Topic: Corrected - High CPU usage after update (Due to OpenLDAP bug)

[andy_wismer]

Hello

I'm trying to help in testing out certain contribs and have two older HP Proliant servers equipped for testing.

My "productive" servers all run on VMWare ESXi (At the moment still ESXi 5.1 SP2) and I'm NOT seeing this behaviour on the VMs.
All productive systems are using SME in "single disk" mode, I prefer to let ESXi and the underlying Hardware handle disk redundancy.

However, the Proliants are displaying problems which I'm not sure about. I've been using SME since 6.x and Linux almost 10 years more, so I do have a bit of experience...

The environment:

Server1:
HP Proliant ML110 G2, 4 GB RAM, 2 SATA Disks with 500 GB. NO Hardware RAID.
Installed OS: SME 9.0 x64

Server2:
HP Proliant ML110 G3, 2.5 GB RAM, 2 SATA Disks with 500 GB. NO Hardware RAID.
Installed OS: SME 9.0 i586

Process:
SME 9 installed, Disks with RAID1 as per SME default.
After installation, Disks take some time to finish mirroring, CPU goes from 1.0 to almost 0.
Contribs installed, still all as expected.
Update to latest Kernel and all per yum update.
After reboot, the CPU usage remains at ca. 50%, or to quote htop:

  1  [||||||||||||||||||||||                          38.9%]     Tasks: 149, 11 thr; 1 running
  2  [|||||||||||||||||||||||||||||||||               59.2%]     Load average: 1.85 1.75 1.73
  Mem[||||||||||||||||||||||||||||               409/2401MB]     Uptime: 06:17:34
  Swp[                                             0/2543MB]

  PID USER      PRI  NI  VIRT   RES   SHR S CPU% MEM%   TIME+  Command                                                         
17719 root       20   0  5348  1920  1160 R  1.4  0.1  0:39.71 htop                                                           
  862 root    20   0   112    32    20 S  0.5  0.0  0:44.14 runsv ldap
 2473 qpsmtpd    20   0 21140 16016  3612 S  0.5  0.7  0:02.60 /usr/bin/perl -Tw /usr/bin/qpsmtpd-forkserver -u qpsmtpd -l 0.0.
 1667 root    20   0  3560   672   484 S  0.0  0.0  0:04.70 irqbalance --pid=/var/run/irqbalance.pid
 6679 root    20   0 25124  8228  2532 S  0.0  0.3  0:03.16 /usr/bin/perl -w /usr/bin

This is from Server2, after a fresh install with NO contribs (just to verify it's from the SME OS and NOT from a contrib).
SME9admin was installed afterwards to "graph" the CPU usage - it didn't change the "usage"...

During the time the disks are mirroring, htop will display a certain cpu usage, but will not show any app / command actually using this - it's the kernel doing the mirroring, not a specific app.
What I'm seeing looks like this on both servers.
CPU usage very high, but no app actually causing this.
Disks are mirrored, as per cat /proc/mdstat...

My suspects are:
irqbalance --pid=/var/run/irqbalance.pid

I haven't tried this on my dell box I have lying around for tseting, that one's a bit newer with a Core2Duo CPU and 4 GB RAM.
But both HP Proliants display the same behaviour.
I have several 80GB, 160GB disks lying around which I use for testing.
Refitting those same servers with the latest SME 8.1 AND several contribs does NOT display this behaviour, the CPU remains at say 5% (instead of 50+% !!!)

Has anyone else seen this or similiar behaviour after updating?

@Stephdl
Your SME9admin contrib on VMWare is one of the reasons I'm using REAL servers to test this out, The noticed problem was a side-effect of testing...
I get fan info and graphs, but nothing else. phpsysinfo does show all the sensor values on the same hardware...

Thanks to the team for a great Server!
Regards
Andy

It seems this has to do with the OpenLDAP bug: http://bugs.contribs.org/show_bug.cgi?id=8635
Doing the command
chown ldap:ldap /var/lib/ldap/__db.*
solved the situation - for now...


 

[ReetP]

Hi Andy,

it's great that you're testing ! We need all the help we can get.

Please follow this sort of issue up on the bug tracker - that's where the real testing and dev work gets done. We can then track exactly who is doing what and help where necessary.

If you want to pose dev related questions then please do so on the dev list.

Just ask if you have any questions.

B. Rgds
John

[stephdl]

@Stephdl
Your SME9admin contrib on VMWare is one of the reasons I'm using REAL servers to test this out, The noticed problem was a side-effect of testing...
I get fan info and graphs, but nothing else. phpsysinfo does show all the sensor values on the same hardware...

Je ne posséde rien, tout appartient à la communauté. That would mean : I own nothing, everything is owned by the community.

[andy_wismer]

Hello

@ReetP At first I wasn't sure if this was a hardware issue or a bug, like I mentionned, they're not exactly new servers. (Disks are new though... .
I was able to pinpoint the Bug causing this, but since the devs are already aware of the problem, I didn't add to the bug info - as I maybe should have.

@stephdl Moi, je ne posséde rien, sauf le gratitude pour la communauté open source (et sme!), et le pétit feedback que je peux rétourné aux communauté.
Me, I own nothing but the gratidude for the open source community (and sme) and the little bit of feedback I can give in return.
Oui, je peux parler français, mais écrire, c'est un autre chose...
If you're interested, I can provide you with root access to such a testing server. I've been using SME9admin and it's predecessors for some time, and if that's what's needed to perfect an almost perfect add-on... Drop me a line or E-Mail...

Open Source: Don't bitch about stuff not working or not quite perfect.
But if you can, improve the code and give back your contribution to the comunity...



Best regards to the team and all the sme comunity!
Andy Wismer