Squid with diskd won't start

aboehme

Hi there,

i am running pfsense 2.1.2 64Bit with squid3 package (not more packages). The last weeks there were no problems, but suddenly squid won't work anymore. The cache.log says, there is no space left on my device, but there is enough. I've tried deleting the whole cache and starting squid -z to rebuild the cache, without success. Switching to aufs cache system works for the moment, but i still want to use diskd.
I've read about about the following kernel options:

kernel.msgmni = 64
kernel.msgmax = 16384

Will they help? Does anyone got the same error?

Thats what the cache.log says:

2014/05/04 00:02:28| Starting Squid Cache version 3.1.22 for amd64-portbld-freebsd8.3…
2014/05/04 00:02:28| Process ID 64386
2014/05/04 00:02:28| With 11095 file descriptors available
2014/05/04 00:02:28| Initializing IP Cache...
2014/05/04 00:02:28| DNS Socket created at [::], FD 11
2014/05/04 00:02:28| DNS Socket created at 0.0.0.0, FD 12
2014/05/04 00:02:28| Adding domain localdomain from /etc/resolv.conf
2014/05/04 00:02:28| Adding nameserver 127.0.0.1 from /etc/resolv.conf
2014/05/04 00:02:28| User-Agent logging is disabled.
2014/05/04 00:02:28| Referer logging is disabled.
2014/05/04 00:02:29| Unlinkd pipe opened on FD 18
2014/05/04 00:02:29| Local cache digest enabled; rebuild/rewrite every 3600/3600 sec
2014/05/04 00:02:29| Store logging disabled
2014/05/04 00:02:29| Swap maxSize 10240000 + 786432 KB, estimated 848187 objects
2014/05/04 00:02:29| Target number of buckets: 42409
2014/05/04 00:02:29| Using 65536 Store buckets
2014/05/04 00:02:29| Max Mem size: 786432 KB
2014/05/04 00:02:29| Max Swap size: 10240000 KB
2014/05/04 00:02:29| storeDiskdInit: msgget: (28) No space left on device
FATAL: msgget failed
Squid Cache (Version 3.1.22): Terminated abnormally.
CPU Usage: 0.013 seconds = 0.013 user + 0.000 sys
Maximum Resident Size: 5632 KB
Page faults with physical i/o: 0

Thanks for your help!

Dvdis

I'm having the same problem.

I already removed and recreated the cache directory, but not resolved.

I use diskd configuration and do not intend to change.

My disk space:
Filesystem Size Used Avail Capacity Mounted on
/dev/ad0s1a 16G 1.2G 13G 8% /
devfs 1.0k 1.0k 0B 100% /dev
/dev/md0 3.6M 62k 3.3M 2% /var/run
devfs 1.0k 1.0k 0B 100% /var/dhcpd/dev

djzort

i have just come across this problem and solved it.

as per http://wiki.squid-cache.org/Features/DiskDaemon one of these values arent compiled in with the right value for diskd

just go

echo "kern.ipc.msgssz=64" >> /boot/loader.conf.local

(note the >> so you dont wipe other stuff you care about)

then reboot.

run this to verify, you should see diskd running

ps ax | grep diskd
42919 ?? S 0:00.03 diskd 43539460 43539461 43539462

squid will then run with diskd.

dear pfsense maintainers, perhaps this value should be up'd by default at compile time as per the squid wiki?

djzort

further to my last post, it seems that more tweaking is needed

for a squid config with the "Local Cache/Level 1 subdirectories" set to 64, these are the sysctls which worked for me


kern.ipc.msgseg: 512
kern.ipc.msgssz: 64
kern.ipc.msgtql: 8192
kern.ipc.msgmnb: 8192
kern.ipc.msgmni: 128
kern.ipc.msgmax: 32768

some are defaults so you just need to add the following to /boot/loader.conf.local


kern.ipc.msgssz=64
kern.ipc.msgmni=128
kern.ipc.msgtql=8192

These are based on this page http://wiki.squid-cache.org/Features/DiskDaemon#FreeBSD

Here are the formulas if you want to adjust the L1 value

L1=xxx
msgmnb=8192
msgni=L1*2
msgseg=512
msgssz=64
msgtql=L1*75 # i rounded this up to 8192

Read the squid wiki page for full explanations.

djzort

this config ran for about a week, then the same "out of space' error occurred again.

something is screwy with diskd vs freebsd

for now ive gone back to the ufs option, but im sure i will be inspired to fiddle again soon