Question about varnish memory usage

Per Buer perbu at
Thu Sep 18 21:28:40 CEST 2014

On Thu, Sep 18, 2014 at 5:54 PM, Adam Schumacher <
adam.schumacher at> wrote:

> Per,
> Thanks for the reply
> >It seems you are using Transient storage. Transient is usually uncapped.
> You can cap Transient but you should probably figure out why Varnish is
> using Transient storage. Usually it will >allocate Transient if
> >1) The objects TTL < 10s
> >2) Failed to allocate storage
> From varnishstat I see:
> SMA.Transient.g_bytes      5525388          .   Bytes outstanding
> If that is the same Transient storage you are referring to, I don’t see
> how varnishd could be taking up as much VM as it is.  Or am I missing
> something?

No, you're not. I read your email too fast. Seems your transient usage is
relatively decent and not in the gigabytes.

Wrt, Dridis point about threads might also not be that relevant as only
700-800 threads are being created. Session workspace is virtual until
allocated so I don't think that is where your memory is disappearing.

24 thread pools is a bit odd (you only need two) but don't know of any
leaks it might trigger.

I'd stay away from classic hashing as cribit is stable and you're the only
one I know that uses it.

 When is the varnishstat taken? Right before running out of memory? That
would probably be useful.

> >As a hotfix I would cap Transient at, say, 2G. "varnishd -s
> Transient=malloc,2G". You can probably up your main storage quite a bit
> with this in place.
> I’ll give it a shot and let you know.  It usually takes a couple days
> before filling up all the physical memory.

Well, I'm somewhat doubtful it it will actually work.

*Per Buer*
CTO | Varnish Software AS
Cell: +47 95839117
We Make Websites Fly!
[image: Register now]
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the varnish-misc mailing list