Hi,
Following up on a discussion in bug 1184804 I took a look at the sysctl
values each kernel ships in a separate file.
vm.dirty_ratio is set to 20 on pae and x86_64 default kernels. AFAICS
this is the builtin default of the kernel so can be dropped.
kernel.hung_task_timeout_secs is set to zero. There's also a
DEFAULT_HUNG_TASK_TIMEOUT kconfig setting. So I think this would be
better set as kconfig option.
That leaves some IPC settings that were set in 2006 due to bug 146656:
| Name | where | setting | default |
|---------------+-------+--------------------+--------------------+
| kernel.msgmax | all | 65536 | 8192 |
| kernel.msgmnb | all | 65536 | 16384 |
| kernel.shmmax | 32bit | 0xffffffff | 0xfeffffff |
| kernel.shmall | 32bit | 0xffffffff | 0xfeffffff |
| kernel.shmmax | 64bit | 0x0fffffffffffff00 | 0xfffffffffeffffff |
| kernel.shmall | 64bit | 0x0fffffffffffff00 | 0xfffffffffeffffff |
All of them were supposed to be higher than the default. The shmmax and
shmall on 64bit systems are actually lower though if I calculate
correctly (upstream default is ULONG_MAX - (1UL << 24)). So I guess
those two can be dropped on 64bit systems?
Is the 32bit limit relevant? If not, shm* wouldn't need to be set at all
via sysctl.
Then we'd only have msgmax and msgmnb left.
In the bug the upstream defaults were called "totally useless". Maybe
makes sense to take the change upstream then? :-)
In any case the setting is not kernel version or flavor dependent. So
even if not for upstream, we could set msgmax and msgmnb in aaa_base
instead. It already ships distro defaults:
https://github.com/openSUSE/aaa_base/blob/master/files/usr/lib/sysctl.d/50-…
Thoughts?
cu
Ludwig
--
(o_ Ludwig Nussel
//\
V_/_ http://www.suse.com/
SUSE Software Solutions Germany GmbH, GF: Felix Imendörffer
HRB 36809 (AG Nürnberg)
https://www.suse.com/support/kb/doc/?id=000019780
I found this report from last year that seems to match a problem encountered on one of our non-opensuse servers. I dropped by irc://libera.chat/opensuse to see if I could find the bug and patch to check the source of our CIFS module to see if it was missing.
They told me to ask the kernel list.
Can you help? Thanks.