How to scale journald correctly?

[lennart@xxxxxxxxxxxxxx (Lennart Poettering)]

On Di, 17.07.18 09:25, Martin Häcker (mh at sntl-publishing.com) wrote:

> On 13.07.18 17:07, Lennart Poettering wrote:
> > journald is designed to just write out what is being thrown at it,
> > with the speed the underlying device permits. It is using relatively
> > large reception buffers for this, so that clients don't have to stall
> > on journald. If you flood journald heavily it will eventually cause
> > clients to block though.
> 
> Is there a known bug about this in version 219? Because what we see is
> that even with a moderate write load of about 100-500 kb / s at some
> point journald seems to snap and starts reading and writing at full disk
> throughput speed (which is about 50mb/s on our servers).
> 
> That smells like a bug to me.
> 
> As we're using the long term support version fo SystemD in CentOS 7.5, I
> guess that this behavior should be  fixed / discussed / workarounds
> documented - but I'm not entirely sure what the right way to proceed
> would be here. Please advise.
> 
> For reference, I'm referring to the behavior documented in this bug
> report: https://bugzilla.redhat.com/show_bug.cgi?id=1599658

219 is really old, we don't really focus on that upstream here. I know
that RHEL has backported a number of fixes, but you'd have to enquire
the RHEL/CentOS people about this...

Lennart

-- 
Lennart Poettering, Red Hat