[Samba] performance problem on bridgehead DC

MORILLO Jordi j.morillo at educationetformation.fr
Fri Mar 16 13:11:08 UTC 2018 

Hi Louis,

Move up to 4.7 is planned soon but I'm little scared to update bridgehead if it actually had performance problem.
I would like to fix all problem before updating.

My smb.conf is really simple:

[global]
        workgroup = PR
        realm = pr.educationetformation.fr
        netbios name = DC000
        server role = active directory domain controller

        bind interfaces only = yes
        interfaces = lo eth0

        printcap name = /dev/null
        load printers = no

        dns forwarder = 10.1.5.1 10.1.1.12

        # Needed for FreeRadius PEAP MSCHAP
        ntlm auth = yes
        raw NTLMv2 auth = yes

[netlogon]
        path = /var/lib/samba/sysvol/pr.educationetformation.fr/scripts
        read only = No
        guest ok = yes

[sysvol]
        path = /var/lib/samba/sysvol
        read only = No
        guest ok = yes

King regards
-----Message d'origine-----
De : samba <samba-bounces at lists.samba.org> De la part de L.P.H. van Belle via samba
Envoyé : vendredi 16 mars 2018 12:13
À : samba at lists.samba.org
Objet : Re: [Samba] performance problem on bridgehead DC

Jessie also has the problem, even before the meltdown patches. 
The kernel 3.16.x had ksoftirqd probems, eating up cpu cicles. 
Make sure you use kernel 4.9.0-6 

And i suggest move up to samba 4.7.x or 4.7.4 from debian/buster, Or use my 4.6.12/4.7.6 packages for stretch, or ask if Tranquil.it can make them. 
The 4.6.14 is still in the make atm for stretch, 4.6.14 for jessie is done. 

And last, without any info from smb.conf we cant help much..
We need to see your settings also.. 

Greetz, 

Louis


> -----Oorspronkelijk bericht-----
> Van: samba [mailto:samba-bounces at lists.samba.org] Namens MORILLO Jordi 
> via samba
> Verzonden: vrijdag 16 maart 2018 11:52
> Aan: samba at lists.samba.org
> Onderwerp: [Samba] performance problem on bridgehead DC
> 
> Hi everybody !
> 
> I am encountering performance problem on my bridgehead DC.
> I have 19 DC (Debian Stretch / Samba 4.6.7 from Tranquil.it
> repo) and they all synchronized on a main bridgehead DC.
> 
> This performance problem first appears when the bridgehead DC was on 
> Debian Jessie and kernel was updated with meltdown/spectre patch from 
> Debian (3.16.51-3+deb8u1) I added "nopti" option to grub boot to 
> resolve the performance problem.
> Last night, I upgrade this Bridgehead DC from Jessie -> Stretch (samba 
> package 4.6.7 from Tranquil.it is the same for Jessie and Stretch), I 
> also keep "nopti" option but problem is back again.
> 
> There is 2 processes eating lots of CPU, samba-tool drs showrepl takes 
> minutes, system is slow, load average is at 2 constantly
> 
> dc000:~# htop
>   PID USER      PRI  NI  VIRT   RES   SHR S CPU% MEM%   TIME+  Command
> 28923 root       20   0  641M 42604 19064 S 39.6  2.1  
> 1h26:53 /usr/sbin/samba
> 28930 root       20   0  641M 36340 13136 R 54.4  1.8 
> 45:00.58 /usr/sbin/samba
> 
> dc000:~# samba-tool processes | egrep "(28923|28930)"
> rpc_server             28923
> rpc_server             28923
> rpc_server             28923
> rpc_server             28923
> rpc_server             28923
> rpc_server             28923
> rpc_server             28923
> rpc_server             28923
> rpc_server             28923
> dreplsrv               28930
> 
> dc000:~# strace -p 28923 -f
> strace: Process 28923 attached
> strace: [ Process PID=28923 runs in x32 mode. ]
> strace: [ Process PID=28923 runs in 64 bit mode. ] fcntl(15, F_SETLKW, 
> {l_type=F_UNLCK, l_whence=SEEK_SET, l_start=332, l_len=1}) = 0 
> fcntl(15, F_SETLKW, {l_type=F_RDLCK, l_whence=SEEK_SET, l_start=368, 
> l_len=1}) = 0 fcntl(15, F_SETLKW, {l_type=F_RDLCK, l_whence=SEEK_SET, 
> l_start=5073496, l_len=1}) = 0 fcntl(15, F_SETLKW, {l_type=F_UNLCK, 
> l_whence=SEEK_SET, l_start=368, l_len=1}) = 0 fcntl(15, F_SETLKW, 
> {l_type=F_RDLCK, l_whence=SEEK_SET, l_start=368, l_len=1}) = 0 
> fcntl(15, F_SETLKW, {l_type=F_UNLCK, l_whence=SEEK_SET, 
> l_start=5073496, l_len=1}) = 0 fcntl(15, F_SETLKW, {l_type=F_UNLCK, 
> l_whence=SEEK_SET, l_start=368, l_len=1}) = 0 fcntl(15, F_SETLKW, 
> {l_type=F_RDLCK, l_whence=SEEK_SET, l_start=376, l_len=1}) = 0 
> fcntl(15, F_SETLKW, {l_type=F_RDLCK, l_whence=SEEK_SET, 
> l_start=5125448, l_len=1}) = 0 fcntl(15, F_SETLKW, {l_type=F_UNLCK, 
> l_whence=SEEK_SET, l_start=376, l_len=1}) = 0 fcntl(15, F_SETLKW, 
> {l_type=F_RDLCK, l_whence=SEEK_SET, l_start=376, l_len=1}) = 0 
> fcntl(15, F_SETLKW, {l_type=F_UNLCK, l_whence=SEEK_SET, 
> l_start=5125448, l_len=1}) = 0 fcntl(15, F_SETLKW, {l_type=F_UNLCK, 
> l_whence=SEEK_SET, l_start=376, l_len=1}) = 0 fcntl(15, F_SETLKW, 
> {l_type=F_RDLCK, l_whence=SEEK_SET, l_start=380, l_len=1}) = 0 
> fcntl(15, F_SETLKW, {l_type=F_RDLCK, l_whence=SEEK_SET, 
> l_start=1545572, l_len=1}) = 0 fcntl(15, F_SETLKW, {l_type=F_UNLCK, 
> l_whence=SEEK_SET, l_start=380, l_len=1}) = 0 fcntl(15, F_SETLKW, 
> {l_type=F_RDLCK, l_whence=SEEK_SET, l_start=380, l_len=1}) = 0 
> fcntl(15, F_SETLKW, {l_type=F_UNLCK, l_whence=SEEK_SET, 
> l_start=1545572, l_len=1}) = 0 fcntl(15, F_SETLKW, {l_type=F_RDLCK, 
> l_whence=SEEK_SET, l_start=2265976, l_len=1}) = 0 fcntl(15, F_SETLKW, 
> {l_type=F_UNLCK, l_whence=SEEK_SET, l_start=380, l_len=1}) = 0 
> fcntl(15, F_SETLKW, {l_type=F_RDLCK, l_whence=SEEK_SET, l_start=380, 
> l_len=1}) = 0 .......
> 
> dc000:~# egrep "(28923|28930)" /proc/locks
> 16: POSIX  ADVISORY  READ  28923 ca:01:132971 168 EOF
> 17: POSIX  ADVISORY  WRITE 28923 ca:01:132971 8 8
> 18: POSIX  ADVISORY  READ  28923 ca:01:132978 168 EOF
> 19: POSIX  ADVISORY  WRITE 28923 ca:01:132978 8 8
> 20: POSIX  ADVISORY  READ  28923 ca:01:132976 168 EOF
> 21: POSIX  ADVISORY  WRITE 28923 ca:01:132976 8 8
> 22: POSIX  ADVISORY  READ  28923 ca:01:132970 168 EOF
> 23: POSIX  ADVISORY  WRITE 28923 ca:01:132970 8 8
> 24: POSIX  ADVISORY  READ  28923 ca:01:132968 168 EOF
> 25: POSIX  ADVISORY  WRITE 28923 ca:01:132968 8 8
> 26: POSIX  ADVISORY  READ  28923 ca:01:132940 168 EOF
> 27: POSIX  ADVISORY  WRITE 28923 ca:01:132940 8 8
> 28: POSIX  ADVISORY  READ  28923 ca:01:132930 168 EOF
> 29: POSIX  ADVISORY  WRITE 28923 ca:01:132930 8 8
> 31: POSIX  ADVISORY  WRITE 28923 00:12:387271 0 EOF
> 54: POSIX  ADVISORY  WRITE 28930 00:12:389224 0 EOF
> 
> dc000:~# ls -la /proc/28923/fd/15
> lrwx------ 1 root root 64 mars  16 09:44 /proc/28923/fd/15 -> 
> /var/lib/samba/private/sam.ldb.d/DC=DOMAINDNSZONES,DC=PR,DC=ED
> UCATIONETFORMATION,DC=FR.ldb
> 
> I also have many access on :
> dc000:~# ls -la /proc/28930/fd/14
> lrwx------ 1 root root 64 mars  16 09:44 /proc/28930/fd/14 -> 
> /var/lib/samba/private/sam.ldb.d/DC=PR,DC=EDUCATIONETFORMATION
> ,DC=FR.ldb
> dc000:~# ls -la /proc/28930/fd/15
> lrwx------ 1 root root 64 mars  16 09:44 /proc/28930/fd/15 -> 
> /var/lib/samba/private/sam.ldb.d/DC=DOMAINDNSZONES,DC=PR,DC=ED
> UCATIONETFORMATION,DC=FR.ldb
> dc000:~# ls -la /proc/28930/fd/16
> lrwx------ 1 root root 64 mars  16 09:44 /proc/28930/fd/16 -> 
> /var/lib/samba/private/sam.ldb.d/DC=FORESTDNSZONES,DC=PR,DC=ED
> UCATIONETFORMATION,DC=FR.ldb
> 
> dbcheck is fine on all DC (10 000 objects) All dc are synced from the 
> bridgehead DC Bridgehead is running with 8 virtual CPU and 2Go of Ram
> 
> 
> If you have any idea, you are welcome :) Thanks
> --
> To unsubscribe from this list go to the following URL and read the
> instructions:  https://lists.samba.org/mailman/options/samba
> 
> 


-- 
To unsubscribe from this list go to the following URL and read the
instructions:  https://lists.samba.org/mailman/options/samba

More information about the samba mailing list