phk at phk.freebsd.dk
Sat Jul 8 19:08:05 CEST 2006
In message <ujru05sav7b.fsf at cat.linpro.no>, Dag-Erling =?iso-8859-1?Q?Sm=F8rgra
>"Poul-Henning Kamp" <phk at phk.freebsd.dk> writes:
>OK. I consider "set backend.port" a priority for reasons which I'll
>explain below; I'll try to fix it myself during the weekend.
Yes, obviously this is so basic that it needs to work, I was not
suggesting abandonning VCL :-)
>> (start running live in VG.no ?)
>When I spoke with Anders earlier today, he expressed interest in
>having Varnish running continously with a low load. This will allow
>us to assess the stability of the code and identify any memory leaks.
Yes, this is obviously a good idea, but we need a program to sort
the logfiles into "everything looks sane" and "something looks out
of pattern" before we can really benefit.
>I'd like to run Varnish with valgrind / electric fence / jemalloc with
>debugging turned on / something similar for at least a few hours and
>see what comes up.
Yes, we should do that, but considering how little memory management
we actually do, we should not spend too much time. There are there
are less than 50 places where memory is actually allocated in the
>BTW, feel free to instrument Varnish with FlexeLint comments.
I generally prefer not to splatter flexelint comments in source
and just live with the few warnings I cannot silence.
>> (Invite alpha testers ?)
>As regards testing, I would like to roll a release of the live-test-1
>tag (or something very close to it, with backend.port working properly
>and debugging mode off by default) and stick it in ports to attract
I would prefer to not do until a bit more work (as indicated in
my list) has been done, but then I certainly think it is a good
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.
More information about the varnish-dev