[Users] Problem in 3.3 with shmmax configuration

[Bob Doolittle]

On 11/05/2013 02:13 AM, Sandro Bonazzola wrote:
> Il 04/11/2013 17:17, Bob Doolittle ha scritto:
>> On 11/04/2013 05:43 AM, Sandro Bonazzola wrote:
>>> Il 02/11/2013 17:51, Bob Doolittle ha scritto:
>>>> Hi,
>>>>
>>>> I'm setting up Engine for the 2nd time - the first time I answered a configuration question wrong. So I did:
>>>>
>>>> engine-setup
>>>> engine-cleanup
>>>> engine-setup
>>>>
>>>> Things worked, until I rebooted the system. I found that postgresql would not startup, and was failing with "could not create shared memory segment:
>>>> Invalid Argument".
>>>>
>>>> I resolved this issue by creating a file /etc/sysctl.d/10-shmmax.conf, containing the line:
>>>> kernel.shmmax = 1000000000
>>>>
>>>> (I read somewhere that postgresql recommends setting shmmax to 1/4 of physical memory, and I have 4GB)
>>>>
>>>> 1. Is this a known bug? If not, should I file one? If so, how do I do that? :)
>>> Which version are you installing?
>> 3.3, on Fedora 19.
>>
>>> Can you please attach all 3 logs from above sequence (setup, cleanup, setup)?
>>> I think something may have gone wrong on second setup execution while setting shmmax.
>> Unfortunately I no longer have those logs.
>>
>> *However* last night I powered down my node and engine. Last night engine would at least come up.
>>
>> Now when I boot up my engine I get the same error. So my guess is that the shmmax setting isn't being configured in a persistent fashion somehow, or
>> is somehow reverting.
>>
>> My most recent setup log is here (I hate to send large logs to lists):
>> https://dl.dropboxusercontent.com/u/35965416/ovirt-engine-setup-20131103141618.log
> 2013-11-03 14:18:18 DEBUG otopi.plugins.ovirt_engine_setup.system.sysctl plugin.execute:441 execute-output: ('/sbin/sysctl', '-n', 'kernel.shmmax')
> stdout:
> 35554432
>
> setup detected that your shmmax is already configured with a good value so it didn't change the configuration.
> If you set above value before running setup, please retry running setup without setting it or setting it to a value less than 35554432.
> setup will detect it's too low and will create needed configuration files for fixing it.


Except that when the system booted postgresql failed to come up because 
it failed to allocate the shm segment it wanted, and when I checked 
/proc/sys/kernel/shmmax the value was much lower than that. I have again 
set the value in syscfg and things are working now.

So my guess is that something is reverting the value after engine-setup 
has set it, or else somehow it was not set in a persistent fashion by 
engine-setup. Where is the shmmax value stored by engine-setup? Should 
there be a file in /etc/syscfg.d? There was not.

-Bob