[Samba] 3.0.4: smbd's + nscd's = 100% CPU; load > 4
Hansjoerg Maurer
Hansjoerg.Maurer at dlr.de
Tue Jul 6 10:39:28 GMT 2004
Hi,
I had a similiar Problem , and a loglevel of 4 shows , that samba was
trying to look up
a user nobody and a user Administrator, all the time.
If I killed nscd the load of the ldap server becomes high...
I added these user to my ldap backend, and the problem disappears.
Greetings
Hansjörg
Dragan Krnic schrieb:
>>>>>the new 3.0.4 Samba installation seems to work fine
>>>>>except that from time to time but at least a couple
>>>>>of times a day one or more smbd processes start
>>>>>running at 20%-40% CPU each and 6 nscd processes
>>>>>then share the remaining CPU power. System 70%-80%
>>>>>users the rest 20%-30%. Load rises fast to over 4.
>>>>>
>>>>>I'm sure that each such process is just idling,
>>>>>but why does it engage so much nscd processing?
>>>>>
>>>>>As soon as I kill the excessive smbd process(es)
>>>>>the situation drops to normal, i.e. load < 0,1
>>>>>no perceptible CPU%.
>>>>>
>>>>>Does anyone know what's happening?
>>>>>
>>>>>
>>>>What does strace say ? Can you attach with gdb to
>>>>a CPU bound process and give a backtrace ?
>>>>
>>>>
>>>Ah backtrace, to see the steps to the black hole?
>>>OK. Will do, sooen as I get back to office (tomorrow).
>>>
>>>
>>A pot watched never boils.
>>But as soon as it happens again, I'll consult
>>strace and gdb to see how and why it happens.
>>
>>
>
>It does, when you stop watching.
>I was away yesterday and what do I see
>this morning:
>
> top - 09:55:35 up 6 days, 17:28, 6 users, load average: 3.59, 3.82, 3.15
> Tasks: 182 total, 3 running, 179 sleeping, 0 stopped, 0 zombie
> Cpu(s): 21.6% user, 78.4% system, 0.0% nice, 0.0% idle
> Mem: 2060704k total, 2004324k used, 56380k free, 185272k buffers
> Swap: 2402296k total, 5236k used, 2397060k free, 1068752k cached
>
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ Command
> 12808 robf 25 0 2996 2596 2244 R 23.6 0.1 4:46.98 smbd
> 12741 robf 22 0 3028 2628 2280 R 20.6 0.1 5:44.55 smbd
> 2354 root 15 0 724 716 536 S 15.3 0.0 150:31.23 nscd
> 2356 root 15 0 724 716 536 S 14.9 0.0 150:35.87 nscd
> 2352 root 15 0 724 716 536 S 9.0 0.0 151:14.08 nscd
> 2353 root 15 0 724 716 536 S 6.3 0.0 150:39.89 nscd
> 2355 root 15 0 724 716 536 S 6.3 0.0 150:31.82 nscd
> 2350 root 15 0 724 716 536 S 3.7 0.0 150:49.78 nscd
>
>I attached both of the smbd processes to gdb and
>backtrace was always:
>
> #0 0x402e5328 in read () from /lib/libc.so.6
> #1 0x40343b90 in __DTOR_END__ () from /lib/libc.so.6
> #2 0x4031d58b in __nscd_getpwnam_r () from /lib/libc.so.6
> #3 0x402c130d in getpwnam_r@@GLIBC_2.1.2 () from /lib/libc.so.6
> #4 0x402c0e6f in getpwnam () from /lib/libc.so.6
> #5 0x081298fd in get_memberuids ()
> #6 0x08129b09 in _samr_query_groupmem ()
> #7 0x081215d1 in api_samr_query_groupmem ()
> #8 0x081358b2 in api_rpcTNP ()
> #9 0x08135632 in api_pipe_request ()
> #10 0x0812fc10 in process_request_pdu ()
> #11 0x0812fdec in process_complete_pdu ()
> #12 0x08130069 in process_incoming_data ()
> #13 0x08130220 in write_to_internal_pipe ()
> #14 0x081301a4 in write_to_pipe ()
> #15 0x080883e0 in api_fd_reply ()
> #16 0x080885b9 in named_pipe ()
> #17 0x080891bc in reply_trans ()
> #18 0x080c80e0 in switch_message ()
> #19 0x080c8172 in construct_reply ()
> #20 0x080c8491 in process_smb ()
> #21 0x080c9004 in smbd_process ()
> #22 0x081f812e in main ()
> #23 0x402268ae in __libc_start_main () from /lib/libc.so.6
>
>After killing both smbd processes with -9 the top soon
>stabilizes at:
>
> top - 10:12:01 up 6 days, 17:45, 6 users, load average: 0.03, 0.19, 1.17
> Tasks: 175 total, 2 running, 173 sleeping, 0 stopped, 0 zombie
> Cpu(s): 0.0% user, 0.7% system, 0.0% nice, 99.3% idle
> Mem: 2060704k total, 2034272k used, 26432k free, 185272k buffers
> Swap: 2402296k total, 5236k used, 2397060k free, 1100020k cached
>
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ Command
> 15182 root 15 0 956 956 700 R 0.3 0.0 0:03.67 top
>
>Unfortunately I didn't trace the nscd processes.
>What a shame! I'll do it next time.
>
>Nobody complained yet about reduced performance.
>
>It's hard to tell when this behaviour started.
>The upper bound seems to be 9 hours,
>the combined run times of the nscd processe,
>some time during the night when the computers
>were totally quiet. The lower bound based on
>the run times of the smbd processes is more
>like half an hour ago.
>
>This is the fourth out of 5 times that the same user,
>"robf", is involved as the effective UID of the smbd process.
>The other one time was root's own smbd.
>
>Jeremy, can I provide more information?
>
>Cheers,
>Dragan
>
>
--
_________________________________________________________________
Dr. Hansjoerg Maurer | LAN- & System-Manager
|
Deutsches Zentrum | DLR Oberpfaffenhofen
f. Luft- und Raumfahrt e.V. |
Institut f. Robotik |
Postfach 1116 | Muenchner Strasse 20
82230 Wessling | 82234 Wessling
Germany |
|
Tel: 08153/28-2431 | E-mail: Hansjoerg.Maurer at dlr.de
Fax: 08153/28-1134 | WWW: http://www.robotic.dlr.de/
__________________________________________________________________
There are 10 types of people in this world,
those who understand binary and those who don't.
More information about the samba
mailing list