On 9/03/2015 4:38 p.m., johnzeng wrote:
Hello Dear All :
I face a problem recently , When i config wccp ( tproxy ) environment (
via using squid 3.5.2 ) ,
if i disable cache_dir rock part ,and it will be success for wccp(
tproxy) , and enable cache_dir aufs
#cache_dir rock /accerater/webcache3/storage/rock1 2646 min-size=4096
max-size=262144 max-swap-rate=250 swap-timeout=350
but if i enable cache_dir rock part ,and it will be failure for wccp(
tproxy) and enable cache_dir aufs
cache_dir rock /accerater/webcache3/storage/rock1 2646 min-size=4096
max-size=262144 max-swap-rate=250 swap-timeout=350
Whether some of my config is error , if possible , please give me some
advisement
For starters,
WCCP is a network protocol Squid uses to inform remote routers that it
is active and what traffic it can receive.
rock is a layout format for bits stored on a disk.
... they are *completely* unrelated.
This is my config
thanks
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------
coredump_dir /accerater/logs/webcache3/
unlinkd_program /accerater/webcache3/libexec/unlinkd
pid_filename /accerater/logs/webcache3/opmizer1/cache.pid
workers 2
cpu_affinity_map process_numbers=1,2 cores=1,3
cache_dir rock /accerater/webcache3/storage/rock1 2646 min-size=4096
max-size=262144 max-swap-rate=250 swap-timeout=350
cache_dir rock /accerater/webcache3/storage/rock1 2646 min-size=4096
max-size=262144 max-swap-rate=250 swap-timeout=350
You are telling Squid to start two controllers to the database file
/accerater/webcache3/storage/rock1 from *each* worker. There is zero
benefit from this and the two controllers may enounter collisions as
they compete for acces to the disk without sharing atomic locks. That
leads to cache corruption.
Remove one of those two lines.
if ${process_number} = 1
cache_swap_state /accerater/logs/webcache3/opmizer1_swap_log1
Dont use cache_swap_state.
access_log stdio:/accerater/logs/webcache3/opmizer1_access.log squid
Use this instead (mind the wrap):
access_log
stdio:/accerater/logs/webcache/opmizer${process_number}_access.log squid
cache_log /accerater/logs/webcache3/opmizer1_cache.log
Use this instead:
cache_log /accerater/logs/webcache3/opmizer${process_number}_cache.log
cache_store_log stdio:/accerater/logs/webcache3/opmizer1_store.log
You should not need cache_store_log at all.
Either remove it or use this instead (mind the wrap):
cache_store_log
stdio:/accerater/logs/webcache3/opmizer${process_number}_store.log
url_rewrite_program /accerater/webcache3/media/mediatool/media2
store_id_program /accerater/webcache3/media/mediatool/media1
Why do you have different binary executable names for the two workers
helpers?
If they are actually different, then the traffic will have different
things applied randomly depending on which worker happened to accept the
TCP connection. If they are the same, then you only need to define them
once and workers will start their own sets as needed.
unique_hostname fast_opmizer1
snmp_port 3401
Use this instead:
unique_hostname fast_opmizer${process_number}
snmp_port 340${process_number}
All of the above details can move up out of the per-worker area.
#cache_dir rock /accerater/webcache3/storage/rock1 2646 min-size=4096
max-size=262144 max-swap-rate=250 swap-timeout=350
cache_dir aufs /accerater/webcache3/storage/aufs1/${process_number} 5200
16 64 min-size=262145
else
#endif
if ${process_number} = 2
cache_swap_state /accerater/logs/webcache3/opmizer2_swap_log
access_log stdio:/accerater/logs/webcache3/opmizer2_access.log squid
cache_log /accerater/logs/webcache3/opmizer2_cache.log
cache_store_log stdio:/accerater/logs/webcache3/opmizer2_store.log
url_rewrite_program /accerater/webcache3/media/mediatool/media4
store_id_program /accerater/webcache3/media/mediatool/media3
unique_hostname fast_opmizer2
snmp_port 3402
Same notes as for worker 1.
#cache_dir rock /accerater/webcache3/storage/rock2 2646 min-size=4096
max-size=262144 max-swap-rate=250 swap-timeout=350
cache_dir aufs /accerater/webcache3/storage/aufs1/${process_number} 5200
16 64 min-size=262145
endif
endif
http_port 127.0.0.1:3220
http_port 3221 tproxy
wccp_version 2
wccp2_router 192.168.2.1
wccp2_forwarding_method 1
wccp2_return_method 1
wccp2_assignment_method 1
wccp2_service dynamic 80
wccp2_service dynamic 90
wccp2_service_info 80 protocol=tcp flags=src_ip_hash priority=240 ports=80
wccp2_service_info 90 protocol=tcp flags=dst_ip_hash,ports_source
priority=240 ports=80
Both workers are telling the WCCP router in different packets that they
are available on the same IP:port. In theory that should work fine,
since the router is just getting twice as many updates as it needs to
keep the proxy registered. In practice some people are finding that
certain routers cant cope with the extra registration operations.
tcp_outgoing_address 192.168.2.2
Be aware that TPROXY spoof the client outgong address. This line has no
effect on the TPROXY intercepted traffic. Only the traffic received on
port 3220 will be using this outgoing address.
Your WCCP rules need to account for that by not depending on the packet
IP addresses.
Amos
