The mbuf cluster stats are clearly a bug in the statistics, but
the mbuf current/peak/max (current part) might not be and that
could indicate a leak.
Could you generate a kernel core with the mbuf susystem
in this state and throw it onto leaf ?
-Matt
Matthew Dillon
<[EMAIL PROTECTED]>
:Toma¾ Bor¹tnar pravi:
:> Actually around 3000 reqs/s and server machine (dual Opterons 246) is
:> over 70% idle :)
:>
:> Toma¾
:dk# wc -l access*
: 114050311 access.log
: 70914 access81.log
: 114121225 total
:
:So 114 millions of requests...
:
:while still running with clients (2):
:dk# netstat -m
:79362/2738/7232 mbufs in use (current/peak/max):
: 69825 mbufs allocated to data
: 9537 mbufs allocated to packet headers
:4180869772/4180869772/3616 mbuf clusters in use (current/peak/max)
:2492356 Kbytes allocated to network (195% of mb_map in use)
:0 requests for memory denied
:0 requests for memory delayed
:0 calls to protocol drain routines
:
:I stopped clients, because they were not getting much action - less then 60
reqs/s!
:
:dk# date
:Fri Nov 25 13:21:09 CET 2005
:dk# netstat -m
:79308/2743/7232 mbufs in use (current/peak/max):
: 69769 mbufs allocated to data
: 9539 mbufs allocated to packet headers
:4180851574/4180851574/3616 mbuf clusters in use (current/peak/max)
:2455961 Kbytes allocated to network (257% of mb_map in use)
:0 requests for memory denied
:0 requests for memory delayed
:0 calls to protocol drain routines
:
:dk# netstat -m
:79311/2746/7232 mbufs in use (current/peak/max):
: 69772 mbufs allocated to data
: 9539 mbufs allocated to packet headers
:4180851540/4180851540/3616 mbuf clusters in use (current/peak/max)
:2455894 Kbytes allocated to network (256% of mb_map in use)
:0 requests for memory denied
:0 requests for memory delayed
:0 calls to protocol drain routines
:
:So resources are not really comming back... But servers (2) are still running.
One on port 80 and the other on 81. Both
:lighttpd 1.4.8 in kqueue mode.
:
:
:Toma¾
