Lot of Apache idle workers after an attack

To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Subject: Lot of Apache idle workers after an attack
From: Olivier Bonvalet <olivier.bonvalet@xxxxxxxxxx>
Date: Sat, 20 May 2006 13:23:31 +0200

I have a problem : after a small attack (DDoS), the AMD64 Apache2 nodes have a lots of idle workers :
  "29 requests currently being processed, 541 idle workers"

Some precisions :
- it's a LVS-TUN setup, with 6 nodes (3 i386 and 3 amd64)
- other apache2 servers which are not behind LVS haven't got this problem
- i386 servers behind LVS haven't this problem
- the number of idle workers don't decrease after some hours
- MinSpareServers is 30, and MaxSpareServers is 50
- a graceful correct the problem, but since this attacks are frequents since 24 hours, I'm searching to way to "correct" this
- the website on this cluster works fine, so it's not a "big problem"
- I use Apache 2.0.54, on Debian Sarge

Have you got any idea ?

I try this :
- switch to the backup LVS. Nothing change.
- reduce the "tcp_fin_timeout" to 3 seconds. It has decreased the number of "TIME_WAIT" connections, but Apache 2 don't change

Thanks in advance,

<Prev in Thread] Current Thread [Next in Thread>
  • Lot of Apache idle workers after an attack, Olivier Bonvalet <=