[blivet][all branches] Reduce some of the overhead in blivet's logging (#1155984)

Samantha N. Bueno sbueno+anaconda at redhat.com
Fri Sep 11 19:31:32 UTC 2015


On Fri, Aug 21, 2015 at 01:01:07PM -0700, Brian C. Lane wrote:
> On Fri, Aug 21, 2015 at 01:25:38PM -0400, Samantha N. Bueno wrote:
> > The primary issue these patches solve is to reduce the amount of wait
> > time between clicking "Create partitions" in the custom part spoke and
> > then seeing those results show up on the screen. With a higher number
> > of disks, e.g. 12, you could wait up to seven or eight seconds for the
> > GUI to refresh. I imagine that with more disks, the problem becomes even
> > more pronounced.
> > 
> > There's definitely more to do here, but it's tedious. The current set
> > reduces the amount of wait time from about seven seconds to about two.
> > 
> > Thanks to dlehman for combing over the logs with me and helping out
> > with identifying some problem areas.
> 
> Do you have a baseline with all the logging removed/disabled so we know
> what to aim for? 2s is still pretty slow, but WAY better than 8 :)

I missed this, sorry.
I didn't have a baseline when I wrote this; I recall that I tried
just setting blivet's loglevel to critical when I began working on this
just to see, but even though debug statements are not being written they
are still being evaluated -- so there went that idea.
I didn't have it in me to manually go through and comment out every
logging statement in blivet to get a good idea of a baseline for current
7.x.
On 7.0 however, the amount of wait time was almost negligible; assuming
I'm analyzing the logs correctly, it appears to be less than one second --
start:
09:08:30,193 DEBUG anaconda: page clicked: New Red Hat Enterprise Linux 7.0 Installation

end:
09:08:30,224 DEBUG anaconda: finished updating space display

So I'd say that 31ms is a pretty good baseline.

Samantha


More information about the anaconda-patches mailing list