Ludovic Courtès <ludo@gnu.org> writes:

> Hi,
>
> Mathieu Othacehe <othacehe@gnu.org> skribis:
>
>>> I have now copied the database to a tmpfs mounted directory to make sure
>>> that those inconsistent duration are only caused by the I/O pressure on
>>> berlin.
>>
>> This helps a lot. The Cuirass web service has been running smooth since
>> two days, without any inconsistent query times.
>
> Interesting.
>
>> I'm considering using a tmpfs backed database for good. The problem is
>> that we would need a save/restore mechanism in case Berlin
>> reboots.
>
> Hmm sounds risky, no?
>
> I wonder if we could instead ensure no I/O-intensive workload runs that
> machine.  I’m thinking in particular of the derivations that produce
> ISO/qcow images that are not offloaded but maybe should.
>
> WDYT?  Do you think that’d be enough?  Or is tmpfs our only hope?

I think Ricardo mentioned that the machine running Cuirass uses an SSD
for the root filesystem, so moving the database there may help?