[squid-users] Strange problem with NTLM_AUTH

Hi all;
We have Squid 2.5.STABLE9 running with 30 ntlm_auth helpers Version
3.0.10-1.fc3. The problem is that on the squid´s cachemgr.cgi->NTLM User
Authenticator Stats is possible to verify that the ntlm_auth processes
are slowly having the flag R (Reserved or Deferred) set and never being
used again (the number of requests stops and the time starts growing).
This problem goes until there is no more ntlm_auth process available and
squid restarts itself, restarting all ntlm_auth too. After the restart,
everything is goes back to normal and the problem slowly repeats. We
detected that the R flag appears more agressively if the domain controller
is under more load (like running a backup script).
Well, the questions are:
1- What does the flag reserved mean?
2- Any ideas why the R flag is spreading throught all the ntlm_auth
processes, like processes 11 and 12 below (you can see the 11 and 12 are
locked for a long time and process 13 is receiving more requests ?

# FD PID # Requests Flags Time Offset
Request
1 8 8656 475909 R 10.140 0 (none)
2 9 8657 632482 0.093 0 (none)
3 10 8658 363615 0.412 0 (none)
4 11 8659 64199 R 311498.132 0 (none)
5 12 8660 33142 R 311497.891 0 (none)
6 13 8661 121226 0.932 0 (none)
7 14 8662 58971 0.913 0 (none)
(...)

Just for the record, I originally have sent this message to the SAMBA
list, and Andrew Bartlett (NTLM_AUTH coder) replyed me as below:
"It might be that we need to have a better way to have ntlm_auth tell
Squid that there is a problem now, but it might go away (previous helper
designs had to be restarted for that to happen, but ntlm_auth can
recover on it's own).

Andrew Bartlett"

Any help is greatly appreciated;
Best regards;

Rafael Sarres de Almeida
Seção de Gerenciamento de Rede
Superior Tribunal de Justiça
Tel: (61) 319-9342
Received on Tue Jun 21 2005 - 08:36:25 MDT