Typical isnt it? You post an idea and then find the problems with the
idea later :). Sheesh.
Thanks for your ideas about DNS and autoproxy scripts, but I would
really rather _ONE_ proxy address with no fuss and bother on the
client or messing with DNS. A proxy end solution is a the best
solution IMHO.
I wrote:
| I cant see any way for squid on proxy1 to send, say 50%, of its misses
| to proxy 2. I think running a 'nanoproxy' on the 9.20 address that has
| both real caches as its parent and uses round-robin but no local
| disk space it self looks too hard (well it would be easy, but I cant
| see an easy way to do it without using more disk space :( and I just
| dont want that complexity).
|
| My solution is looking like writeing something like pluggw which sits on
| 9.20 and shares the load by redirecting the requests to 9.10 or 9.11 (rando
>| mly
| or something).
|
Having a pluggw option will break the access logs. *sigh* Having two
or even three squids running will make logfile parsing more complex,
but I can live with that. :)
Ok, take number two.
Have proxy1 with proxy2 as its sibling (proxy only) for 5 mins and then
reconfigure it to have proxy2 (proxy only) as its parent for 5 mins.
Keep swapping for ever. Problem: I dont trust it not to fail during
swapover.
Or.... could have proxy2 get a subset list of objects from proxy1, transfer
them to itself from proxy1 then remove the objects from the log on proxy1
and run check_cache.pl to nuke then off the disk (or somehow 'purge' them).
This would migrate objects from proxy1 to proxy2. Proxy1 would then
use proxy2 as a sibling (proxy-only). Problem: I dont trust it to not
barf on objects during the clean.
Or.... Running a third proxy that had both proxy1 and proxy2 as
proxy-only parents is seeming better and better actually.
Or.... Run proxy1 and proxy2 as siblings. Have the main proxy address
move between the two once per day. This means one box will collect a
day's worth of proxying and then the other will the next day.
Swap at 4am to minimise connection loss. Problem: cache requests in
progress at swap time will be lost when the
IP number switches.
So, I picked number 3, because it seemed easier and I wasnt convinced
that I wouldnt lose data using number 1 and it saved programming number 2.
Number 4 just seemed sad :).
I ran a proxy on 9.20 with:
cache_host 134.178.9.11 parent 3128 3130 round-robin proxy-only
cache_host 134.178.9.10 parent 3128 3130 round-robin proxy-only
But 9.11 always gets the requests!!
In my logs I get:
(on 10)
884823653.491 0 134.178.9.20 UDP_MISS/000 32 ICP_QUERY http://web/ - NONE/- -
884823655.514 0 134.178.9.11 UDP_MISS/000 32 ICP_QUERY http://web/ - NONE/- -
(on 11)
884823655.052 0 134.178.9.20 UDP_MISS/000 32 ICP_QUERY http://web/ - NONE/- -
884823657.219 152 134.178.9.20 TCP_MISS/200 5618 GET http://web/ - DIRECT/web text/html
and there is nothing I could do to make it distribute the load between
the two hosts. I tried adding weights. I tried reordering the cache_hosts
in the config file. I tried removing ICP and everything else I could
think of, but nothing seemed to be able to get it to round-robin the
requests (even with the cache_host option of that name), infact it was
always .11 that did the DIRECT request. I tried default round-robin
no-query proxy-only in almost every configuration I could think of.
Bahumbug. Meh. Forget that then.
Maybe I will end up just having one box proxy .com and the other
everthing else (I would really rather a more even share). No. Dont
like that one.
That leaves having them swap the task of being the main server once a
night.... its ugly, but atleast it keeps testing my
switchover/failover capability. :)
*shrugs*
Someone must have done this without a cisco redirector. :)
cheers,
cam
-- / `Rev Dr' C.Blackwood@bom.gov.au skeptic, virtual goth \ < [+61 3] 9669 4268 BSD Unix, C/C++, genetics, ATM > \ The BoM. http://explorer.ho.bom.gov.au/~cameron/ / ____ finger korg@nod.zikzak.net for PGP/Geek Code and stuff ____ On the side of the software box, in the "System Requirements" section, it said "Requires Windows 95 or better". So I installed Linux.Received on Wed Jan 14 1998 - 21:16:35 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:38:26 MST