No buffer space available

[Fernando Schapachnik]

Hi,
	First of all, I hope this is on-topic. Please let me know otherwise.

	Scenario: 4.7p10 machine, 128 Mb RAM, running 3 squid processes (each
one with a different ACL set on a different port). Everything fine. Now I'm
trying to end up with one process with a merge of the ACLs, receiving the load
of the previous 3. ACLs are in files and account for 168 Kb.

	After few seconds of running the unified process, squid reports 

comm_open: socket failure: (55) No buffer space available

	which maps to:

if ((new_socket = socket(AF_INET, sock_type, proto)) < 0)

	The system doesn't seem to be lucking mbufs (according to netstat -b),
nor memory, nor file descriptors. I recompiled kernel with
MAXSSIZ="(256*1024*1024)" to get the stack size lifted, but no change.

	Top shows:

----------------------------------------------------------------------------
last pid:   453;  load averages:  0.48,  0.30,  0.13                       up
0+00:02:32  08:48:00
36 processes:  1 running, 35 sleeping
CPU states:  0.4% user,  0.0% nice,  0.4% system,  0.0% interrupt, 99.2% idle
Mem: 52M Active, 4608K Inact, 14M Wired, 3912K Cache, 22M Buf, 45M Free
Swap: 500M Total, 13M Used, 487M Free, 2% Inuse

  PID USERNAME  PRI NICE  SIZE    RES STATE    TIME   WCPU    CPU COMMAND
  187 nobody      2   0 46440K 45624K poll     0:07  0.59%  0.59% squid
  453 root       28   0  1900K   956K RUN      0:00  0.56%  0.15% top
----------------------------------------------------------------------------

# more /proc/187/rlimit
cpu -1 -1
fsize -1 -1
data 536870912 536870912
stack 268435456 268435456
core -1 -1
rss -1 -1
memlock -1 -1
nproc 896 896
nofile 1792 1792
sbsize -1 -1
vmem -1 -1

	Each squid individual process has more or less the same memory
footprint of the merged one.

	Squid mailing lists show post with similar problems but no solutions.

	Any ideas? Thanks in advance.



Fernando Schapachnik