Question on Amavis and ramdisk

[Willem Jan Withagen]

On 21-8-2017 15:23, Mark Martinec wrote:
> 2017-08-21 14:47 Willem Jan Withagen wrote:
>> In the amavis rc.d file is noted:
>> "========================================================"
>> "WARNING: using ramdisk is reported to be unstable and"
>> "thus it is highly recommended to be turned off."
>> "========================================================"
>>
>> And this warning seems there since 2012....
>>
>> Is this warning still valid?
>> And if YES, could somebody try and enlighten me as to what is unstable
>> on this config?
>>   - Is the ramdisk itself unstable?
>>   - Or is it the fact that upon a crash de ramdisk is lost and email
>>     might be lost?
> 
> I don't really know what was the reason for this warning, but I can
> guess that it's because the port creates a mdmfs ram disk of a fixed
> size for the %%AMAVISDIR%%/tmp file system, and any fixed size small
> disk may eventually run out of space, either during some peak mail
> traffic rush-in, of perhaps when soma stale temporary files happen
> to be left undeleted and accumulating, while this goes unnoticed
> for some time.

Oke, so these are the regular sysadmin troubles of a (too) small a disk.
Nothing a good purge can not fix....
Full disk might indeed lead to a sort of DOS situation.

> Using tmpfs instead of mdmfs could avoid some of the above concerns
> if you really want to use a ram disk. In my experience with a
> modern host and file systems, especially with SSD, there is no longer
> any substantial speedup by using a ram disk instead, so I don't
> think it is worth sacrificing memory for a ram disk, which could
> better be used by file system caches etc.

'mmmm this system runs in VM on KVM in OpenStack.
And one of the bottlenecks is amavis flushing all mail to a disk. And
that pushes the virtual disk to their limit.
Switching some ramdisk on really does help in the load.
>>   - Or is it the fact that upon a crash de ramdisk is lost and email
>>     might be lost?
> 
> No, mail should not be lost. Write failures would be noticed and
> a feeding mailer would receive a temporary failure (smtp status 450),
> so mail should stay in the mailer's queue for a later retry.
> But left unattended for days, this would result in mail non-delivery
> notification to the sender.

Right. As far as I know will postfix only note amavis complete if Amavis
has really reported as such, and only then tkae the file from the
waiting queue. So a crash would not result in wrong deliveries.

So perhaps the wording on that message should be less strong and
prohibitive?

--WjW