[rt-users] Page Load Hangs in 3.4.4

[Joop van de Wege]

On Tue,  1 Nov 2005 16:50:27 -0500 (EST)
rt-users-request at lists.bestpractical.com wrote:

> I've come across an interesting situation where impatient users get 
> punished for clicking around too fast.  If a page load is interrupted, 
> especially one returning a good-sized chunk of data (like a ticket 
> search or detail), the socket connection gets put into a CLOSE_WAIT 
> state that takes a *long* time (hours) to go away.  For that user, RT 
> becomes unusable until s/he restarts the browser, at which point 
> everything is fine again.
> 
> Apparently this is a known HTTP flaw, but it is much more pronounced 
> here than on any of my other webapps (none of which use Mason).  Both 
> the web and database servers have plenty of CPU/RAM/swap, etc.  Anything 
> I can do to speed up database queries and pageloads?  Sorry this 
> question is so general, but so is the problem.
> 
> RT 3.4.4
> Apache 1.3.33
> Oracle 9.2.0.4
> Solaris 8
I'm having the same config except we are running RT 3.4.3 and haven't
had this problem. I just tried a couple of heavy operation like
searching on ticket content and aborting the search but I can start a
new query right away. Maybe it is because we're running RT using
mod_perl (Apache/1.3.33 (Unix) PHP/4.4.0 mod_fastcgi/2.2.12 mod_perl/1.29 mod_ssl/2.8.23 OpenSSL/0.9.8)
Further info that might help. System is a Sun/Sparc Solaris8 420R with
4CPU and 4Gb of memory. Its main use is running RT with a couple of
other light webapps such as Openwebmail/WebCalender/FaqOmatic. Oracle
9.2.01, relevant parts from init.ora:
large_pool_size=8388608
shared_pool_size=115343360 
db_block_size=8192
db_cache_size=25165824
db_file_multiblock_read_count=16 
So almost no tuning, except for a context index on attachments.content
and a modification in Searchbuilder to use that index when searching for
ticket content.

Joop


-- 
Joop van de Wege <JoopvandeWege at mococo.nl>