/ Zope / Apsis / Pound Mailing List / Archive / 2005 / 2005-02 / COOKIE affinity unreliable. Need clarification on 2K Cookies and 63 byte hash key

[ << ] [ >> ]

[ COOKIE affinity unreliable. Need clarification on ... ] [ Does Pound support CRL checking? / Damien Dougan ... ]

COOKIE affinity unreliable. Need clarification on 2K Cookies and 63 byte hash key
"Alexander N. Spitzer" <aspitzer(at)deploy.com>
2005-02-15 22:57:45 [ FULL ]
Someone in one of our remote offices is testing Pound, and is seening
that after about 20 users, users start getting sent to the wrong server.

Here is what I know from reading the Archives:

There is a 2K limit on cookies, which can be addressed by increasing MAXBUF

see:
http://www.apsis.ch/pound/pound_list/archive/2004/2004-08/1093017284000#1093017284000

in that same article, I see:

"The session code keeps only the first 63 bytes and it may get badly
confused"


*******************
THE ACTUAL QUESTION
*******************
1) Is the 63byte limit on Cookies on the WHOLE "Cookies:" line, or only
for the cookie defined as "Session COOKIE XXXXX 600"?

I am assuming the latter, but one of the things I noticed when sniffing
the headers was that the cookies seemed to appear in different orders
depending on the page, so I just wanted to make sure that that was not
the issue.

2) is the 2K limit on the WHOLE "Cookie:" line, or is that for EACH
Cookie? (this is actually moot because the whole "Cookies:" line is less
that 2K... but this is for my own understanding...

3) Any other ideas why someone might see issues with "Cookie Stickyness"
after 20+ users?


Here is what was tested in the remote office:
-------------------------------------------
ListenHTTP 192.168.54.178,7002
Client 120
LogLevel 2
UrlGroup ".*"
BackEnd 192.168.54.181,7001,3
BackEnd 192.168.54.177,7001,2
Session COOKIE SessionId 200
Server 600
EndGroup
-------------------------------------------


Version 1.8 was tested on Fedora FC3... compiled locally, no ssl...


-alex

------------------------
Alexander N. Spitzer
Deploy Solutions

[...]

MailBoxer