dual cache, dual link again

From: Graham Toal <gtoal@dont-contact.us>
Date: Wed, 16 Apr 1997 16:35:56 -0500 (CDT)

Well, as it does every month or two, disaster struck and one of
our links to the net went down today. Since this was the first time
since I switched from CERN to squid, and I'd had a very busy few
weeks, I hadn't prepared in advance how to switch one of the caches
from using the other (whose link was up) as a sibling to being
a parent, thus bypassing the problem.

I made what I thought were the obvious changes:

I changed the cache_host command on the broken host
from 'sibling' to 'parent', and I added "inside_firewall vt.com"
so that it would not try to access outside sites directly
but would fetch them unconditionally from the parent.

I didn't change anything on the other cache which was working OK.

However when I restarted the server, it did not manage to
find any pages. I was in too much of a panic to stop to
work out why, as I had to go reinstall a CERN cache and
get it to fetch from the other squid instead.

Any obvious suggestions as to where I screwed up?

Also, for Wayne - it occurred to me that a sufficient
test (you were asking how the squid could tell) that
an external link is down would be to ping a particular
router in our upstream connection. The most common
outage cause is that our direct link to sprintlink goes
down, it's very seldom something farther upstream.
(Mainly because they're caught and fixed by sprint as a
matter of some urgency, but little guys like us being offline
aren't noticed until we call in a fault report :-);
also most upstream problems are rerouted around by Sprint)

Finally, Wayne added permission for my two caches to
access the national caches on the net today. Let's say
I make the dangerous assumption that if the appropriate
link is up, I can guarantee to reach those caches.
What would the commands be to set up the situation
I've been talking about where both my caches can use
the other if their link (national parent) is down?

To do this, do I have to fetch *all* pages from the
parents and lose the possibility of going directly?
(I can live with that if you can, but I don't want
to burden your servers unnecessarily *or* slow
down access to closer pages if I can help it, though
since they're only 70ms away it probably doesn't hurt
that much)

This connectivity stuff in squid is complicated and
I haven't grasped it all yet. I am however reading the
docs any chance I get. Excuse me if I occasionally try
to shortcut the process by asking here first - I have
36 different jobs in my queue at the moment and can only
allow myself about half an hour per day to work on the squid
caches :-(


Received on Wed Apr 16 1997 - 14:46:06 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:35:00 MST