply; 4+ messages in thread
From: Denis V. Nagorny @ 2012-08-17 10:26 UTC (permalink / raw)
  To: linux-nfs

15.08.2012 11:54, Denis V. Nagorny пишет:
> Hello,
>
> Using Scientific Linux 6.1 (I think it's equal to RH EL 6.1) we met 
> the strange issue.  Several last months we have problem. After one or 
> two days of successful work, files on nfs server begins to be randomly 
> unacessible.
> I doesn't mean that files becames hidden or something like this. It 
> means that attempts to open some random files may be unsuccessful. 
> Usually restart of nfs server makes situation better but for several 
> days only. There are no any messages about errors in logs on server 
> and clients machines. Can anybody point me how can I try to understand 
> what happens at least. Sorry for my english.
>
> Denis.

Hello again,

I've made some additional experiments. It looks like nfs clients can be 
in one of two states: "quite stable" and "quite unstable". Clients are 
usually stable but after some heavy job with a lot of I/O with NFS 
server clients become "quite unstable" and fails even with single file 
operations with NFS server. In this state I can't unmount NFS shares and 
so on.  I've tried to analyse with wireshark and found that in unstable 
state there are a lot of NFS4ERR_EXPIRED answers from NFS server.  In 
one of experiments I've changed NICs in both machines involved - result 
the same. So I'm still looking for the ways to understand the problem.
Can anybody give me any advices?

Denis

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: Randomly inaccessible files through NFS
  2012-08-17 10:26 ` Denis V. Nagorny
@ 2012-08-17 10:50   ` Adrien Kunysz
  2012-08-17 12:28     ` Denis V. Nagorny
  0 siblings, 1 reply; 4+ messages in thread
From: Adrien Kunysz @ 2012-08-17 10:50 UTC (permalink / raw)
  To: Denis V. Nagorny; +Cc: linux-nfs

I would try to tcpdump all NFS traffic starting when the client is in
the "stable" state (including the MOUNT call). Once it's in the
"unstable" state, I would stop the capture then try to figure out
exactly at what point it switched from "stable" to "unstable" (maybe
figure out when exactly the NFS4ERR_EXPIRED start to happen) and track
it down to a specific NFS pattern.

I don't know much about NFS really so I cannot be more specific. Yes,
this probably requires lot of storage to capture all the traffic and
lot of time to analyse the captured data.

On Fri, Aug 17, 2012 at 11:26 AM, Denis V. Nagorny
<dvnagorny@compcenter.org> wrote:
> 15.08.2012 11:54, Denis V. Nagorny пишет:
>
>> Hello,
>>
>> Using Scientific Linux 6.1 (I think it's equal to RH EL 6.1) we met the
>> strange issue.  Several last months we have problem. After one or two days
>> of successful work, files on nfs server begins to be randomly unacessible.
>> I doesn't mean that files becames hidden or something like this. It means
>> that attempts to open some random files may be unsuccessful. Usually restart
>> of nfs server makes situation better but for several days only. There are no
>> any messages about errors in logs on server and clients machines. Can
>> anybody point me how can I try to understand what happens at least. Sorry
>> for my english.
>>
>> Denis.
>
>
> Hello again,
>
> I've made some additional experiments. It looks like nfs clients can be in
> one of two states: "quite stable" and "quite unstable". Clients are usually
> stable but after some heavy job with a lot of I/O with NFS server clients
> become "quite unstable" and fails even with single file operations with NFS
> server. In this state I can't unmount NFS shares and so on.  I've tried to
> analyse with wireshark and found that in unstable state there are a lot of
> NFS4ERR_EXPIRED answers from NFS server.  In one of experiments I've changed
> NICs in both machines involved - result the same. So I'm still looking for
> the ways to understand the problem.
> Can anybody give me any advices?
>
> Denis
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: Randomly inaccessible files through NFS
  2012-08-17 10:50   ` Adrien Kunysz
@ 2012-08-17 12:28     ` Denis V. Nagorny
  0 siblings, 0 replies; 4+ messages in thread
From: Denis V. Nagorny @ 2012-08-17 12:28 UTC (permalink / raw)
  To: linux-nfs

One more observation,
It looks like NFS4ERR_EXPIRED messages are delivered for the process 
blocked in the kernel:

Aug 17 13:18:41 srvmpidev03 kernel: INFO: task bcast2:6338 blocked for 
more than 120 seconds.
Aug 17 13:18:41 srvmpidev03 kernel: "echo 0 > 
/proc/sys/kernel/hung_task_timeout_secs" disables this message.
Aug 17 13:18:41 srvmpidev03 kernel: bcast2        D 000000000000000e     
0  6338      1 0x00000084
Aug 17 13:18:41 srvmpidev03 kernel: ffff880c238b76e8 0000000000000082 
0000000000000000 ffffffffa03a9eed
Aug 17 13:18:41 srvmpidev03 kernel: ffff880621561080 ffff880603c79aa0 
ffff880603c79bc0 00000001004eaf4f
Aug 17 13:18:41 srvmpidev03 kernel: ffff880c23f325f8 ffff880c238b7fd8 
000000000000f598 ffff880c23f325f8
Aug 17 13:18:41 srvmpidev03 kernel: Call Trace:
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffffa03a9eed>] ? 
__put_nfs_open_context+0x4d/0xf0 [nfs]
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8110d310>] ? 
sync_page+0x0/0x50
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff814daf13>] 
io_schedule+0x73/0xc0
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8110d34d>] sync_page+0x3d/0x50
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff814db77f>] 
__wait_on_bit+0x5f/0x90
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8110d503>] 
wait_on_page_bit+0x73/0x80
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8108e140>] ? 
wake_bit_function+0x0/0x50
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8110d5aa>] 
__lock_page_or_retry+0x3a/0x60
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8110e73f>] 
filemap_fault+0x2df/0x500
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff81136fe4>] 
__do_fault+0x54/0x510
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff81137597>] 
handle_pte_fault+0xf7/0xb50
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff811544ca>] ? 
alloc_pages_current+0xaa/0x110
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff81045b77>] ? 
pte_alloc_one+0x37/0x50
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff811381c8>] 
handle_mm_fault+0x1d8/0x2a0
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff810414e9>] 
__do_page_fault+0x139/0x480
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8112dbc0>] ? 
vma_prio_tree_insert+0x30/0x50
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8113ac8c>] ? 
__vma_link_file+0x4c/0x80
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8113b45b>] ? 
vma_link+0x9b/0xf0
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8113d9e9>] ? 
mmap_region+0x269/0x590
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff814e007e>] 
do_page_fault+0x3e/0xa0
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff814dd425>] 
page_fault+0x25/0x30
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8126e3af>] ? 
__clear_user+0x3f/0x70
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8126e391>] ? 
__clear_user+0x21/0x70
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8126e418>] 
clear_user+0x38/0x40
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff811c4e1d>] padzero+0x2d/0x40
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff811c6e3e>] 
load_elf_binary+0x88e/0x1b10
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff811330f1>] ? 
follow_page+0x321/0x460
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8113839f>] ? 
__get_user_pages+0x10f/0x420
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff811c38dc>] ? 
load_misc_binary+0xac/0x3e0
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff81179f2b>] 
search_binary_handler+0x10b/0x350
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8117b0b9>] 
do_execve+0x239/0x310
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8126e4ca>] ? 
strncpy_from_user+0x4a/0x90
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff810095ca>] 
sys_execve+0x4a/0x80
Aug 17 13:18:41 srvmpidev03 kernel: [<ffffffff8100b5ca>] 
stub_execve+0x6a/0xc0


17.08.2012 14:50, Adrien Kunysz пишет:
> I would try to tcpdump all NFS traffic starting when the client is in
> the "stable" state (including the MOUNT call). Once it's in the
> "unstable" state, I would stop the capture then try to figure out
> exactly at what point it switched from "stable" to "unstable" (maybe
> figure out when exactly the NFS4ERR_EXPIRED start to happen) and track
> it down to a specific NFS pattern.
>
> I don't know much about NFS really so I cannot be more specific. Yes,
> this probably requires lot of storage to capture all the traffic and
> lot of time to analyse the captured data.
>
> On Fri, Aug 17, 2012 at 11:26 AM, Denis V. Nagorny
> <dvnagorny@compcenter.org> wrote:
>> 15.08.2012 11:54, Denis V. Nagorny пишет:
>>
>>> Hello,
>>>
>>> Using Scientific Linux 6.1 (I think it's equal to RH EL 6.1) we met the
>>> strange issue.  Several last months we have problem. After one or two days
>>> of successful work, files on nfs server begins to be randomly unacessible.
>>> I doesn't mean that files becames hidden or something like this. It means
>>> that attempts to open some random files may be unsuccessful. Usually restart
>>> of nfs server makes situation better but for several days only. There are no
>>> any messages about errors in logs on server and clients machines. Can
>>> anybody point me how can I try to understand what happens at least. Sorry
>>> for my english.
>>>
>>> Denis.
>>
>> Hello again,
>>
>> I've made some additional experiments. It looks like nfs clients can be in
>> one of two states: "quite stable" and "quite unstable". Clients are usually
>> stable but after some heavy job with a lot of I/O with NFS server clients
>> become "quite unstable" and fails even with single file operations with NFS
>> server. In this state I can't unmount NFS shares and so on.  I've tried to
>> analyse with wireshark and found that in unstable state there are a lot of
>> NFS4ERR_EXPIRED answers from NFS server.  In one of experiments I've changed
>> NICs in both machines involved - result the same. So I'm still looking for
>> the ways to understand the problem.
>> Can anybody give me any advices?
>>
>> Denis
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html


^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2012-08-17 12:28 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-08-15  7:54 Randomly inaccessible files through NFS Denis V. Nagorny
2012-08-17 10:26 ` Denis V. Nagorny
2012-08-17 10:50   ` Adrien Kunysz
2012-08-17 12:28     ` Denis V. Nagorny

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).