Re: multiple A records

Duane Wessels wrote:
>
> >>> Compile 1.1.11 with -DRETRY_CONNECT=1 (it's the default in 1.2.alpha)

> >It would have been nice if this feature had been announced!!!! (And why
> >isn't it the default in 1.1.11 anyway?).

> I don't think it works perfectly just yet. I ran it on the NLANR
> caches and had some coredumps** with it. YMMV. So rather than spend a
> lot of time getting it working in 1.1, I spent time to get it working
> in 1.2.
:) cool!

are you runnin 1.2.alpha anywhere loaded yet?

> >>can't we get ipcacheCycleAddr to return addrs.count and then pass that
> >>instead of using the arbitrary number 4? (or am I missing something -
> >probably,
> For one thing it was just me tinkering around. 4 seemed like a reasonable
> number.
thought so :)

> If a host only has one address, why not retry connecting to the same
> address a few times as well? And if host has 20 addresses, should
> we try all 20 before giving up?
>
> Obviously it needs to be more configurable. So tell me how you would
> like to configure it in squid.conf.

We might want to think about regex-type settings that
allow us to change settings for problem sites... I presume that this
isn't trivial though...

#server_retry regex tries-per-server total-tries time-spent(seconds)
#their servers seem to go down in batches, and stay down, so try each
#site once for a maximum of 20 times
server_retry microsoft.com 1 20 0
#default? tries each server 3 times, cycles through all servers, no timeout
server_retry . 3 0 0

time-spent would put a limit on the time that it tries to connect - people
will get irritated if it takes 20 minutes for it to tell them "sorry, I can't
connect"......

> **Probably the main reason for most coredumps in Squid is when
> we issue a callback and the 'callback_data' memory has been freed.
> This is why we have things like ipcache_unregister() etc. But
> note there is no "connect_unregister()."
how does Michael Pelletier's patch handle this? or does it rely on
his malloc not "truly" giving it back to the OS and therefore still
having the right values until by chance we next xmalloc it?

> For squid-1.2 I've got a scheme where any generic callback data can
> registered and unregistered. Before a callback function is made, we
> check to see if the data area is still valid. If not, the callback is
> not made. This helps quite a bit, as you might imagine.
yep.

        Oskar
Received on Tue Jul 29 2003 - 13:15:42 MDT